## Detailed walkthrough
### [11:18](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=678s) Introduction and Structure of Investor Day
Tesla Chief Financial Officer Zach Kirkhorn takes the stage at Tesla’s global headquarters in Austin, Texas, greeting both the live audience and those tuning in online. Kirkhorn reflects on Tesla's historical trajectory, describing distinct cycles defined by product advancement, technological breakthroughs, and aggressive volume scaling—culminating most recently in the global expansion and localized production of the Model Y program.
Looking ahead, Kirkhorn emphasizes that the day’s focus is explicitly long-term rather than tied to quarterly metrics: “We don't want to talk about this quarter or next quarter. We want to go further out into the future.” He outlines a three-part structure for the event:
- **Part 1:** The high-level macro view addressing what is physically required to transition the entire planet to sustainable energy generation and consumption.
- **Part 2:** Tesla’s concrete execution across every company discipline, featuring direct presentations from leadership teams spanning vehicle design, powertrains, software, and manufacturing.
- **Part 3:** A synthesis of what these combined technical and operational efforts mean for Tesla as a corporate entity.
Kirkhorn closes his opening remarks with standard safe-harbor reminders regarding forward-looking statements before welcoming CEO Elon Musk and Senior Vice President of Powertrain and Energy Engineering Drew Baglino to present Master Plan Part 3.
---
[13:31](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=811s) Master Plan Part 3: Abundance vs. Austerity
Elon Musk opens Master Plan Part 3 by laying out the primary message of the presentation: humanity does not face an impossible environmental dead end. Transitioning Earth to sustainable energy does not demand ecological degradation or forced human sacrifice.
> “The thing that I think is we wanted to convey probably more importantly than anything else that we talked about here is that there is a clear path to a sustainable energy Earth.” — Elon Musk
Musk pushes back firmly against the prevailing notion that sustainability requires extreme privation: “It doesn't require us to be austere and stop using electricity and sort of be in the cold or anything.” Instead, Tesla's models show that a fully electrified system can sustain a human population significantly larger than the current 8 billion people in material abundance. Musk expresses disbelief that even highly educated individuals frequently view the climate transition as impossible under current global demographics without harsh countermeasures. To substantiate these claims, Tesla plans to release a detailed white paper documenting all empirical assumptions, datasets, and calculations behind Master Plan 3.
---
[15:23](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=923s) The Inefficiency of the Fossil Fuel Economy
Drew Baglino frames the existing status quo, characterizing the modern fossil-fuel-driven energy economy as fundamentally dirty and inefficient. Over 80% of primary energy globally is sourced from fossil fuels, yet merely one-third of that energy translates into actual useful work or heat. Musk explains that this basic thermodynamic reality is routinely misunderstood by both general observers and engineers:
```mermaid
flowchart LR
accTitle: Fossil Fuel Energy Losses in Vehicles
accDescr: Diagram contrasting primary energy input from crude oil with end-use kinetic energy delivered to the road.
subgraph WellToPump["Upstream Extraction & Processing"]
direction TB
A["Crude Oil Extraction"] --> B["Refining Process"]
B --> C["Transport & Distribution"]
end
subgraph InternalCombustion["Vehicle Conversion"]
direction TB
D["Combustion in ICE"]
E["Waste Heat: ~75% to 80%"]
F["Kinetic Motion: <20%"]
end
WellToPump --> D
D --> E
D --> F
```
In an internal combustion engine vehicle, only 20% to 25% of the chemical energy contained in gasoline becomes kinetic motion; the remainder radiates away as waste heat. When factoring in the upstream energy spent extracting crude oil, refining it, and transporting fuel to retail stations, less than 20% of the original energy from the extracted oil ever pushes the vehicle forward.
Musk points out that critics of electrification make a fundamental modeling error: they assume that a green civilization must generate the full 165 petawatt-hours (PWh) of primary energy currently consumed each year by the fossil economy. Because electric motors, heat pumps, and batteries minimize thermal losses across production and end-use stages, an electrified global economy cuts the total required primary energy in half. Even under conservative and intentionally pessimistic modeling assumptions, society needs roughly half the primary energy input of the combustion baseline.
---
[18:10](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1090s) The High-Level Blueprint for Earth’s Transition
Musk and Baglino detail the high-level inputs required to construct this sustainable system worldwide. The cornerstone of the plan is stationary and mobile battery storage to balance the intermittency of solar and wind installations:
```mermaid
flowchart TD
accTitle: Macro Requirements for Sustainable Energy Infrastructure
accDescr: High-level requirements including storage, generation, land, and capital needs.
subgraph Storage["Energy Storage"]
S1["240 TWh Total Battery Storage"]
S2["Electric Vehicles + Stationary Grids"]
S3["30 TW Clean Power Capacity (8:1 Storage-to-Power Ratio)"]
S1 --- S2
S2 --- S3
end
subgraph Footprint["Land & Capital Scale"]
F1["$10T Global CapEx (Over 10–20 Years)"]
F2["<0.2% of Earth's Land Area for Solar & Wind"]
F3["Mining Footprint Net Reduction vs. Fossil Fuels"]
F1 --- F2
F2 --- F3
end
Storage --> Footprint
```
- **Storage Capacity:** Approximately 240 terawatt-hours (TWh)—equivalent to 240,000 gigawatt-hours (GWh)—of total battery energy storage split between electric transport and grid storage.
- **Renewable Power Generation:** 30 terawatts (TW) of renewable generation capacity, reflecting roughly an 8:1 ratio of stored energy to continuous generation capacity.
- **Capital Expenditure:** While raw manufacturing modeling points to roughly $6 trillion in CapEx across the entire supply chain (mining, refining, battery manufacturing, recycling, and vehicle factories), Tesla rounds this figure up to a conservative $10 trillion. Relative to the roughly $100 trillion global annual GDP, this represents approximately 1% of world economic output if spent over 10 years, or 0.5% over 20 years.
- **Land Footprint:** Supplying global power through wind and solar requires less than 0.2% of Earth’s surface area. Musk notes that incident solar energy equates to roughly 1 GW per square kilometer; when divided by a capacity factor of 4 to 5 to account for day-night cycles, the land footprint remains modest. Furthermore, wind and solar can share the same land footprint or be deployed offshore over Earth's oceans, which comprise 70% of the surface.
- **Material Extraction:** Refuting concerns over raw material scarcity, Baglino and Musk state that electrifying the globe will ultimately require extracting significantly less raw material and ore from the Earth than the ongoing, non-recyclable extraction cycle of the fossil fuel economy.
---
[21:31](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1291s) The Five Pillars of Electrification
To achieve complete global sustainability, Baglino outlines five discrete operational focus areas:
1. **Re-power the existing electrical grid** with wind, solar, and utility-scale stationary storage.
2. **Switch personal and commercial fleets to electric vehicles**, which dramatically cuts transport energy demand.
3. **Switch residential, commercial, and industrial space/water heating to heat pumps**, capitalizing on thermodynamic multiplication of energy input to heat output.
4. **Deploy high-temperature heat delivery and thermal storage** for heavy industrial and chemical manufacturing, supplemented by targeted green hydrogen production where hydrogen serves as an irreplaceable chemical reactant.
5. **Sustainably fuel planes and maritime vessels**, leveraging synthetic clean fuels for long-haul duty cycles while capitalizing on increasing battery energy density to electrify shorter routes.
Musk adds that as volumetric and gravimetric battery energy densities continue to climb, every category of transportation will eventually become fully electrified.
[22:20](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1340s) Rocket Fuel and Re-Powering the Grid with Renewables
The discussion begins with an aside on the rare exceptions to direct electrification, specifically rockets. Elon Musk explains that while you cannot plug a rocket directly into a battery pack, rocket propellant itself can still be electrified indirectly: with electricity, water, and carbon dioxide, one can synthesize methane ($\text{CH}_4$) and oxygen ($\text{O}_2$). Musk points out that this exact Sabatier-style process is what will be used on Mars, where an atmosphere rich in $\text{CO}_2$ and widespread water ice provide the raw ingredients to produce fuel. As Musk concludes, “So, ultimately, even rockets can be electrified.”
Turning to the terrestrial transition under Master Plan Part 3, Drew Baglino details the first major pillar: re-powering the existing electrical grid with renewable power. Across the presentation slides, estimates are displayed for the necessary terawatt-hours of storage, terawatts of generation capacity, and trillions of dollars of capital expenditure. Baglino stresses that this transformation is not a distant vision but is already well underway. In the United States, 60% of all new generation capacity added to the grid in 2022 came from solar energy, with solar deployments expanding at a 50% year-over-year rate. At that pace, Baglino argues, the transition to clean generation will occur far faster than most expect.
- In 2022, solar power represented 60% of all newly added utility generation capacity in the United States.
- U.S. solar deployment grew at a 50% year-over-year rate as of 2022.
---
[23:32](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1412s) Fleet Electrification, Autonomy, and End-Use Efficiency
Switching the global transport fleet to electric vehicles represents the second pillar, directly yielding a 21% reduction in total fossil fuel consumption. Global EV production surged 59% year-over-year in 2022, crossing what Baglino calls an “awesome milestone” of 10% global market share. Musk underscores that the shift will be comprehensive, predicting that the entire automotive fleet will inevitably become fully electric and fully autonomous. To emphasize how outdated traditional vehicles will feel, Musk remarks:
> “And so, writing a non-autonomous gasoline car is going to be analogous to riding a horse and using a flip phone. That’s basically the going to be the situation.”
Addressing fleet scale, Baglino and Musk discuss how autonomous operation alters vehicle demand. Today, approximately 2 billion passenger cars and commercial trucks operate globally. However, autonomous fleets operate at dramatically higher utilization rates, substantially reducing the absolute number of vehicles needed to deliver the same passenger miles. Tesla’s modeling takes a conservative stance by largely omitting these aggressive fleet-downsizing assumptions from their core battery volume calculations. Even so, the model assumes a reduced active passenger fleet of around 1.4 billion vehicles (compared to the existing 2 billion) supported by an annualized global production rate of approximately 85 million vehicles across all manufacturers, with Tesla targeting an output of 20 million electric vehicles per year.
```mermaid
flowchart TD
accTitle: Well-to-Wheel Efficiency Comparison
accDescr: Comparison between Tesla Model 3 well-to-wheel electrical efficiency and Toyota Corolla internal combustion fuel cycle.
subgraph EV["Electric Vehicle Path (Tesla Model 3)"]
direction TB
E1["Renewable Power Generation"] --> E2["Grid Transmission & Charging"]
E2 --> E3["High-Efficiency Powertrain"]
E3 --> E4["4x Well-to-Wheel Efficiency"]
end
Baglino highlights the physics driving this transition: end-use efficiency. On a well-to-wheel basis, a Tesla Model 3 is four times more energy-efficient than a comparable internal combustion car like a Toyota Corolla. To make this tangible, Baglino provides a culinary comparison: a Model 3 can travel more than one mile using merely the energy required to bring a pot of pasta water to a boil, and another mile on the energy required to cook the pasta. Musk points out the striking physical reality that a 4,000-pound vehicle can be propelled for miles on the energy consumed by preparing just one single pound of pasta, noting that thermal phase change and heating require far more raw energy than mechanical motion.
- 21% reduction in global fossil fuel consumption achieved by switching global vehicles to electric drives.
- Global EV output increased 59% year-over-year in 2022, capturing a 10% worldwide automotive market share.
- Global operating fleet: roughly 2 billion cars and trucks today; modeled down to roughly 1.4 billion units under widespread autonomy.
- Global vehicle production run rate modeled around 85 million vehicles per year, with Tesla targeting 20 million vehicles annually.
- Well-to-wheel efficiency: Model 3 delivers four times the efficiency of a Toyota Corolla across the complete energy supply chain.
---
[26:56](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1616s) Heat Pumps for Homes, Businesses, and Industry
The third lever in Master Plan Part 3 is the widespread deployment of heat pumps across residential, commercial, and industrial facilities. Currently, heat pumps fulfill only 10% of building heating requirements globally, with annual installation rates expanding at roughly 10% year-over-year. Baglino identifies heat pumps as the single lowest-hanging fruit in the entire decarbonization plan when measured by capital cost per unit of fossil fuel displaced.
Demystifying the hardware, Baglino explains that heat pumps do not generate thermal energy by burning a fuel source like a natural gas furnace; rather, they move existing ambient heat from the outside environment into the interior, functioning mechanically as an air conditioner or refrigerator operating in reverse. This thermodynamic leverage delivers a threefold (3x) reduction in primary energy consumption compared to conventional heating systems. Furthermore, modern industrial heat pumps can deliver temperatures up to 200°C. Noting that heat pump thermal loops are now standard equipment across Tesla's vehicle lineup, Musk hints at potential product expansion, quipping: “And at some point, we might make a heat pump for our home.”
- Building heating: Heat pumps satisfy 10% of building heating today, growing at a 10% year-over-year install rate.
- Thermal range: Commercial and industrial heat pump architectures can serve heat loads up to 200°C.
- Energy reduction: Switching from fossil furnaces to heat pump systems delivers a 3x reduction in total energy consumed for heating.
---
[28:18](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1698s) High-Temperature Industrial Heat and Green Hydrogen
Addressing hard-to-abate industrial sectors, Baglino notes that over 50% of all industrial heat demand requires temperatures exceeding 400°C, with heavy industrial processes such as cement, steel, chemical refining, plastics, and fertilizer manufacturing requiring process temperatures upward of 1,500°C. Because renewable energy generation is intermittent, electrifying these sectors relies on purpose-built thermal storage equipment. During daily peak generation windows when renewable electricity is abundant and cheap, electric resistive heating creates high-temperature heat that is absorbed into thermal storage media. Materials such as carbon graphite remain thermally stable up to 2,800°C, while silicon carbide and other refractories handle loads around 1,500°C. This stored thermal energy can then be steadily discharged into continuous 24/7 manufacturing operations.
Beyond direct thermal electrification, green hydrogen is required for chemical feedstock reactions and metal refining. Currently, around 120 million tons of hydrogen are consumed annually, almost entirely derived from fossil fuel reforming. Green hydrogen produced via water electrolysis can substitute directly for metallurgical coal in steelmaking by running Direct Reduced Iron (DRI) furnaces instead of traditional blast furnaces.
While acknowledging the role of green hydrogen in heavy industry, Musk emphasizes his longstanding skepticism regarding its application in mobility:
> “My personal opinion is that hydrogen will not be used meaningfully in transport, but—and shouldn’t be. If you’re going to use a chemical fuel, you should use CH4, not H2.”
- Over 50% of all industrial process heat demand occurs at temperatures greater than 400°C.
- Heavy industrial operations (steel, cement, chemical synthesis) routinely demand temperatures reaching 1,500°C.
- Carbon graphite refractory materials maintain physical stability at operating temperatures up to 2,800°C.
- Current hydrogen usage: approximately 120 million metric tons per year, predominantly derived from unabated fossil sources.
---
[30:23](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1823s) Electrifying Maritime Shipping and Aviation
The final wedge of the plan targets maritime shipping and aviation. Although representing a smaller portion of aggregate global emissions—with maritime shipping accounting for approximately 3% of global carbon dioxide—it remains an essential component for complete decarbonization. Baglino points out that maritime shipping is particularly well-suited to electrification; even with lower-density lithium iron phosphate (LFP) chemistry, long-haul cargo ships can be engineered to run entirely on battery-electric powertrains.
For aviation, short-haul regional flights can be electrified using current cell technology. While intermediate long-haul flights may rely on sustainable aviation fuels (SAF) synthesized from excess renewable electricity and captured $\text{CO}_2$, Musk stresses that true long-range electric aircraft and vessels require ground-up systems engineering. Much like automotive design, where simply retrofitting an internal combustion chassis with heavy battery modules creates an inefficient vehicle, planes and ships must be built with structural battery packs that form the load-bearing airframe or hull. Musk explains that once aircraft are optimized around structural packs, long-range electric flight becomes commercially viable using battery cells with energy densities around 450 watt-hours per kilogram ($\text{Wh/kg}$)—performance levels that are already commercially available today, albeit at premium price points that will decline over time.
- Maritime shipping generates 3% of global carbon dioxide emissions and can utilize LFP battery chemistries.
- High-efficiency electric aircraft require ground-up structural battery architectures rather than retrofitted airframes.
- Battery energy density threshold for long-range electric flight is approximately 450 Wh/kg, a specification achievable with existing high-end cell chemistries.
[32:06](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1926s) Feasibility, Required Growth Rates, and Capital Investment
Reviewing the overarching numbers of Master Plan 3—30 terawatts of renewable power, 240 terawatt-hours of storage, and a $10 trillion capital requirement—Drew Baglino addresses the obvious investor question: is any of this actually feasible? Giving away the punchline immediately, he declares: “Spoiler alert, it's entirely feasible.” When broken down by required expansion rates against current trajectories, the gaps close surprisingly fast. Solar and wind deployment only need to scale 3x, a target well within reach given that both industries are already expanding at a breakneck pace. Electric vehicle production needs an 11x scale-up, but global EV output surged roughly 60% year-over-year in the prior year alone. Stationary energy storage must expand dramatically, yet Tesla’s own stationary storage business has sustained a 65% compound annual growth rate (CAGR) since 2016, with the broader industry accelerating in tandem. Tesla’s long-term targets remain pegged at manufacturing 20 million electric vehicles per year and deploying 1 terawatt-hour of stationary storage per year as fast as humanly possible.
When evaluating the $10 trillion price tag, context makes the figure far less intimidating. As Elon Musk noted, it represents about 10% of a single year's global gross domestic product spread over a multi-decade transition. Viewed from another angle, it amounts to roughly 60% of what society spent just in the preceding year maintaining and expanding fossil fuel infrastructure. Building out a clean, electrified energy system requires less ongoing capital investment year-over-year than simply sustaining the legacy fossil fuel economy.
- Required renewable deployment increase: 3x for solar and wind
- Required EV manufacturing growth: 11x (against a baseline of 60% year-on-year global growth)
- Tesla energy storage CAGR: 65% since 2016
- Long-term annual targets: 20 million EVs/year and 1 TWh/year of stationary storage
- Sustainable energy transition capital cost: ~60% of current annual fossil fuel capital expenditures
[33:57](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2037s) Global Land Footprint and Environmental Displacement
The physical space required to harness this renewable power is remarkably minimal, occupying less than 0.2% of Earth's total land surface. To put that footprint in perspective, intensive agriculture currently occupies roughly 12.5% of the planet's land area—an order of magnitude larger than everything required for solar, wind, and storage combined. While farms are a familiar sight during road trips, they are far from everywhere, illustrating how negligible a 0.2% land allocation really is.
Furthermore, solar and wind installations do not need to displace productive farmland, clear-cut forests, encroach upon jungles, or infringe on delicate ecological reserves. They can be installed across barren, uncultivated, sparsely populated desert regions and land otherwise unsuitable for commercial development. Musk underlines that moving away from combustion does not compromise the natural world, arguing that: “Transition to a sustainable energy economy would result in a substantial reduction in current ecological impact.”
```mermaid
flowchart TD
accTitle: Global Land Footprint Comparison
accDescr: Visual comparison of intensive agricultural land use versus the land required for complete sustainable energy infrastructure.
subgraph EarthLand["Earth's Total Land Surface Area (100%)"]
direction TB
Farming["Intensively Farmed Land: ~12.5%"]
Renewables["Sustainable Energy Infrastructure: < 0.2%"]
OtherLand["Remaining Deserts, Forests, Urban & Other: ~87.3%"]
end
Renewables -.->|"Requires >60x less space than"| Farming
```
[35:02](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2102s) Resource Extraction, Material Abundance, and Battery Chemistries
A persistent misconception about clean energy is that the mining and mineral extraction required to build batteries and solar panels will strip the earth bare. Baglino dispels this by contrasting global extraction mass flows. Today, humanity mines roughly 68 gigatons of ore, fossil fuels, and minerals annually—visualized as 68 individual 1-gigaton haul trucks. In a fully electrified economy, the vast bulk of those haul trucks vanishes because fossil fuel extraction ceases entirely. Replacing coal, oil, and gas with minerals for renewables and batteries results in a significant net reduction in total earth moved each year. While specialized mining and refining capacity must still be built out, the total physical mass moving through global supply chains will be lower than what is handled today.
Comparing cumulative raw material demand through 2050 against known United States Geological Survey (USGS) reserves confirms that humanity will not exhaust its geological bank. In fact, economic history demonstrates that resource estimates expand as extraction increases. Whenever exploration budgets rise, geologists discover far more material than was previously registered, meaning known reserves of key battery metals have grown steadily since 2000 rather than depleted.
```mermaid
flowchart LR
accTitle: Global Resource Extraction Comparison
accDescr: Diagram illustrating total annual material extraction dropping as fossil fuel extraction is replaced by battery and clean energy minerals.
Current -->|"Transition eliminates fossil extraction"| Clean
BatteryMetals --> RecyclingLoop
```
Musk tackles widespread market anxiety regarding battery minerals, starting with lithium. Despite common claims of scarcity or geopolitical chokeholds, lithium is widely distributed across the globe: “There's enough domestic material to electrify Earth.” Rather than mining raw ore, the true industrial bottleneck lies in refining spodumene or brines into battery-grade lithium hydroxide and lithium carbonate—a processing constraint solvable through standard chemical engineering investments.
Turning to other chemistries, nickel remains the most technically constrained element, though the master plan requires only about 30% of current known nickel reserves, which themselves continue to expand. High-energy-density nickel cathodes will be reserved strictly for weight-sensitive applications: long-range aircraft, transoceanic shipping, and heavy, ultra-long-range terrestrial transport. The workhorse of global electrification will instead be iron-based cells (such as lithium iron phosphate). Iron is the single most abundant element on Earth by mass, sitting alongside oxygen at the planet's core and crust. As Musk jokes, the planet is essentially a “muddy rust ball,” guaranteeing that society will never run short of iron cathode materials.
- Total current global resource extraction: ~68 gigatons per year
- Net material extraction decreases under a sustainable energy economy
- USGS known reserves of key energy transition materials have grown since 2000
- United States domestic lithium reserves alone could theoretically electrify the entire planet
- Primary supply bottleneck: industrial chemical refining into lithium hydroxide and lithium carbonate
- Projected nickel demand: ~30% of known global reserves
- Nickel applications: aircraft, marine vessels, long-range heavy transport
- Mass-market vehicle and storage chemistry: iron-based cathodes
[39:15](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2355s) Battery Recycling and the Core Thesis of Master Plan 3
The resource story ultimately culminates in closed-loop recycling. Once the initial infrastructure and battery fleets are deployed, the ongoing demand for newly mined ore falls by an order of magnitude or more. Unlike fossil fuels, which are incinerated once into atmospheric pollution, battery elements remain intact inside their cells, ready to be recovered and fed directly back into cathode manufacturing lines.
Wrapping up the presentation of Master Plan 3, Musk shares the broader philosophy behind the day's disclosures. Rather than catering narrowly to Wall Street analysts or Tesla equity holders, the plan is designed for anyone invested in the future of the planet itself. The conclusions are anchored in fundamental physics, thermodynamic reality, and engineering feasibility rather than speculative optimism. Musk concludes the overview with an emphatic promise to the room: “Earth can and will move to a sustainable energy economy, and will do so in your lifetime.”
[40:21](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2421s) Part 2: Vehicle Design with Franz von Holzhausen and Lars Moravy
With Master Plan 3 laid out, Franz von Holzhausen (VP of Vehicle Design) and Lars Moravy (VP of Vehicle Engineering) take the stage to introduce the execution phase. Von Holzhausen recalls joining Tesla back in 2008 to build an in-house design studio from scratch—a function that simply did not exist at the company prior to his arrival. At that time, the electric vehicle landscape was virtually barren, and the fledgling team operated with only a handful of engineers and stylists.
Designing the original Model S forced the team to rethink vehicle architecture without the safety net of an established automotive factory. Operating without a plant gave the tiny design studio free rein to lead technical development, inventing concepts that challenged legacy auto norms: engineering rear jump seats so a sleek sedan could seat seven people, engineering flush motorized door handles that slid into the door skins, and embedding a massive central touchscreen console before touch interfaces were standard in automotive cabins.
Winning the Motor Trend Car of the Year award in 2013 for the Model S marked an incredible triumph for a clean-sheet vehicle from a startup. Yet looking back, Moravy and von Holzhausen acknowledge that their original development model was fundamentally traditional and strictly linear: Franz's team created the styling, Lars's engineering team figured out how to package the components, and only then did manufacturing teams attempt to build it. In those early days, as Franz points out, they were finalizing vehicle blueprints before they even knew where on Earth they were going to manufacture the car.
[42:27](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2547s) Fremont Lessons and Model 3 “Production Hell”
Franz von Holzhausen and Lars Moravy look back at Tesla’s manufacturing origins in Fremont, comparing the early days to “flying a plane and putting the wings and building the engine at the same time.” When scaling up to Model 3, the design brief was clear: build a smaller, more affordable, and more efficient vehicle that retained everything drivers loved about the Model S and Model X. While Tesla successfully united design, engineering, and manufacturing under one roof, an abrupt mid-stream push toward total automation collided with a car originally engineered around manual assembly, spiraling the factory into what became known as “production hell.”
Moravy reflects candidly on the missteps and hard-won expertise of that period. Tesla had to rip out numerous automated systems across the Fremont floor because attempting to automate processes conceived for human hands proved disastrously difficult. Some targeted automated stations did survive, including one Moravy built with a small engineering team that is still running faster than ever today. Despite those early design constraints and manufacturing hurdles, Model 3 emerged as the best-selling electric vehicle in history, with its sibling, Model Y, on track to surpass it.
> “Like Franz said, automating something that we designed to be built manually is super hard.” — Lars Moravy
---
[44:21](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2661s) The Cybertruck Paradigm: Letting Materials Dictate Design
With Cybertruck, Tesla flipped the traditional product development sequence upside down by starting directly with the manufacturing process. The vehicle's architecture was dictated entirely by its primary material: full hard stainless steel. Because ultra-hard stainless steel cannot be stamped or drawn into conventional curved body panels, the design had to embrace sharp, linear bending operations that were virtually alien to modern automotive manufacturing language. Rather than crippling the product, this severe constraint resulted in an unmistakable silhouette and eliminated two of the largest, most capital-intensive areas of a car factory: the stamping shop and the paint shop.
Von Holzhausen emphasizes that these manufacturing constraints never compromised the truck's utility, delivering full truck functionality while radically shrinking the factory footprint. Moravy reassures the audience that the prototypes displayed at the event are genuine production-intent vehicles, affirming that Cybertruck is coming this year.
> “Forming full hard stainless steel isn't rocket science, but it sure isn't easy, and it limited the way we could do it.” — Lars Moravy
- Stamping presses and paint shops are completely eliminated for the Cybertruck body exterior.
- Panel forming relies on planar bends rather than complex stampings.
- Cybertruck was confirmed to launch later in 2023.
---
[45:55](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2755s) Questioning 100 Years of Assembly Line Orthodoxy
Tesla's core organizational advantage lies in collapsing boundaries: design, vehicle engineering, manufacturing, and automation teams sit in the same room and report to a single leader, eliminating cross-departmental finger-pointing. Moravy contrasts this integrated setup with legacy automotive manufacturers, who still adhere to the sequential manufacturing paradigm introduced by Henry Ford in 1922. For a century, the industry standard has followed an identical path: stamp panels, weld them into a body-in-white framing station, bolt on doors, paint the entire shell, remove the doors again, and send human workers into cramped body openings to install wiring, instrument panels, and seats before finally hauling the doors back to reattach them.
This legacy sequence forces people and robots into awkward ergonomic postures while transporting an oversized five-meter shell through vast distances with little value-added work happening along the way. While Japanese rivals recently evaluated the Model Y and declared it an “engineering work of art,” Moravy asserts that Tesla cannot rest on those laurels if it intends to achieve orders-of-magnitude scaling and fulfill Master Plan 3.
```mermaid
flowchart LR
accTitle: Traditional 100-Year-Old Vehicle Assembly Sequence
accDescr: Diagram illustrating the traditional automotive manufacturing sequence from stamping through final assembly, highlighting wasteful door re-handling.
subgraph BodyShop["Body & Paint"]
A["Stamp Individual Panels"] --> B["Framing Station (Body-in-White)"]
B --> C["Hang Doors"]
C --> D["Full Body Paint Shop"]
end
subgraph AssemblyShop["Final Assembly Inefficiencies"]
D --> E["Remove Doors"]
E --> F["Install Interior Through Openings"]
F --> G["Underbody & Powertrain Lift"]
G --> H["Install Seats & Glass"]
H --> I["Reinstall Doors Once More"]
end
```
> “Henry Ford first invented this assembly line in 1922. It's been 100 years, and it's really hard to make a change after 100 years.” — Lars Moravy
> “That that was humbling, but at Tesla, it's not good enough.” — Lars Moravy
---
[48:26](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2906s) The Unboxed “Parallel” Manufacturing Process
Tesla began breaking the linear assembly mold on Model Y by introducing massive single-piece front and rear giga castings—deleting hundreds of stamped parts—and integrating the structural battery pack as the structural floor. By mounting the interior seats and center console directly onto the battery pack before marrying the pack to the chassis, Tesla reduced the final assembly line length by roughly 10%. For its next-generation platform, Tesla takes this parallel philosophy to its logical extreme: an “unboxed” manufacturing strategy where subassemblies are built and validated completely independently in parallel rather than in a sprawling serial line.
Instead of navigating around a bulky, five-meter vehicle, technicians and automated cells work on isolated subcomponents: the front casting, rear casting, individual side panels, and the battery pack floor. This yields a 44% increase in operator density and a 30% jump in space-time efficiency because operators spend their shifts actually assembling components rather than walking around empty car frames. Each side panel is painted and outfitted independently, the interior is loaded onto the floor without spatial restrictions via pure top-down access, and the painted exterior panels, doors, and floor meet only once at the very end of the line. Moravy clarifies that while the illustrative slide uses a Model Y body to demonstrate the concept, this unboxed method is engineered specifically for Tesla's upcoming next-generation platform.
```mermaid
flowchart TD
accTitle: Tesla Next-Generation Unboxed Manufacturing Process
accDescr: Parallel modular assembly flow where distinct vehicle subassemblies are built simultaneously and joined only once in final assembly.
subgraph ParallelWork["Simultaneous Modular Subassemblies"]
direction TB
P1["Front Casting & Subsystems"]
P2["Rear Casting & Drive Units"]
P3["Structural Battery Pack + Seats & Floor Interior"]
P4["Independent Left & Right Sides (Painted & Outfitted)"]
P5["Doors (Pre-Assembled)"]
end
subgraph FinalIntegration["Single-Step Final Assembly"]
ParallelWork --> M1["Single Marriage: Structural Floor Joins Front & Rear"]
M1 --> M2["Attach Outfitted Body Sides"]
M2 --> M3["Install Doors & Closures Once and Only Once"]
end
```
> “We balance parallel and series manufacturing in a way where we only do things that are necessary with a much shorter final line blocking a lot less of the entire rest of the factory so we can optimize material flow using the best practices.” — Lars Moravy
- **44% higher operator density** by working on open, parallel modules rather than inside fully framed vehicles.
- **30% improvement in space-time efficiency**, virtually eliminating empty handling steps and unnecessary transit time.
- Assembly robots gain direct vertical access (top-down / bottom-up) rather than contorting through window and door apertures.
- Factory footprint drops by more than **40%**, significantly lowering CapEx per unit of output.
- Combined with engineering and powertrain improvements, this next-gen process targets a **manufacturing cost reduction of up to 50%** (the “two-for-one” concept).
[51:58](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3118s) The Next-Generation Vehicle Tease and Colin Campbell’s Powertrain Introduction
Lars Moravy and Franz von Holzhausen wrap up their design segment by addressing the anticipation surrounding Tesla’s next vehicle platform. Lars mentions that while their track record proves they deliver industry-leading cars within tight constraints, the audience will have to wait for the actual reveal of the next-generation car: “I'd love to really show you what I mean and unveil the next-gen car. But you're going to have to trust me on that until a later date.” He promises that Tesla will always deliver exciting, compelling, and desirable vehicles. Shifting the focus to speed, Lars introduces Colin Campbell, head of powertrain engineering at Tesla.
Taking the stage to applause, Campbell underscores Tesla’s reputation for producing the fastest production vehicles available for the money, whether compared against electric or internal combustion competitors. Highlighting the Model S Plaid, he points out that its powertrain produces more than 1,000 horsepower, remarking that pound for pound, its electric motors generate power comparable to jet engines. Beyond sheer straight-line performance, Campbell stresses efficiency: Tesla vehicles achieve 25% to 30% more range than rival EVs in the same class on an equivalent energy basis. However, at Tesla, efficiency encompasses far more than watt-hours per mile; it governs the entire lifecycle of development, manufacturing, refinement, and scaling.
- **Vehicle Range Advantage**: Tesla vehicles travel 25% to 30% further on the same amount of battery energy than competing electric vehicles in their class.
- **Model S Plaid Output**: Over 1,000 horsepower with power-to-weight motor density rivaling jet engines.
- **Holistic Definition of Efficiency**: Broadened to include hardware weight reduction, supply chain simplification, capital expense minimization, and factory floor footprint.
---
[53:35](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3215s) Continuous Powertrain Evolution and Cross-Disciplinary In-House Engineering
Campbell illustrates Tesla's iterative engineering philosophy using the Model 3 and Model Y powertrain as the benchmark. Since rolling out the platform in 2017, relentless engineering changes have produced a drive unit that is 20% lighter for the exact same power output and uses 25% less heavy rare earth elements. Concurrently, the manufacturing footprint behind him at Gigafactory Texas saw dramatic reductions: the powertrain production facility is 75% smaller and 65% cheaper to construct than the original lines built in Fremont, all while matching the same output and vehicle performance.
```mermaid
flowchart TD
accTitle: Powertrain In-House Integrated Engineering Loop
accDescr: Cross-functional teams collaborate directly to design both hardware and manufacturing equipment simultaneously.
subgraph OptimizationOutput["Real-Time Systems Co-Design"]
E["Custom Hardware & Custom Silicon"]
F["Optimized High-Speed Simulators"]
G["Factory Floor & Tooling Architecture"]
end
A --> E
B --> E
B --> F
C --> G
D --> E
CrossDisciplinaryTeam -->|"Uncompromised Decision-Making"| OptimizationOutput
```
This rapid optimization stems directly from organizational structure. In contrast to legacy automakers who outsource inverters and subcomponents to tier-1 automotive suppliers, Tesla brings every discipline together under one roof. Campbell notes: “We have small and highly capable teams, and to make a critical decision, we can have the battery cell chemists, the mechanical engineers, the manufacturing engineers, the supply chain team, the automation designers, the software programmers, all in one room, working together in real time.” When a customer buys a premium German electric car, the inverter was conceived by outside contractors; at Tesla, the motor designers sit in the exact same room as the automation engineers designing the industrial tooling that will assemble that motor.
- **Model 3/Y Drive Unit Mass**: 20% mass reduction for identical power levels since the 2017 launch.
- **Heavy Rare Earth Reduction**: Reduced by 25% across the current Model 3/Y motor fleet.
- **Factory Footprint & Cost**: Modern powertrain factory lines are 75% smaller and 65% cheaper to install than the 2017 baseline while maintaining equal capacity.
- **Contrast with Legacy Auto**: Avoids fractured third-party Tier-1 supplier sourcing for critical power electronics and motor assemblies.
---
[55:25](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3325s) Power Electronics Innovations: Silicon Carbide and In-House Custom Silicon
Turning to internal hardware breakthroughs, Campbell focuses on power electronics. Every single electron propelling a Tesla passes through transistor packages within the inverter and charging system. Rather than adopting catalog solutions, Tesla engineered a proprietary transistor package capable of extracting twice as much heat as any off-the-shelf component. Doubling thermal extraction efficiency allows the internal silicon carbide (SiC) semiconductor wafer to be made substantially smaller. Because silicon carbide is exceptionally expensive and difficult to scale, shrinking the required die surface area represents an enormous financial and supply chain advantage.
```mermaid
flowchart LR
accTitle: Transistor Package and Processing Evolution
accDescr: Custom thermal packaging reduces SiC die area while custom silicon consolidates microprocessors.
subgraph SiliconCarbideReduction["Thermal & Wafer Optimization"]
H["Off-the-Shelf Transistor"] -->|"2x Thermal Extraction"| I["Custom Tesla Package"]
I -->|"Reduces Die Size"| J["Substantially Smaller SiC Wafer Area"]
end
subgraph MicroprocessorConsolidation["Compute Consolidation"]
K["4x Legacy Microprocessors"] -->|"Consolidation into 1 SoC"| L["1x Tesla Custom High-Power Microprocessor"]
L -->|"Impact"| M["50% Cost Reduction & Computational Agility"]
end
```
Simultaneously, orchestrating high-frequency transistor switching is an intensive computational challenge. Previous architectures required four discrete microprocessors to manage switching topologies. Tesla replaced all four with a single in-house custom microprocessor purpose-built for high-power electronics, halving component cost. Across the board, these power electronic refinements have enabled Tesla to cut both the weight and the cost of onboard charging systems by 50% relative to the original 2012 Model S baseline. Crucially, these high-voltage power electronics serve as the technological backbone not only for vehicles, but also for Tesla Superchargers and Megapack energy storage installations.
Software plays an equally critical role in drive unit optimization. Tesla developed proprietary physics simulation software to model the high-speed rotating magnetic fields inside the stator and rotor. While commercial simulation suites exist, Campbell emphasizes that Tesla’s internal software tools run faster and calculate electromagnetic behavior with far higher fidelity. This computational capability allows powertrain engineers to simulate and evaluate millions of motor permutations automatically, optimizing power density, thermal performance, total mass, manufacturing cost, and high-frequency acoustic characteristics.
- **Transistor Thermal Extraction**: Custom-designed packaging dissipates 2x more heat than commercial off-the-shelf modules.
- **Compute Consolidation**: A single custom Tesla microprocessor replaces 4 separate third-party controllers at 50% of the cost.
- **Charger Cost and Weight**: Cut by 50% in cost and 50% in mass compared to the 2012 Model S system.
- **Proprietary Simulation**: Custom electromagnetic simulation tooling explores millions of geometry iterations to minimize mass, cost, and audible drive unit noise.
---
[58:34](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3514s) Next-Generation Drive Unit: 75% Less SiC, Zero Rare Earths, and $1,000 Unit Cost
Campbell reveals the engineering specifications for Tesla’s next-generation powertrain platform, explicitly designed to eliminate supply bottlenecks and drive down electric vehicle prices for mass adoption. The new drive unit reduces silicon carbide utilization by 75% compared to current architectures without degrading car efficiency or performance. Furthermore, the powertrain is engineered to accept any battery cell form factor and chemistry, freeing manufacturing from specific supply chain dependencies.
Financially and physically, the next-gen platform represents an unprecedented step forward: Campbell states that Tesla has reduced the overall drive unit manufacturing cost down to roughly $1,000, asserting that no other automaker comes close to that price point. Additionally, the next powertrain manufacturing line will occupy 50% less factory footprint than the already streamlined Gigafactory Texas line, enabling rapid expansion of global vehicle assembly plants.
The most notable sustainability milestone centers on magnetics. As global demand for clean energy solutions accelerates, mining rare earths presents escalating supply bottlenecks along with severe environmental and occupational hazards. Campbell announces a major breakthrough: “We have designed our next drive unit, which uses a permanent magnet motor, to not use any rare earth materials at all.” This breakthrough allows Tesla to scale volume to millions of units annually without relying on constrained critical minerals. Campbell concludes by thanking the dedicated powertrain team before welcoming Pete Bannon and David Lau to the stage.
| Metric / Parameter | Current Generation (Model 3 / Model Y) | Next-Generation Powertrain Platform |
|:--- |:--- |:--- |
| **Silicon Carbide (SiC) Content** | Baseline usage in inverter | **75% reduction** in SiC transistor content |
| **Permanent Magnet Chemistry** | Reduced heavy rare earth elements | **0% rare earth materials** (complete elimination) |
| **Battery Chemistry Support** | Specific cell lines / chemistries | **Universal compatibility** with any cell chemistry |
| **Drive Unit Total Cost** | Substantially higher than $1,000 | **~$1,000 per unit** |
| **Factory Floor Footprint** | Already 75% smaller than 2017 Fremont line | **50% smaller footprint** than current Giga Texas line |
---
[01:01:16](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3676s) Electronic Architecture: Transition from Legacy Wiring to Integrated Control
Pete Bannon steps forward to introduce Tesla's low-voltage electronic architecture, opening with a slide showing the sprawling low-voltage system of the original 2012 Model S. That early vehicle was primarily constructed using third-party vendor-sourced electronic control units (ECUs) integrated together by Tesla engineers. The car contained over 300 distinct low-voltage devices, ranging from trivial components like the glove box courtesy lamp to high-bandwidth infotainment computers and critical life-safety actuators, including airbags, steer-by-wire, and brake-by-wire subsystems.
Bannon explains that traditional automotive low-voltage wire harnesses rely on hundreds of individually measured, hand-cut, and crimped wires snapped into plastic connector housings. He characterizes this industry-standard harness assembly method as an inherently manual, tedious, and error-prone process that cannot scale efficiently to meet Tesla’s hyper-scale manufacturing targets. Moving into future vehicle architectures, Tesla's primary low-voltage directive is drastically simplifying wiring harness length, cutting total mass, eliminating manual manufacturing steps, and redesigning physical controller placement throughout the chassis.
- **2012 Model S Architecture Baseline**: Relied on vendor-sourced ECUs driving more than 300 distinct low-voltage endpoints across the car.
- **Spectrum of Low-Voltage Endpoints**: Ranges from ambient cabin convenience lighting to steering, braking, and airbag safety loops.
- **Manufacturing Bottleneck**: Traditional wire harness manufacturing remains highly manual, cut-and-crimped, error-prone, and resistant to robotic automation.
[01:02:14](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3734s) Wire Harness Simplification and In-House Controllers
The section begins by examining the intense engineering friction caused by traditional vehicle wire harnesses. In the early stages of bring-up and vehicle development, dense wire harnesses introduce extraordinary complications: when a subassembly fails to operate, engineers struggle to isolate whether the root cause is software, a controller, the processor, or any of the myriad endpoints hidden inside the physical entanglement. Debugging must happen across every layer at once. Pete Bannon takes over to outline the progression of internal architectural design since the early days of Model S.
```mermaid
flowchart TD
accTitle: Evolution of Tesla Controller Design and In-House Share
accDescr: Progression showing controller development from Model S through next-gen platforms.
subgraph EarlyGen["Early Platforms"]
S["Model S: Traditional Architecture"]
M3["Model 3: Controller merging<br/>-17 kg wiring harness"]
end
subgraph UnifiedGen["Shared Architectures"]
MY["Model Y: Enhanced controllers"]
HeatPump["Model 3 Refresh: Heat pump integration<br/>Shared supply chain"]
MSX["2021 Model S & X Refresh"]
end
S --> M3
M3 --> MY
MY --> HeatPump
HeatPump --> MSX
MSX --> CT
CT --> NextGen
```
Starting with Model 3, Tesla merged separate electronic modules together to cut weight and bundle functionality. Between Model S and Model 3 alone, Tesla trimmed 17 kg from the wire harness. Bannon emphasizes how huge a deal this is by revealing an internal incentive program: “our VP of engineering Lars will deliver a bottle of your favorite spirit to your desk if you manage to save a kilogram of weight from the car.” He quips that this harness reduction cost Lars quite a lot of liquor. The controller improvements made for Model 3 were enhanced for Model Y, then backported to Model 3 when the heat pump was introduced, establishing shared controllers across both product lines to simplify component sourcing. These designs were later enhanced for the 2021 Model S and Model X refreshes.
For Cybertruck, Tesla designed 85% of all onboard controllers internally. On the upcoming next-generation platform, Bannon states: “For the next-gen platform, we're going to finish the job and be designing 100% of the controllers to give us full control of the design and the supply chain at the component level.” This component-level ownership proved crucial during recent automotive supply chain crises and directly feeds back into software capabilities, allowing Tesla to ship features that were never conceived when the hardware was originally laid down. David Lau chimes in to add that Sentry Mode was one of his favorite examples of a last-minute feature enabled by this flexibility.
- 17 kg of wiring mass removed between Model S and Model 3.
- Model 3 and Model Y share electronic controllers, simplifying the global supply chain.
- 85% of Cybertruck controllers designed in-house; targeting 100% for the next-generation vehicle architecture.
---
[01:04:33](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3873s) eFuses, Lithium-Ion Auxiliary Batteries, and Continuous Refinement
Starting in 2017 with Model 3, Tesla replaced physical electromechanical relays and fuses with solid-state eFuses (electronic fuses). Solid-state transistors give low-level software fine-grained control over vehicle power, enabling dynamic capabilities such as load shedding during extreme brownouts or voltage dips, automated software retries for transient faults, and deep historical telemetry of power draws. Lau explains that this approach fundamentally alters vehicle design: “software-controlled hardware is fundamentally about being able to sidestep what otherwise would have been static trade-offs between one attribute or another in a piece of hardware.” Software infuses hardware with intelligence and context-specific behavior rather than locking a component into an arbitrary worst-case operating envelope.
In 2022, Tesla fully phased out traditional 12-volt lead-acid auxiliary batteries across its fleet in favor of low-voltage lithium-ion units fitted with a custom toolless connector. The lithium-ion auxiliary battery is designed to last the entire lifespan of the vehicle, permanently ending sudden dead-battery breakdowns. The toolless connector simplifies factory assembly and field service, incorporating built-in software validation that confirms the connector is seated and locked correctly before a service ticket can be closed.
Bannon compares Tesla's persistent engineering ethos to peeling a carrot: you shave off tiny slivers that seem negligible on their own, but continuous effort yields massive cumulative gains. As an illustration, he walks through the iterative evolution of the center touchscreen between 2017 and 2023.
[01:07:03](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4023s) The 48-Volt Transition and Ethernet-Linked Zonal Controllers
Low-voltage vehicle architecture has been stranded at 12 volts for over sixty years. Because modern car amenities and compute loads draw mounting power, 12-volt systems force electrical harnesses to carry currents exceeding 200 amps, requiring heavy, thick copper runs. Bannon announces that beginning with Cybertruck, all future Tesla vehicle platforms will transition entirely to a 48-volt low-voltage bus. Moving to 48 volts cuts necessary current by a factor of 4. Because resistive power loss is proportional to the square of current ($I^2R$), dropping current fourfold slashes wiring power losses by a factor of 16 ($4^2 = 16$).
```mermaid
flowchart LR
accTitle: 48-Volt Architecture Physics and Packaging Advantages
accDescr: Diagram illustrating how quadrupling voltage drops power dissipation and allows smaller wires and controllers.
subgraph Physics["Electrical Scaling"]
V["48V Architecture<br/>(4x Voltage)"] --> I["Current reduced by 4x<br/>(I / 4)"]
I --> P["Power loss: P = I²R<br/>16x reduction in lost heat"]
end
subgraph Packaging["Hardware Impact"]
P --> Wire["Thinner, lighter wire harnesses"]
P --> Silicon["Smaller eFuses & controllers"]
P --> Thermal["Heat sinks shrunk or eliminated"]
end
```
This drastic reduction in thermal loss allows engineers to shrink wire diameters, reduce controller footprints, downsize eFuses, and eliminate bulky heat sinks entirely. Bannon declares: “48 volts is the future for low voltage design at Tesla and likely the rest of the industry in due course,” openly challenging the rest of the automotive industry and component suppliers to follow Tesla's lead.
Alongside the voltage shift, Tesla is redesigning the network typology from centralized wiring to zonal controllers connected via high-speed Ethernet. In legacy cars, individual wires run across the length and width of the chassis from central body controllers out to single sensors, motors, or lamps. In the Cybertruck design, local endpoints wire directly to the nearest local zone controller, which packets the signal and routes it over an ultra-reliable, low-latency, low-jitter Ethernet backbone. This cuts out nearly all cross-car wiring runs; the next-gen platform will eliminate cross-car low-voltage wiring entirely. Unlike static CAN bus loops, this Ethernet architecture allows instantaneous reconfiguration of network routing and enables engineers to inspect the entire car through a single diagnostic port.
- 48-volt bus replaces the 60-year-old 12-volt industry standard starting on Cybertruck.
- $4\times$ drop in current yields a $16\times$ reduction in low-voltage resistive heat loss.
- Zone controllers convert analog/local signals to network packets over high-speed Ethernet, eliminating cross-car wiring harnesses.
---
[01:09:50](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4190s) Software Architecture, OTA Depth, and Telemetry Loops
David Lau, Tesla's VP of Software Engineering, steps up to discuss the software foundations built since he joined the company in 2012. From the earliest production runs of the Model S, Tesla engineered full Over-the-Air (OTA) update support across every onboard electronic controller, whereas other automakers were barely beginning to update isolated infotainment head units. Combined with deep, anonymized telemetry streaming back from the operating fleet, this closed-loop architecture enables Tesla to monitor real-world vehicle health, detect operational bottlenecks, and rapidly iterate on software based on empirical evidence rather than theoretical assumptions.
These aggregated fleet insights inform both code updates and hard mechanical choices. Lau and Bannon note that fleet telematics revealed that customers rarely opened their motorized sunroofs, leading Tesla to cleanly delete panoramic sliding roofs entirely in favor of fixed glass panels. Similarly, real-world crash telematics feed directly into safety design, allowing Tesla engineers to tune restraint systems and deploy airbags based on actual vehicle dynamics gathered from live collisions rather than relying strictly on standardized regulatory laboratory crash tests.
[01:11:09](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4269s) Fleet Data, Real-World Safety, and Vertical Software Integration
David Lau explains how Tesla uses real-world telemetry from its fleet to simulate and validate vehicle changes far beyond standard regulatory requirements. Displaying a scatter plot, he points out that gray dots represent crash tests mandated by global regulatory and consumer ratings agencies, whereas the colored area enveloping and extending well past those dots represents actual collisions documented across the fleet. Tesla designs and tests against this comprehensive real-world dataset. By leveraging this field telemetry, Tesla updated the seat belt tensioning algorithm over-the-air (OTA) across the entire fleet to reduce passenger injuries. Furthermore, daily data streams allow engineers to analyze fleet driving and charging habits to optimize battery pack sizing for next-generation vehicles.
```mermaid
flowchart TD
accTitle: Real-World Fleet Data Feedback Loop
accDescr: Telemetry from driving and crash data informs simulations and updates over the air.
subgraph DataCollection["Fleet Operations"]
A["123M Miles Driven Daily"]
B["1.9M Daily Charge Sessions"]
C["Real-World Crash Telemetry"]
end
subgraph SimulationAndDesign["Engineering & Simulation"]
D["Recreate Crashes in Simulation"]
E["Optimize Battery Pack Sizing"]
end
subgraph OTA["Fleet Delivery"]
F["Seatbelt Tensioning Algorithm"]
G["Over-the-Air Updates"]
end
A --> D
B --> E
C --> D
D --> F
F --> G
G --> DataCollection
```
David contrasts Tesla's integrated approach with traditional automotive software development, where a vehicle's controllers are sourced from Tier 1 suppliers who outsource to Tier 2 and Tier 3 vendors. In that fragmented model, even minor cross-component modifications require months of coordination before work begins. Instead, Tesla treats the vehicle, backend server infrastructure, and the continuous feedback loop from the fleet as a single unified system:
> “Most engineers think about the vehicle as a complete consolidated fully contained system. But in the software team, we think about the system as including the vehicle as well as all of our back-end server-side applications and infrastructure and the resultant feedback loop from the entire fleet that informs all the decisions we make as an engineering team.”
- Daily telemetry includes over 123 million miles driven and 1.9 million charging sessions.
- Real-world crash simulations run automatically every time a software or hardware change is made to airbags or restraint systems.
---
[01:13:56](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4436s) Predictive Air Suspension and Background Shadow Testing
To illustrate vertical hardware-software synergy, David outlines a feature deployed on Model S and Model X that automatically raises the air suspension for comfort before encountering rough road surfaces. Implementing this required coordinating five distinct vehicle and server components: the Restraint Control Module's Inertial Measurement Unit (IMU) senses road roughness; the Autopilot computer localizes the car via GPS; backend navigation servers aggregate anonymized fleet telemetry to map roughness; the onboard navigation engine predicts upcoming road conditions along the route; and the air suspension controller adjusts the ride height in real time.
```mermaid
flowchart LR
accTitle: Predictive Air Suspension Architecture
accDescr: Data flow from vehicle sensors to navigation servers and back to the air suspension controller.
subgraph SenseAndLocalize["In-Car Sensing"]
A["Restraint Control Module (IMU)"]
B["Autopilot Computer (GPS)"]
end
subgraph CloudProcessing["Tesla Backend"]
C["Navigation Servers & Telemetry Aggregation"]
end
subgraph Actuation["Vehicle Actuation"]
D["Onboard Navigation Engine"]
E["Air Suspension Controller"]
end
A -->|"Road Roughness"| C
B -->|"Location Telemetry"| C
C -->|"Annotated Map Data"| D
D -->|"Predict Rough Road Ahead"| E
```
Before releasing this feature to the public, Tesla deployed prototype algorithms to run passively in shadow mode across the fleet. This silent testing gathered anonymous data on when the feature would have engaged across varying road environments. Pete Bannon jumps in to underscore the value of this architecture for safety-critical functions:
> “Yeah, being able to test new algorithms in the background without an impact in the car is is really critical ability for us, especially for safety critical things like automatic emergency braking.”
Pete adds that this silent background validation was used in 2017 to deploy stability control algorithms across the entire Model 3 fleet with extreme confidence.
---
[01:15:53](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4553s) Manufacturing Integration, Unboxed Assembly, and Automated Service
Software practices also govern Tesla’s internal manufacturing and service operations. Recalling the early days of Model S production, David notes how easy it was to make assembly mistakes—such as failing to seat a connector or pulling an incorrect part from a bin. To eliminate assembly errors on Model 3 and future platforms, Tesla adopted the core software principle of "test early and test often" on the factory line.
Whenever a line associate connects a subassembly to a vehicle, the central car computer detects the connection immediately. It verifies part compatibility, pushes any required software updates, validates calibrations, and executes diagnostics across interconnected subcomponents. If a physical defect is detected, the car's screen flags it and alerts backend command systems so a worker can resolve it on the line before the next assembly stage covers it up.
Pete notes that subassembly testing using Tesla-designed controllers is essential for Lars Moravy’s unboxed vehicle assembly strategy, ensuring every pre-assembled section is fully validated before final marriage.
In customer service, Tesla combines onboard vehicle diagnostics, natural language processing (NLP) on customer-submitted appointment notes, and custom internal software to streamline repair logistics. David reports that this automated pipeline currently diagnoses issues, schedules service, and pre-orders replacement parts for over 33% of all customer service concerns without human pre-triage.
---
[01:18:06](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4686s) Building the Autonomous Fleet Management Platform
David shifts focus from autonomous driving models to the backend platform infrastructure necessary to run a global autonomous fleet. Over several years, Tesla has been deploying foundational building blocks directly into consumer and fleet operations. In 2021, Tesla introduced mobile app phone key sharing via text or email invitation, followed by cloud profile synchronization for seat positions, mirrors, steering settings, media preferences, and navigation destinations across vehicles.
For internal operations, Tesla employees use a proprietary app to locate, monitor, and drive company-owned vehicles across manufacturing, logistics, and delivery sites. Similarly, when a customer receives a service loaner, the vehicle is dynamically provisioned directly to their Tesla app account. With synchronized cloud profiles and mobile key credentials, the loaner mirrors their personal car's setup the moment they open the door.
David notes that all interactions rely on end-to-end encryption and cryptographically signed commands, ensuring security and privacy while ensuring vehicles only execute authenticated instructions. He concludes:
> “So, that's a bit of a preview of all the pieces that we're building looking ahead to our future of autonomous fleets. Synchronization, permissions management, security, and privacy.”
With that, David concludes the software overview and invites Ashok Elluswamy to the stage to present the latest updates on Full Self-Driving (FSD).
[01:20:33](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4833s) Full Self-Driving and the Sustainable Future
Ashok Elluswamy takes the stage, noting that he joined Tesla in 2014 and has been tackling Autopilot and self-driving engineering for nearly nine years. He anticipates the skepticism from investors wondering how vehicle autonomy fits into a master plan focused on global energy sustainability. The answer lies in vehicle utilization: cars spend the vast majority of their operational lives parked idle. Once true autonomous capability is unlocked, a vehicle can continuously serve other passengers instead of sitting motionless in parking lots. This dramatic increase in per-vehicle utility fundamentally diminishes the total number of cars that must be manufactured to fulfill transportation demand, conserving raw materials and energy at a planetary scale.
Elluswamy acknowledges the staggering technical difficulty of solving autonomy, calling it “one of the hardest real-world AI problems out there right now.” Rather than attempting narrow, geofenced solutions, Tesla focuses on a fully general approach. To build a scalable self-driving system, he outlines three essential pillars that must be executed correctly: AI system architecture, data, and compute.
[01:22:04](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4924s) AI Architecture: From Brittle Vision to Foundation Models and Neural Planners
Addressing the first pillar, Elluswamy explains how Tesla systematically transitioned its neural architecture. In the platform's early iterations, the system relied on single-camera, single-frame neural networks whose outputs were stitched together through heuristic, rule-based post-processing. Because this setup proved fragile, Tesla replaced it with multi-camera temporal video neural networks. These unified networks process synchronized, real-time video streams from all eight vehicle cameras simultaneously, directly projecting into a unified 3D vector space. This single representation handles diverse tasks, including occupancy tracking, road curvature, lane topology, obstacle dynamics, and traffic control signals.
```mermaid
flowchart TD
accTitle: Tesla Autopilot Vision and Planning Stack
accDescr: Diagram illustrating eight camera feeds feeding temporal video networks and downstream neural planning.
subgraph SensorInput["Vehicle Sensors"]
CAM["8 Hardware Cameras<br>(Real-Time Video Stream)"]
end
subgraph Perception["Unified 3D AI Perception"]
VNN["Multi-Camera Video Neural Networks<br>(Transformers & Occupancy Networks)"]
OCC["3D Occupancy Grid & Motion Vectors"]
LAN["Topological Lane Connectivity & Semantics"]
end
subgraph Planning["Trajectory & Motion Control"]
NNP["Neural Network Planner<br>(Search Budget: ~50 ms)"]
ACT["Vehicle Control & Trajectory Actuation"]
end
CAM --> VNN
VNN --> OCC
VNN --> LAN
OCC --> NNP
LAN --> NNP
NNP --> ACT
```
To demonstrate the capability of this architecture, Elluswamy showcases Tesla's occupancy network, which reconstructs obstacle volume and velocity in real time. The model precisely traces the swerving, erratic motion of an adjacent semi-truck, feeding accurate kinematic vectors to the path planner to safely preserve clearance. For abstract spatial reasoning—such as lane connectivity and complex junction topologies where naive vision methods falter—Tesla incorporates techniques borrowed from natural language processing and reinforcement learning:
> “These networks use similar techniques such as transformers, attention modules, auto-regressive modeling of tokens similar to what large language models like ChatGPT do out there.”
This machine-learned end-to-end perception replaces brittle manual logic with rich environmental tokens. Tesla extends this approach to motion planning. Traditional heuristic pathing struggles in high-density urban environments where turning maneuvers require yielding to multi-directional vehicular traffic and pedestrians while balancing assertiveness and comfort. Computing thousands of interacting candidate paths under standard algorithms would consume roughly 10 milliseconds per individual configuration, making real-time search impossible. By formulating urban trajectory optimization as an AI-driven search problem, Tesla compresses the entire multi-agent calculation into an onboard real-time compute budget of 50 milliseconds.
[01:24:54](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5094s) The Fleet Data Engine, Auto-Labeling, and Simulation
Elluswamy moves to the second pillar: data. While Tesla possesses an unmatched advantage in its massive connected customer fleet, raw video alone is not usable for training deep neural networks. Supervising complex multi-camera models requires massive quantities of accurate ground-truth labels. Human annotation teams simply cannot scale to produce the necessary volume of high-dimensional 3D annotations. Tesla resolved this bottleneck by engineering an automated labeling pipeline running in back-end data centers.
The auto-labeling engine ingests multi-trip video clips recorded by different vehicles traversing the exact same real-world locations. It aggregates these distinct camera trajectories to generate rich 3D spatial reconstructions, faithfully resolving lane boundaries, curbs, crosswalk geometries, and painted asphalt lettering. Once a photorealistic 3D baseline is constructed, Tesla injects it into a physics simulator. This platform allows engineers to introduce synthetic edge cases, generating adverse weather conditions, blinding glare, and aggressive agent behavior that rarely occur in natural driving.
```mermaid
flowchart LR
accTitle: Tesla Autopilot Data Engine Flywheel
accDescr: The closed-loop cycle of fleet data mining, automated labeling, retraining, and software deployment.
A["Identify Edge Case / Failure Modes"] --> B["Mine Customer Fleet for Video Clips"]
B --> C["Compute Center Auto-Labeling & 3D Reconstruction"]
C --> D["Augment with Synthetic Simulation Scenarios"]
D --> E["Train Large Video Neural Networks"]
E --> F["Deploy Over-The-Air to Fleet"]
F --> A
```
Elluswamy details how this data engine systematically eliminates operational edge cases, citing an issue where vehicles exhibited phantom braking around parked cars:
- Autopilot previously mistook parked roadside vehicles for moving obstacles cutting into the travel lane, triggering defensive braking.
- Engineers queried the fleet to retrieve 14,000 video instances depicting false braking near parked vehicles.
- The auto-labeling pipeline processed the clips, and networks were retrained with the expanded corpus.
- The updated model learned subtle contextual cues—such as an empty driver's seat—to correctly deduce that the target vehicle was parked, resolving the phantom decelerations.
Elluswamy emphasizes the predictability of this flywheel: “On the chart on the right side, you can see that every time we add data, the performance improves.”
[01:28:10](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5290s) In-Car Hardware, Training Compute Infrastructure, and FSD Rollout
Executing this architecture and data flywheel requires vast computational resources, both in the car and across Tesla's data centers. Inside every customer vehicle, the custom Autopilot computer delivers roughly 150 TOPS of compute to evaluate vision models and planners in real time. In the cloud, model training and label generation run on dedicated supercomputing clusters:
- **14,000 GPU Cluster:** Tesla operates a 14,000-GPU cluster in its data center, split between automated fleet data labeling (roughly 30%) and deep learning model training (roughly 70%).
- **Storage Cache:** The pipeline currently processes against a 30-petabyte video caching cluster, which is rapidly expanding toward 200 petabytes.
- **Training Scale:** The occupancy network alone digested 1.4 billion video frames during training.
- **Dojo Integration:** Elluswamy notes that these cloud compute figures will surge dramatically once Tesla's custom-built Dojo training computer comes fully online.
Regarding public deployment, Elluswamy states that Tesla has deployed Full Self-Driving (FSD) Beta to virtually all customers who purchased it in the United States and Canada, representing approximately 400,000 vehicles. While the software remains Level 2 supervised, it handles unprotected turns, traffic light navigation, pedestrian yielding, and end-to-end trip execution. Fleet tracking indicates that drivers operating with FSD Beta engaged are roughly five to six times safer than the US national driving average. Solving real-world AI, he concludes, is what will ultimately unlock true driverless capability and transform vehicle fleet economics.
[01:30:12](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5412s) Tesla Bot (Optimus) Update: Actuator Progression and Locomotion
Elluswamy invites Elon Musk back to the stage to present an update on the Tesla Bot (Optimus) humanoid robotics program. The presentation transitions to a video demonstration showing Optimus hardware undergoing development in the laboratory. Musk reacts to the visual of disembodied mechanical limbs cycling through bench tests, observing with dry humor, “It's kind of weird seeing the arms and legs just separate. We have a whole lab full of arms and legs.”
Musk puts the visual in perspective by reminding investors of the rapid development cadence since the robot's public unveiling:
> “Bearing in mind that when we did AI Day, this version of Optimus didn't walk at all. So, the rate of improvement here I think is quite significant.”
He jokes that while the robot is obviously not leaping across obstacles or performing parkour, the humanoid platform is now independently walking around Tesla facilities, using in-house actuators, custom battery packs, and the same real-world AI vision computer that powers the vehicular Autopilot stack.
[01:31:18](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5478s) Optimus, Real-World AI, and Economic Implications
Elon Musk explains that Tesla's primary differentiator in robotics is its leadership in real-world artificial intelligence. Because a Tesla vehicle can be understood essentially as a "robot on wheels," the transition to Optimus is simply creating a "robot on legs." The very same computer hardware and software running Tesla's Full Self-Driving stack are placed directly into Optimus. Musk points out that humanoid robots have little practical utility if an engineer must manually program every individual action; the machine must navigate spaces autonomously, complete tasks dynamically, and understand plain instructions either visually by demonstration or through direct spoken commands.
Beyond software, Tesla applies its core competencies in design-for-manufacturing directly to humanoid hardware. When attempting to source components for Optimus, the team was surprised to discover that none of the vast numbers of off-the-shelf motors and gearboxes existing globally were suitable for a humanoid robot. As a result, the electric motors, gearboxes, power electronics, and battery packs inside Optimus are completely custom-designed. The powertrain engineers responsible for the high-performance motors in the Model S Plaid designed the robotic actuators, positioning Tesla to scale useful humanoid robots faster than any competitor.
Looking forward, Musk suggests the potential ratio of humanoid robots to humans could realistically exceed one to one, given extensive industrial and domestic applications. In such a scenario, traditional economic definitions cease to apply, as gross domestic product has historically been measured as output per person multiplied by the total human population. If production output per unit becomes detached from human labor constraints, economic growth potential becomes virtually boundless. While acknowledging civilization remains far from advanced Kardashev scale thresholds, Musk predicts Optimus will ultimately become significantly more valuable to Tesla long-term than its entire vehicle business.
“The thing that's I think Tesla brings the table that others don't have is that we have um we have the uh real-world AI. I We're We're the most advanced in real-world AI.” — Elon Musk
- **Optimus Hardware**: Powered by the same autonomous vehicle computer and neural net architecture used in Tesla cars.
- **Actuators**: 100% custom-designed by the Tesla powertrain team (including Plaid motor designers), as zero off-the-shelf commercial components met efficiency and packaging constraints.
- **Economic Scale**: Potential for humanoid robots to outnumber humans (>1:1 ratio), unlocking virtually unlimited economic output.
```mermaid
flowchart LR
accTitle: Tesla Real-World AI and Hardware Flywheel
accDescr: Diagram illustrating how Tesla transfers automotive AI and manufacturing advantages directly into the Optimus humanoid robot.
subgraph CoreTech["Tesla Core Technology"]
AI["Real-World AI & FSD Computer"]
PT["Plaid Powertrain Engineering"]
MFG["Design for High-Volume Manufacturing"]
end
subgraph Applications["Hardware Platforms"]
Car["Tesla Vehicles<br>('Robot on Wheels')"]
Optimus["Optimus Humanoid<br>('Robot on Legs')"]
end
AI --> Car
AI --> Optimus
PT --> Optimus
MFG --> Optimus
Optimus --> Home
Optimus --> Ind
Home --> Econ
Ind --> Econ
```
---
[01:35:40](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5740s) Scaling Global Charging and Manufacturing Efficiencies
Rebecca Tinucci takes the stage to present Tesla's charging infrastructure strategy, emphasizing that an effortless charging experience has been treated as the fundamental linchpin of EV adoption since day one. Tesla addresses charging holistically across both residential (single-family homes, apartments, and condominiums) and en-route scenarios (daily commuting and long-distance road trips). In 2022, Tesla delivered 9 terawatt-hours of total charging energy, with more than 50% supplied through residential AC wall connectors. Away from home, drivers had access to an extensive footprint of 80,000 public Tesla charging points worldwide, including 40,000 DC Superchargers.
Having built infrastructure for a decade when other automakers avoided capital investment, Tesla developed industry-leading cost structures. Tesla’s deployment costs are typically 20% to 70% below competitors for both Supercharger installations and AC hardware lines. This advantage stems from deep vertical integration: designing and manufacturing internal hardware, utilizing shared components across product ecosystems, and directly executing site construction and management. In Buffalo, New York, Giga New York pre-assembles four-post Supercharger units onto pre-cast concrete slabs. These pre-fabricated units are trucked directly to destination sites and hoisted into position with a crane, slashing site deployment expenses by 15% and reducing physical on-site installation timelines to mere days.
“Our costs are often 20%, if not 70% lower than alternatives, and that goes for both our supercharging hardware and deployments, and our AC charging product lines.” — Rebecca Tinucci
- **2022 Energy Delivered**: 9 TWh across all charging methods.
- **AC Home Dominance**: Greater than 50% of all delivered energy originated from home AC chargers.
- **Global Network Scale**: 80,000 total charging points, including 40,000 Supercharger stalls.
- **Pre-Fabricated Supercharger Units**: Four-post modules built at Giga New York, trucked on-site, and craned in, reducing installation time to days and capital deployment costs by 15%.
---
[01:38:02](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5882s) Network Utilization, Trip Planner, and Network Opening
Operating sites efficiently is equally critical to driving down the cost curve. Over recent years, Tesla has decreased its operating cost per kilowatt-hour by 40%, largely through optimizing site utilization—the volume of charging sessions and kilowatt-hours pushed through each individual post. Because maximizing utilization risks causing customer wait times, Tesla relies on its proprietary vehicle Trip Planner. Unlike third-party network operators who only see site status, or rival automakers who only monitor car battery telemetry, Tesla holds end-to-end visibility across both vehicle location and real-time station congestion.
By actively routing traveling vehicles toward underutilized stations and away from congested locations, Trip Planner increased global site utilization by 30% while simultaneously cutting customer wait times in half. Average session visit durations were concurrently reduced by 30% through coordinated improvements across battery preconditioning software, vehicle charge curves, and dispenser hardware. Looking forward, Trip Planner will transition from reactive, real-time site updates to predictive occupancy modeling by analyzing the stream of vehicles actively navigating toward specific stations, effectively acting as an air traffic control system for global charging.
Looking toward the overarching goal of Master Plan 3—a completely electrified global fleet requiring an estimated 9 petawatt-hours of charging annually—Tesla has begun opening its Supercharger network to non-Tesla EVs globally. Over 50% of European Superchargers are open to all EVs, initial expansion has begun in Australia for the Asia-Pacific market, and the first 10 pilot stations opened in the United States just prior to the presentation. For non-Tesla drivers, sessions are initiated seamlessly via the Tesla smartphone app. To accommodate varying vehicle inlet standards in North America without replacing existing cables, Tesla engineered the "Magic Dock," an integrated hardware adapter built directly into the stall that automatically couples with the CCS connector when requested by non-Tesla vehicles.
“Ultimately, our vision for Trip Planner is that it's the air traffic controller for electric vehicle charging across all infrastructure on a global basis.” — Rebecca Tinucci
- **Cost and Time Reductions**: Per-kWh charging operating costs cut by 40%, stall utilization raised by 30%, average site visit times decreased by 30%, and wait times reduced by 50%.
- **Future Fleet Demand**: An estimated 9 PWh of annual charging capacity required for a fully electrified global transport fleet.
- **Network Opening**: Over 50% of Supercharger stalls opened to all EVs across Europe, first sites opened in Australia, and the initial 10 pilot stations activated in North America using integrated Magic Dock adapters.
[01:41:30](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6090s) Next-Gen Supercharging and Shifting to Daytime Renewables
Concluding her presentation on charging infrastructure, Rebecca Tinucci explains the deployment of Tesla's fourth-generation (V4) Supercharger posts, which began rolling out in Europe first. While seemingly subtle, these new posts incorporate longer cables specifically designed to reach charge ports on non-Tesla electric vehicles regardless of where they are positioned on the car. Beyond expanding physical access, the larger strategic priority for Master Plan Part 3 is ensuring that the global charging load is powered cleanly: over the preceding two years, Tesla procured enough renewable energy to offset 100% of customer charging, but the long-term goal requires reshaping charging habits so electricity demand aligns with when renewable generation actually occurs.
Across aggregate grid models in the United States, matching the solar and wind generation profile means moving significantly more charging to daytime hours. To accomplish this, Tesla's strategy focuses on blanketing the locations where vehicles are naturally parked throughout the day with low-cost, convenient AC destination charging. While Superchargers and home charging remain foundational pillars, daytime AC charging allows vehicles to soak up midday renewable power directly. Supporting Master Plan Part 3 will ultimately require scaling industry-wide annual capacity to 9 petawatt-hours (PWh), opening the network to non-Tesla vehicles, powering the fleet with renewables, and continually elevating the charging user experience.
- **V4 Supercharger Posts**: Debuting in Europe with longer cables to eliminate physical compatibility issues for non-Tesla vehicle port locations.
- **Renewable Offset**: Tesla has procured enough renewable energy over the past two years to offset 100% of global customer charging demand.
- **Master Plan Part 3 Scale**: The global electric vehicle transition requires scaling industry-wide battery and vehicle charging capacity to 9 PWh annually.
Rebecca Tinucci introduces Karn Budhiraj and Roshan Thomas to present Tesla’s supply chain strategy. Karn introduces his role managing supply chain operations for electronics, powertrain, and batteries, alongside indirect purchasing, construction procurement, and global warehousing and distribution (noting he has been with Tesla for 10 years). Roshan, a 14-year Tesla veteran, oversees vehicle commodities, solar, logistics, supply planning, and capital equipment. Together, they explain how Tesla’s supply chain differs fundamentally from a traditional automotive procurement model. Because Tesla designs so many of its subcomponents from scratch rather than purchasing off-the-shelf parts, the relationship with suppliers cannot be treated as a purely commercial or transactional interaction.
Instead, Tesla relies on an internal organization called Supply Industrialization Engineering (SIE). Comprising mechanical, electrical, and industrial engineers, this team takes an engineering drawing, designs the manufacturing concept, selects the necessary equipment, and physically embeds themselves inside supplier facilities for weeks or months at a time to bring up production lines from scratch. Once lines achieve targeted rates, yields, and cost structures, the team works on automation and further yield optimization. This hands-on integration gives Tesla real-time visibility across approximately 10,000 supplier factories worldwide. As Karn Budhiraj observes:
> “Now, this is a huge strategic advantage because we manage every detail of our supply chain, because we're so integrated with our partners, that we know about issues before they happen.”
```mermaid
flowchart TD
accTitle: Supply Industrialization Engineering Flow
accDescr: Diagram showing how Tesla takes a custom subcomponent drawing through line bring-up to automated production.
subgraph BringupPhase["Bring-up Phase"]
B --> C["Manufacturing Concept &<br/>Equipment Selection"]
C --> D["On-Site Bring-Up<br/>(Weeks/Months at Supplier)"]
end
subgraph ScalingPhase["Scaling Phase"]
D --> E["Achieve Rate, Quality &<br/>Yield Targets"]
E --> F["Automation & Yield<br/>Improvement"]
end
```
---
[01:46:44](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6404s) The Reality of In-House Manufacturing and Supplier Consolidation
Managing supply chains at Tesla is an unforgiving discipline because the company manufactures its vehicles in-house rather than outsourcing production to contract manufacturers. In this environment, missing a single component stops an entire vehicle assembly line, escalating immediately across the organization. As Karn Budhiraj remarks:
> “Perfection is a passing grade. It's like a you know, if one part doesn't show up, that line goes down, and our CEO finds out about it in less than 20 minutes.”
To manage this risk across scaling programs, Tesla streamlined its supplier network between vehicle generations. During the original Model S and Model X programs, the company deliberately cast a wide net across many vendors to evaluate their technical, financial, and cultural capabilities. That experience acted as a rigorous filter: only suppliers with both the desire and the capability to move at Tesla's pace were carried forward into the high-volume Model 3 and Model Y programs. This consolidation sharply reduced the tier-one supplier base even as total manufacturing volumes multiplied, a strategy Tesla intends to replicate on its next-generation platform.
While tier-one suppliers deliver finished assemblies directly to Tesla’s factories (totaling roughly 8,000 parts), the underlying complexity escalates dramatically down through tier-two suppliers and beyond—all the way to tier-six raw material sources like lithium and cobalt mines. The tier-two layer contains the massive universe of discrete silicon chips, resistors, capacitors, and diodes that feed every subassembly. To demonstrate the sheer operational complexity hidden beneath simple terminology, Karn introduces the "Car Computer," pointing out that while the top of the Autopilot board is already famous, the underside is packed just as densely with components that each demand meticulous supply chain tracking.
- **Accountability**: With in-house vehicle assembly, any missing part halts the line, triggering an alert to the CEO within 20 minutes.
- **Supplier Filtering**: Model S/X served as an operational testbed to prune underperforming vendors; only top-performing suppliers graduated to the Model 3/Y supply base.
- **Supply Chain Depth**: Tesla directly manages tier-one suppliers down through tier-six raw material extractors (lithium, cobalt, and nickel) to support scale.
- **The Car Computer**: Represents extreme tier-two complexity, with double-sided surface mount circuit boards densely populated with thousands of active and passive components.
[01:50:32](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6632s) Computer Architecture and Manufacturing Automation
The assembly process for the car's computer highlights the immense tier-two complexity embedded within a single tier-one part number. Mounted on the opposite side of the liquid-cooled heat sink from the Full Self-Driving (FSD) board sits the multimedia cluster (MCU) board. Like its counterpart, the MCU board is a double-sided, eight-layered printed circuit board. Because both boards run extremely hot during peak compute operations, they must be bonded directly to a central, hermetically sealed, liquid-cooled heat sink. After physical assembly, software is flashed onto the system before shipping to Tesla as a unified part number containing over 7,000 components. Production operates at extreme speed: a component is placed onto a car computer every 1.4 milliseconds along a manufacturing line stretching the length of an entire football field.
```mermaid
flowchart TD
accTitle: Car Computer Assembly Flow
accDescr: Assembly flow of Tesla car computer showing MCU board, FSD board, heat sink bonding, and automation stages.
subgraph Assembly ["Assembly & Processing"]
BOND["Bond to Heat Sink<br>(7,000+ Components)"]
FLASH["Flash Software"]
end
subgraph AutomationProgression ["Cost Reduction Stages"]
S1["1. Remove Labor"] --> S2["2. Full Automation"] --> S3["3. Lights-Out Factory"]
end
MCU --> BOND
HS --> BOND
FSD --> BOND
BOND --> FLASH
FLASH --> AutomationProgression
```
Initially, early production runs relied heavily on manual labor to manage the sheer complexity. However, once target quality, build rates, and yields stabilized, Tesla shifted entirely toward labor reduction and automation. As Karn Budhiraj put it: “The only way to control cost is by removing labor. The first step is removing labor. The second step is fully automating. The third step is turning off the lights and letting the factory run, ideally.” While automation currently stands at roughly 95%, efforts to automate further remain ongoing.
- One car computer assembly integrates more than 7,000 subcomponents into a single tier-one part.
- Components are populated at an average rate of one every 1.4 milliseconds on a line the length of an American football field.
- Tesla embeds its own supply engineers directly on supplier shop floors to ramp capacity, monitor quality, and maintain granular oversight.
[01:52:38](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6758s) Global Inbound Logistics and Nearshoring Strategy
Roshan Thomas stepped in to illustrate the massive inbound operational scale required across Tesla's tier-one and tier-two networks. In the prior year alone, Tesla moved approximately 16 million pallets of materials from global suppliers to its production factories—enough physical volume, if lined up edge-to-edge, to stretch halfway around the Earth's circumference. Furthermore, the electronic supply chain alone channels more than 1 billion individual components every week (52 billion annually) into manufacturing plants. Alongside factory logistics, the supply chain team orchestrates just-in-time component deliveries to 685 service centers worldwide to maintain the operating fleet.
To mitigate international disruptions, Tesla maintains a footprint across roughly 45 countries, aggressively dual-sourcing and triple-sourcing mission-critical subcomponents. Karn emphasized that Tesla's overarching geographic goal is to bring manufacturing closer to local factories: “Our approach really has been to bring the manufacturing of subcomponents closer to the point of consumption. It makes the supply chain more green cuz you're not burning diesel, moving stuff around. And it also gives us the security of supply of having multiple sources in case one factory is taken offline.”
[01:54:10](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6850s) Weathering COVID-19, Port Congestion, and the Chip Shortage
Roshan and Karn reflected on the past three years, with Roshan joking that the two of them “used to look a lot younger pre-COVID.” During pandemic-era shutdowns, Tesla's supply chain personnel acted far beyond traditional procurement roles, directly negotiating with municipal leaders, mayors, and state governments across disparate jurisdictions to permit safe factory restarts. Because cars cannot ship with missing components and jurisdictional lockdowns occurred asynchronously across the world, supply teams and Tesla's government relations staff tackled factory closures sequentially on an ongoing basis.
```mermaid
flowchart LR
accTitle: Supply Chain Crisis Timeline
accDescr: Sequence of global supply chain disruptions faced by Tesla from COVID to the chip shortage.
C19["COVID-19 Outbreaks<br>& Regional Lockdowns"] --> PC["West Coast Port Congestion<br>(Lead times 2x, ETA accuracy <35%)"]
PC --> CS["Global Chip Shortage<br>(Fab capacity constraints)"]
CS --> SOL["Tesla Playbook:<br>Direct Tier-2 relations, in-house design, stable demand signal"]
```
Immediately following the initial pandemic waves, severe maritime shipping congestion slammed the West Coast. Shipping logistics lead times doubled, and container vessel arrival accuracy plunged below 35%, leaving estimated times of arrival (ETAs) roughly 75% inaccurate. Shortly thereafter, the automotive chip shortage hit the industry.
While the broader automotive sector shrank overall during this period, Tesla achieved sustained vehicle production growth. Karn pointed out that Tesla's resilience stemmed from its core engineering culture: designing electronics in-house forced Tesla over a decade ago to establish direct relationships with tier-two silicon suppliers rather than hiding behind tier-one middle tiers. Consequently, when parts shortages emerged, engineering and supply leads knew exactly whom to contact to unblock allocations. Furthermore, Tesla’s aggressive demand commitment stood out: “We did not let the pandemic phase us. We were full throttle. And it was pedal to the metal in terms of our demand signal.” Because silicon fabrication plants require billions of dollars and years to construct, semiconductor foundries rewarded Tesla's unwavering, multi-year demand signals.
- Shipping accuracy plummeted during port gridlock, leaving vessel ETAs off by approximately 75%.
- Tier-three and tier-two capacity expansion timelines must precede tier-one line installations by months or years due to long semiconductor manufacturing lead times.
- Weekly production grew from zero to over 40,000 cars through the crisis period.
[01:58:16](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7096s) Scaling Silicon Needs to 20 Million Vehicles
Looking forward toward scaling output to 20 million cars per year, Karn addressed the persistent industry question of whether Tesla's ambitious targets would break global semiconductor capacity. A modern Tesla equipped with Full Self-Driving hardware integrates roughly four times the silicon surface area of a standard legacy vehicle. At a run rate of roughly 2 million vehicles annually, Tesla consumes around 700,000 twelve-inch equivalent silicon wafers per year.
Given that total global fabrication capacity stands at roughly 135 million 12-inch wafer equivalents, Tesla currently represents approximately 0.5% of worldwide wafer production. Even when scaling by a factor of ten to 20 million vehicles per year, Tesla's silicon footprint will remain manageable without monopolizing global foundries—particularly as upcoming vehicle generations incorporate architectural simplifications to streamline part counts and eliminate redundant discrete controllers across the vehicle.
| Metric | Current Run Rate (~2M Vehicles) | Global Semiconductor Industry |
|:--- |:--- |:--- |
| **Tesla 12-inch Wafer Consumption** | ~700,000 wafer equivalents / year | ~135,000,000 wafer equivalents / year |
| **Tesla Share of Total Capacity** | ~0.5% | 100% |
| **Silicon Multiplier vs. Standard Vehicle** | ~4× silicon content (with FSD Computer) | Baseline 1× |
[01:59:19](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7159s) Applying the Electronic Playbook to Mechanical Systems
Even under the most pessimistic assumptions, Tesla's projected growth will require approximately 8 million semiconductor wafers. However, because the broader industry is scaling to between 20 million and 200 million wafers by that point, Tesla's demand will still represent less than 5% of the total market share. By continuing to run their established playbook—collaborating closely with partners, investing directly in fabrication facilities, and securing dedicated capacity—the supply chain team is confident that semiconductor supply will not act as an impediment to reaching their target of 20 million vehicles.
The supply chain team has also taken a page out of the electronics playbook and applied it directly to mechanical engineering. Previously, Tesla vehicles relied on a dispersed thermal system connected by a complex web of hoses. Drawing inspiration from integrated circuit boards, Tesla's engineers designed a highly integrated heat pump system known as the "super manifold." This design packaged 60% of the thermal system into a component roughly the size of a basketball, consolidating about 100 individual components, 50 sealing interfaces, and several distinct manufacturing processes into a single, tight assembly.
This highly integrated system was launched in March 2020, right at the outbreak of the COVID-19 pandemic. The engineering and supply chain teams were forced to manage over 100 components from different suppliers around the globe completely remotely. Initially, Tesla set up a manual assembly line, which suffered from severe throughput and quality issues. They quickly pivoted to a semi-automated line, which resolved the throughput bottlenecks but failed to eliminate persistent quality defects. Realizing that full automation was the only viable path forward, the team built a complete 3D simulation of the complex sub-assemblies. This allowed them to specify exactly what equipment their suppliers needed to purchase and test, optimizing the entire end-to-end line virtually before physical implementation.
The fully automated line was deployed in less than eight months, yielding a 99% reduction in the required labor force. A production line that would have required 1,000 associates to build the super manifold at full volume now requires only 10. Furthermore, quality improved by an order of magnitude, driving the defect rate down to less than 0.005%, with a new super manifold rolling off the line every 7 seconds.
“This is how we are thinking about complex assemblies and we are going to push this type of thinking into our supply base more and more who will be building our complex assemblies,” the speaker explained, noting that automation is the only way to confidently scale to 20 million vehicles in a world of rising labor costs and high turnover.
- **Super Manifold Integration:** Consolidates 60% of the thermal system, containing roughly 100 components and 50 sealing interfaces, into a basketball-sized unit.
- **Labor Reduction:** Automation reduced the required assembly staff from 1,000 associates to just 10.
- **Quality and Speed:** The automated line achieved a defect rate of less than 0.005% and a cycle time of 7 seconds per unit.
- **Automation Cost Trend:** The cost of implementing advanced automation is projected to decrease by 3x over the next 10 years.
[02:03:57](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7437s) Supply Chain Strategy and the Road to 20 Million
Tesla's supply chain strategy for executing Master Plan 3 rests on four core pillars: simplifying supply chain design, establishing deeper control over the lower tiers of the supply chain, growing responsibly and sustainably alongside long-term partners, and aggressively automating complex assemblies. The team acknowledges that scaling from zero to 40,000 cars per week was an incredibly painful process filled with trial, error, and what Elon Musk refers to as "mental scar tissue." However, because that foundational work is now complete, the prospect of scaling tenfold from 40,000 to 400,000 cars per week is no longer intimidating.
“Our leadership team is galvanized behind the mission, and we're going to get it done,” the speaker declared, thanking the audience and handing the stage over to Tom Zhu and Drew Baglino to discuss global manufacturing.
```mermaid
flowchart TD
accTitle: Tesla Supply Chain Strategy
accDescr: Flowchart illustrating the core pillars of Tesla's supply chain strategy for Master Plan 3.
A["Supply Chain Strategy"] --> B["Design Simplification"]
A --> C["Deeper Tier Control"]
A --> D["Responsible & Sustainable Growth"]
A --> E["Aggressive Automation"]
D --> F["Long-Term Partners"]
E --> G["3D Simulation & Automated Lines"]
```
[02:05:06](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7506s) Global Manufacturing and the 4 Million Milestone
Tom Zhu, who joined Tesla in 2014 and previously ran the company's operations in Greater China and the Asia-Pacific region, introduced himself as the leader now responsible for global production, sales, delivery, and service. Representing all of Tesla's Gigafactories, Zhu detailed the company's current manufacturing footprint: four vehicle factories operating across three continents, supported by a manufacturing workforce of over 65,000 employees. This total includes the team at Gigafactory Nevada in Reno, who produce the powertrains, battery packs, and drive units that feed the vehicle assembly plants. Currently, Tesla's total installed capacity across these factories is approximately 2 million cars per year, a number the company continuously aims to increase through footprint optimization.
Zhu highlighted Tesla's speed in factory construction, pointing to Gigafactory Shanghai as a prime example. Built on a greenfield site in just 9.5 months, the factory began delivering its first customer vehicles only three months later. Zhu explained that this rapid pace was made possible by learning from the "production hell" experienced at the Fremont factory. By designing straight production lines with minimal elevation changes and turns, deleting redundant buffers, and utilizing a highly capable in-house construction team, Tesla has streamlined its factory builds. This same in-house team has managed construction across Giga Berlin, Giga Texas, and Giga Nevada. Looking ahead, the modular design of Tesla's next-generation vehicle platform will allow for even more densified Gigafactories, enabling the company to build more cars from a smaller physical footprint.
Zhu also shared a major milestone: early on the morning of March 1, 2023, Tesla officially built its 4 millionth vehicle. The milestone car was manufactured right on the floor of Gigafactory Texas, where the Investor Day event was being held. Zhu emphasized the exponential acceleration of Tesla's production velocity, noting how the time required to build each subsequent million vehicles has dramatically shrunk.
“Took us 12 years to build the first million, and about 18 months to build the second million. The third million took us 11 months and just a shortly less than a 7 months we built the fourth million cars,” Tom Zhu said, praising the global manufacturing team for their efforts.
- **Global Footprint:** Four vehicle factories on three continents with an installed capacity of approximately 2 million vehicles per year.
- **Giga Shanghai Construction:** Completed in 9.5 months from greenfield to factory completion, with customer deliveries starting 3 months later.
- **Production Acceleration:**
- **1st Million:** 12 years
- **2nd Million:** 18 months
- **3rd Million:** 11 months
- **4th Million:** Less than 7 months
[02:08:48](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7728s) The Mechanics of Ramping a Gigafactory
Ramping a Gigafactory to full production is an incredibly complex task that requires coordinating 600 robots, thousands of processes, and a massive workforce—which Tom Zhu joked could eventually consist of "5,000 human and 5,000 Optimus" bots. To manage this complexity, Tesla focuses on two primary operational metrics: Overall Equipment Effectiveness (OEE) and cycle time.
Tesla sets a strict passing grade for its vehicle factories: an OEE of 90% and a cycle time of 45 seconds. OEE measures a production line's performance by evaluating equipment uptime, machine speed, and product quality, comparing actual high-quality output against planned productive time. A 45-second cycle time means that a finished vehicle must roll off the final assembly line every 45 seconds. Minimizing cycle times and maximizing OEE allows Tesla to ramp its factories rapidly, which is critical for capturing the economies of scale necessary to lower production costs.
“And the faster we ramp, the faster we can get the economy of scale,” Tom Zhu explained, preparing to show how Giga Shanghai successfully optimized its labor and manufacturing processes to achieve these targets.
- **Target Factory Metrics:** 90% Overall Equipment Effectiveness (OEE) and a 45-second cycle time.
- **Cycle Time Output:** A target of one finished vehicle rolling off the assembly line every 45 seconds.
- **Ramp Variables:** Successful factory ramps require the synchronization of over 600 robots and hundreds of distinct manufacturing processes.
[02:10:01](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7801s) Manufacturing Efficiency, First Principles, and Customer Feedback Loops
Tom Zhu continues reviewing production line improvements across Tesla's global footprint, pointing to hours per car during ramps and noting a brief dip in Q2 2022 caused by the COVID-19 shutdown. Turning to the Fremont Model Y shop, Zhu emphasizes that despite being a six-year-old facility, the on-site team has systematically optimized material flow, removed single points of failure, and driven higher throughput, which directly lowers labor hours per unit. In fact, Zhu reveals that Fremont continues to set records: “Yesterday they just had a new factory daily record,” prompting applause from the room as he congratulates the team.
Explaining how Tesla persistently drives down cycle times, Zhu invokes Elon Musk's manufacturing ethos: “We follow the philosophy Elon shared about building rocket. Which is questioning. Basically find the right problem to solve. And we start with delete, simplify, then we try to accelerate, pressurize the line, and find the if the the solution actually work. And at last we think about the automation.” He shares concrete examples of this doctrine in practice:
- **Shanghai Paint Shop:** Engineers realized there was an overlapping baking range between the PVC sealer and the top coat, which originally ran through two separate ovens. By merging these into a single combined process, Shanghai reduced cycle time, eliminated an oven step, and cut both energy consumption and CO₂ emissions by 9%.
- **Fremont Model S and Model X commonization:** Teams created a common body structure and headlamps across the Model S and X lines, cutting roughly 40 distinct parts and trimming cycle time by about 10%.
- **Berlin 5G Private Network:** Gigafactory Berlin rolled out a shop-floor private 5G network that resolved approximately 90% of over-cycle issues on a specific general assembly (GA) station, an upgrade Tesla plans to deploy globally.
```mermaid
flowchart LR
accTitle: First Principles Optimization Flow
accDescr: Tesla's five-step engineering loop applied to manufacturing processes.
A["Question & Define"] --> B["Delete"]
B --> C["Simplify"]
C --> D["Accelerate & Pressurize"]
D --> E["Automate"]
```
Zhu details how Tesla has unified global operations under a single umbrella—linking production directly to sales, delivery, and service. This structural unification creates a rapid feedback loop where real-world customer escalations directly drive shop-floor process improvements: “With this direct feedback loop we be able to turn customer escalations and the feedback into a quick actions and improvement on the shop floor immediately.” In just the prior six months, this mechanism led to substantial reductions in service visit duration, early-ownership complaints, and appointment wait times. Looking ahead, Zhu states that four factories are insufficient to hit the company's long-term goal of 20 million cars per year, requiring continuous construction of new plants and lines running unified, shared manufacturing practices.
---
[02:13:51](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8031s) 4680 Cell Scaling and Factory Footprint Reductions
Drew Baglino takes the stage to address cell manufacturing, connecting the vehicle expansion goals to an equivalent push in battery manufacturing. Reminding the audience of the Battery Day presentation where Tesla showed dry electrode powder turning into film using a literal spoon, Baglino jokes: “Let's just say there's no spoon now.” Attendees on the day's factory tour saw the fully automated dry electrode machine in Giga Texas running powder and foil in, and coated electrode out, with zero manual intervention. Per tool, the Texas equipment achieves more than 20 times the peak productivity of the Kato Road pilot tools shown during Battery Day.
Baglino explains that transitioning from conventional 2170 cylindrical cell layouts to the 4680 architecture yields a massive 5x reduction in factory footprint and volume. Further engineering refinements from the Kato pilot line to Giga Texas, and onward into Giga Nevada, have shrunk the factory volume by a total factor of 10x relative to traditional facilities. He attributes this leap to close collaboration among five internal teams: product design, manufacturing design, process design, equipment design, and facility design.
```mermaid
flowchart TD
accTitle: Five-Way Collaborative Design Loop
accDescr: The five core disciplines Tesla integrates to shrink factory footprint and scale cell manufacturing.
subgraph IntegratedDesign["Integrated Holistic Design"]
PD["Product Design"]
MD["Manufacturing Design"]
PRD["Process Design"]
ED["Equipment Design"]
FD["Facility Design"]
end
IntegratedDesign --> OUT["10x Volumetric Footprint Reduction & Lower CapEx/GWh"]
```
- **Parts and process reduction:** Drastic pruning of internal cell components and manufacturing steps lowered CapEx per gigawatt-hour.
- **Scale objectives:** The streamlined cell footprint is designed to support Tesla's macro target of 240 terawatt-hours of total transition storage, 1 terawatt-hour per year of stationary storage, and 20 million vehicles annually.
---
[02:16:19](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8179s) Upstream Materials: Lithium Refining and Cathode Production
Beyond cell assembly, Baglino reviews Tesla's selective vertical integration into raw materials. He showcases the 50 GWh/year Corpus Christi lithium refinery in Texas, where ground has already been broken: “The facility will start commissioning by the end of 2023 this year. This is a good example of of something where we're we're we're basically talking about breaking ground and and and starting commissioning within 10 months and with with with actual production within 12 months.” Situated on a direct rail connection 30 minutes from the port of Corpus Christi, the refinery uses a direct soda ash leach process, entirely bypassing the acid roasting stage and eliminating the hazardous waste streams typical of legacy refining. It is optimized for spodumene rock but engineered to accept diverse primary and secondary feedstock feeds.
Baglino also highlights the 60 GWh/year cathode facility being built directly behind the main Giga Texas building. Constructed on an identical 10-month timeline, tooling installation was underway for line one, with commissioning slated to start the following quarter. Baglino clarifies that Tesla does not intend to internalize every supply-chain link, acknowledging that “there are lots of competent companies out there,” but steps in where necessary to de-risk key manufacturing innovations, lower CapEx per gigawatt-hour, and accelerate industry-wide scaling speed.
---
[02:18:14](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8294s) Energy and the Megapack Business
Baglino introduces Mike Snyder, head of the Megapack organization, to present on Tesla's stationary energy division. Snyder, who has been with Tesla for nearly nine years, explains that the company's approach from day one was never about merely selling loose battery cells: “From the beginning with this business, we have always focused on building successful projects for our customers and not just the batteries in the box.” Tesla developed an adaptable hardware and software platform engineered to operate reliably in diverse climates across the globe, capable of scaling seamlessly from microgrid island installations to multi-gigawatt-hour utility systems.
Watching a montage of completed installations spanning a decade of deployment, Snyder describes the systems as intuitive, turnkey, and engineered around deep knowledge of site risks and utility integration. He announces that Tesla is now manufacturing its sixth-generation industrial energy product—the Megapack XL—at the dedicated Megafactory in Lathrop, California.
[02:19:52](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8392s) Megapack Engineering, Density, and Factory Scaling
Tesla has deployed over 16 gigawatt-hours of industrial and residential energy products across 50 countries, positioning the Megapack as the clear market leader in efficiency, reliability, and energy density, while remaining the easiest and cheapest system to install. Addressing the physical dimensions of the unit, Mike Snyder joked about why it is designated as "XL": “And we call it XL because it is actually the like largest, heaviest object that you can transfer around the roads of the world without, you know, having to shut them down and get crazy permits and things like this. So, it is extra large and there is going to not going to be an XXL, let's just say it that way.” Customer demand has surged heading into 2023, driven by gigawatt-hour scale project construction, ongoing ramp-up at the dedicated Lathrop Megafactory, and new products slated for the near-term roadmap.
Reaching this stage required a relentless engineering focus on every element of stationary storage delivery. For example, Tesla integrated the electrical wireways directly into the base of the Megapack enclosure, eliminating the traditional, time-consuming construction requirement of trenching conduits and underground cables. Crews can instead lift the cables straight into the base, accelerating deployment and lowering installation expense. Internally, the battery modules and integrated power electronics convert DC battery power into AC grid power using an onboard "brain" capable of interconnecting with any utility grid worldwide straight out of the crate, enabling hundreds of standalone Megapacks to operate seamlessly as a synchronized power plant. This high level of integration delivers upwards of 300 megawatt-hours per acre—twice the power density of a conventional fossil gas peaker plant. As Drew Baglino affirmed: “This is the product that retires the fossil fuels. One power plant at a time.”
- **Global deployment**: Over 16 GWh of industrial and residential storage installed across 50 countries.
- **Land use efficiency**: More than 300 MWh per acre, achieving double the power density of standard gas peaker facilities.
- **Integrated wireways**: Enclosure base wireways eliminate extensive below-ground civil trenching and conduit runs.
```mermaid
flowchart TD
accTitle: Megapack Integration and Grid Connection Flow
accDescr: Diagram showing how Megapack integrates enclosure, batteries, power electronics, and grid connection.
subgraph MegapackUnit ["Megapack XL Enclosure"]
A["Integrated Base Wireways"] --> B["DC Battery Modules"]
B --> C["Power Electronics & Onboard Brain"]
end
subgraph GridOperation ["Utility Grid Interface"]
C -->|"Direct AC Inversion"| D["Universal Grid Interconnect"]
D -->|"Coordinated Control"| E["Multi-Pack Site Fleet (300+ MWh/acre)"]
end
```
[02:22:03](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8523s) Power Electronics and Virtual Machine Mode
Tesla has emerged as a powerhouse in power electronics, having deployed more than 1.4 terawatts of capacity across its vehicle and energy storage fleets combined. On an annualized basis, Tesla manufactures and ships more power electronics than the global solar and wind industries combined. Drew Baglino highlighted how vital these components are to the broader transition, noting that power electronics serve as the fundamental connective tissue between energy generation, stationary storage, and end consumption. These semiconductor devices switch hundreds of thousands of times per second to respond dynamically to transient conditions on the grid or in a vehicle drivetrain.
Because this hardware is entirely software-controlled, it delivers advanced capabilities previously unattainable on alternating current power grids. A prominent breakthrough is Tesla's "Virtual Machine Mode" firmware, which provides synthetic inertia. Mike Snyder explained that Virtual Machine Mode stabilizes the grid much like a vehicle shock absorber smooths out mechanical oscillations and vibrations; without that damping, mechanical oscillations cause vehicles to spin out, while electrical frequency swings trigger grid-wide blackouts. Through pure software emulation, Megapack batteries replicate the physics of massive rotating mechanical turbines without needing a traditional fossil fuel or hydroelectric generating station. Drew Baglino emphasized the transformative nature of the feature: “We have one grid operator that's utilizing virtual machine mode and said they will not operate their grid at 100% renewables unless they have this feature unless they have virtual machine mode working.”
- **Manufacturing scale**: More than 1.4 TW of power electronics deployed across vehicles and stationary storage.
- **Industry throughput**: Tesla ships more power electronics volume annually than the global wind and solar sectors combined.
- **Dynamic inertia**: Virtual Machine Mode replaces spinning synchronous masses with programmable synthetic inertia that dynamically tunes dampening parameters to local grid conditions.
[02:24:22](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8662s) Autobidder Autonomous Energy Trading
Beyond hardware and firmware stabilization, managing battery storage economics requires complex software automation. Tesla created Autobidder, an autonomous energy monetization platform. While the baseline financial strategy involves charging during periods of low power prices and discharging when electricity commands a premium, the operational reality of utility-scale storage demands hundreds or thousands of dispatch decisions every five minutes. This degree of real-time multi-variable optimization far exceeds the operational complexity of solar farms, wind parks, or thermal generation plants.
Because no commercial third-party platform could handle this decision density and real-time execution, Tesla engineered Autobidder in-house. The platform evaluates market pricing, weather forecasts, battery degradation costs, and ancillary service opportunities simultaneously. Mike Snyder stated: “In the market that we've been markets we've deployed them in in Australia and Texas and the UK it has proven to be a market leader. We're continuing to invest in Autobidder as we expand to new markets.”
- **Decision frequency**: Evaluates and executes hundreds to thousands of automated bidding and dispatch decisions every 5-minute interval.
- **Core operational footprint**: Active across competitive wholesale electricity markets including ERCOT in Texas, the National Electricity Market (NEM) in Australia, and the United Kingdom.
- **Algorithmic arbitrage**: Co-optimizes wholesale energy arbitrage alongside frequency regulation and reserve grid services.
[02:25:19](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8719s) Rapid Manufacturing and Project Execution
A foundational driver of Tesla Energy’s growth is rapid execution speed, demonstrated in both factory buildout and field commissioning. At the Lathrop Megafactory in California, Tesla converted a former J.C. Penney distribution warehouse into a specialized, high-volume Megapack manufacturing facility in less than 12 months. This accelerated build was achieved by applying Tesla's vertically integrated playbook: seating automotive production engineers directly alongside industrial facility designers and construction managers so engineering roadblocks could be solved cross-functionally on the spot without multi-tier contractor delays.
That same speed philosophy governs project site construction. By standardizing connections and engineering true plug-and-play architecture, Tesla transformed field commissioning. Mike Snyder detailed the substantial efficiency gains: “Over the past 4 years, we've we've increased the installation speed by 4x. And we've reduced the total labor involved across both construction and manufacturing by 3x.” Minimizing the duration between when a Megapack rolls out of the factory gate and when it synchronizes with the commercial grid is considered paramount for unlocking customer scale.
- **Lathrop conversion**: Industrial site transformed from a retail warehouse to an active Megafactory in under 12 months.
- **Installation speedup**: 4x acceleration in field installation timelines achieved over a four-year horizon.
- **Labor footprint**: 3x reduction in combined field construction and factory labor per megawatt-hour deployed.
[02:26:40](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8800s) Distributed Storage and Tesla Electric
Tesla’s energy strategy balances centralized utility installations with distributed residential systems. Through Tesla Electric—launched commercially in retail deregulated markets such as Texas—the company leverages software derived from Autobidder to pool residential Powerwalls into Virtual Power Plants (VPPs), effectively allowing everyday households to operate collectively as their own decentralized utility provider.
To demonstrate the real-world financial return of this distributed architecture, Drew Baglino shared empirical data collected throughout 2022 across 5,000 enrolled Powerwall and solar customers in South Australia. A standard residential customer on a default retail utility tariff paid an average electricity bill of $140 per month. Adding rooftop solar and a local Powerwall in non-export self-consumption mode cut that monthly cost in half to $70. However, when those same systems were connected to Tesla Electric software and dispatched intelligently to support regional grid stability, the economics inverted completely, turning electricity bills into net payouts. Drew Baglino emphasized the precedent South Australia sets for modern grids: “Solar and wind supplied 70% of South Australia's energy in 2022. That compares to 30% in Texas and 35% in California. But it But this is a This is an indication of where this is all headed.”
| Customer Configuration (South Australia VPP Data) | Net Monthly Customer Electricity Cost |
|:--- |:--- |
| Default Utility Retail Service | **$140 / month** |
| Solar + Powerwall (Isolated / No Grid Interaction) | **$70 / month** |
| Solar + Powerwall Managed via **Tesla Electric** | **Net Positive Cash Flow (Customers Get Paid)** |
- **Sample size**: 5,000 participating residential customers analyzed throughout calendar year 2022 in South Australia.
- **Renewable energy share (2022 benchmark)**: South Australia achieved 70% renewable generation from solar and wind, compared to 35% in California and 30% in Texas.
[02:28:23](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8903s) Tesla Electric, Retail Energy Expansion, and the 1 TWh Storage Ambition
Drew Baglino and Mike Snyder outline the role of retail energy in unlocking fully renewable power grids by orchestrating both centralized and distributed storage resources. The customer-facing vehicle for this effort is Tesla Electric, which initially rolled out to residential customers in Texas who own a Powerwall. The market opportunity extends far beyond this pilot: over one billion people live in competitive retail electricity markets representing more than $2 trillion in annual energy spending. Tesla intends to expand market by market, mimicking the rollout playbook of Tesla Insurance to lower the total cost of ownership (TCO) for customers.
In Texas, Tesla Electric is introducing a new plan by the summer of 2023: unlimited overnight EV home charging for $30 per month for Tesla owners. This offering capitalizes on Texas's massive wind generation, which peaks during nighttime hours. Incentivizing EV owners to consume surplus overnight wind power aligns grid demand with renewable generation, directly driving down ownership costs while supporting Master Plan 3.
```mermaid
flowchart LR
accTitle: Tesla Electric Retail Energy Flow
accDescr: Nighttime wind energy generation powering Tesla vehicles under a flat monthly retail plan.
subgraph Grid["Renewable Generation"]
W["Texas Nighttime Wind"]
end
subgraph Product["Tesla Electric"]
TE["Retail Plan<br/>($30/month)"]
end
subgraph Consumption["End Users"]
EV["Overnight Home EV Charging"]
PW["Powerwall Distributed Storage"]
end
W --> TE
TE --> EV
TE --> PW
```
To meet the global requirement of tens of terawatt-hours of stationary storage modeled in Master Plan 3, Tesla is establishing a long-term production target of 1 terawatt-hour (TWh) per year—a 25x increase over the 40 GWh capacity at the Lathrop Megafactory.
> “So, we have our sights set on annual production rates of 1 terawatt hour, which is 25x our capacity at Lathrop.” — *Drew Baglino*
- **Addressable retail energy market:** Over 1 billion people and >$2 trillion in annual spend across competitive retail markets worldwide.
- **Texas EV overnight charging plan:** $30 per month for unlimited charging, powered by off-peak overnight wind generation.
- **Near-term Megapack demand:** Exceeding 100 GWh per year in 2023, with projected annual growth of over 100 GWh per year over the following few years.
- **Long-term storage production target:** 1 TWh per year across Megapack and stationary storage platforms.
---
[02:31:07](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9067s) Impact at Tesla: Workforce, Culture, and Employee Engagement
Brandon Ehrheart of the Tesla legal team opens the "Impact at Tesla" section by welcoming investors and affirming active engagement between management and the Board of Directors, before introducing Laurie Shelby, Vice President of Environmental Health, Safety, Security, and Sustainability (EHS3). Shelby highlights that Tesla's success rests fundamentally on its workforce of 129,000 employees globally, with more than 50% working in vehicle manufacturing and nearly 60% based in the United States.
Tesla's mission makes it an extraordinarily competitive destination for technical and manufacturing talent. In 2021 alone, the company received 3 million job applicants, holding the top spots alongside SpaceX as the top engineering employers of choice. Shelby describes the internal "Take Charge" program, an engagement platform empowering any worker on the factory floor or in service to flag issues or suggest improvements across safety, people, cost, production, and quality. Empirical factory data demonstrates that as employee-submitted "Take Charge" reports increase, incident rates consistently decline.
> “People who do the work, the people who do the work know how to improve the work.” — *Laurie Shelby*
- **Global headcount:** 129,000 employees.
- **Workforce distribution:** >50% in vehicle manufacturing; ~60% located in the United States.
- **Job applications:** 3 million applicants in 2021; Tesla and SpaceX ranked #1 and #2 preferred engineering employers.
- **Take Charge program:** Direct relationship established between higher bottom-up employee reporting and lower workplace injury rates.
---
[02:34:42](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9282s) Environmental Impact, Operational Footprint, and Reporting Alignment
Addressing sustainability metrics, Shelby stresses that Tesla rejects superficial compliance checklists in favor of measuring real physical impact through vehicles, energy hardware, and software. In 2021, Tesla's customer fleet enabled an avoidance of 8.4 million metric tons of greenhouse gas emissions—the equivalent of removing 1.7 million internal combustion engine (ICE) passenger cars from roads for a year.
Inside Tesla’s own operations, facilities continue to drive year-over-year reductions in water consumption and waste generation while scaling on-site renewable generation across manufacturing plants and the Supercharger network. Looking across the entire lifecycle from 2012 to 2021, the cumulative clean electricity generated by Tesla solar installations exceeded the entire volume of energy consumed by all Tesla vehicles on the road combined with all Tesla factory operations worldwide.
Addressing grid cleanliness criticisms, Shelby points out that even in a worst-case scenario where an EV charges entirely from a 100% coal-fired electric grid, a Tesla Model 3 still produces fewer lifecycle emissions than the average gasoline-powered vehicle. Concluding her segment, Shelby announces that Tesla is formalizing its stakeholder reporting by aligning with the Task Force on Climate-related Financial Disclosures (TCFD) framework, with expanded disclosures planned for the upcoming Impact Report.
- **Customer emissions avoided (2021):** 8.4 million metric tons of GHG emissions (~1.7 million ICE vehicles removed for a full year).
- **Clean generation vs. consumption (2012–2021):** Solar generation produced more cumulative energy than all fleet driving and manufacturing operations consumed.
- **Coal-grid parity:** A Model 3 operated on a 100% coal electric grid still produces lower net emissions than an equivalent average ICE car.
- **Governance & reporting:** Formal adoption and alignment with the TCFD (Task Force on Climate-related Financial Disclosures) framework.
---
[02:37:44](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9464s) Financials: Relentless Cost Reduction on the Model 3 Platform
Chief Financial Officer Zach Kirkhorn takes the stage for the final prepared presentation, reflecting on this being the first formal Investor Day Tesla has held featuring presentations across the entire engineering and executive leadership bench. Kirkhorn notes that while presenters were given free rein to describe their domains in the context of Master Plan 3, every single engineering leader independently focused on structural cost reduction.
> “Because in this industry, in this business, you survive or you die based upon the ability to manage your costs.” — *Zach Kirkhorn*
Kirkhorn focuses on Tesla’s longest-running high-volume production vehicle, the Model 3. After achieving its original nameplate design capacity of 5,000 vehicles per week in mid-2018, Tesla managed to engineer 30% of total unit cost out of the Model 3 platform over the subsequent years through continuous powertrain, electronics, manufacturing, and supply chain iterations.
- **Model 3 ramp milestone:** Reached 5,000 vehicles per week production rate in mid-2018.
- **Cost performance:** Cumulative 30% unit cost reduction achieved on the Model 3 from mid-2018 through 2023.
[02:39:40](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9580s) Disciplined Cost Reduction and Next-Gen Platform Targets
Zach Kirkhorn explains that a deep focus on cost reduction is fundamentally ingrained in Tesla's corporate culture and serves as a primary pillar of the company’s survival and success. Crucially, cutting costs cannot come at the expense of vehicle quality. As Kirkhorn puts it: “when we're working on cost reduction, it's easy just to take costs out and make our products worse. But we have to take costs out and improve our products at the same time.” Comparing a Model 3 manufactured in 2023 to one produced in 2018 reveals an extensive list of product enhancements achieved concurrently with major cost reductions.
There is no single silver bullet for driving out expense. Instead, gains result from scaling production volume 3x, doubling manufacturing productivity at the Fremont factory compared to mid-2018, refining overhead efficiencies, and heavily localizing supply chains, such as at Gigafactory Shanghai. When scaling volumes, Tesla works closely with external suppliers so they can replicate these manufacturing efficiencies, passing along material cost savings that directly enhance vehicle affordability.
Looking forward to Tesla’s next-generation vehicle platform, the company has targeted a 50% cost reduction relative to the Model 3 and Model Y platform—repeating the exact same 50% reduction accomplished when stepping down from the Model S and Model X platform. Kirkhorn emphasizes that linear reductions in vehicle cost unlock exponential demand growth, vastly expanding the population of potential buyers and aligning directly with Master Plan 3. The anticipated 50% cost reduction is balanced across equal buckets across the car: vehicle engineering, battery and powertrain, manufacturing processes, and related overhead.
```mermaid
flowchart LR
accTitle: Platform Cost Halving Progression
accDescr: Step-down progression illustrating a 50 percent cost reduction from Model S/X to Model 3/Y, and targeting another 50 percent reduction for the next-gen platform.
subgraph Gen1["First Generation"]
SX["Model S & Model X Platform"]
end
subgraph Gen2["Second Generation"]
ThreeY["Model 3 & Model Y Platform"]
end
subgraph Gen3["Next-Generation Platform"]
NextGen["Next-Gen Vehicle Platform"]
end
- Model 3 production volume increased 3x between 2018 and 2023.
- Labor productivity at the Fremont factory doubled relative to mid-2018 levels.
- The 50% cost-down target on the next-generation vehicle is split across balanced categories: vehicle architecture, battery and powertrain, and manufacturing.
---
[02:42:01](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9721s) Lifetime Cost of Ownership and Operating Expense Leverage
Beyond the initial sticker price, Kirkhorn underscores that the transition to a sustainable economy depends heavily on total cost of ownership (TCO) per mile over a vehicle's life cycle. Factoring in five-year ownership costs—including financing, insurance premiums, power/charging expenses, maintenance, and regular wear and tear—a base Model 3 in the United States already achieves a lower cost per mile than a Toyota Corolla, the best-selling passenger car in the world. As Tesla rolls out its next-generation platform and introduces dedicated robotaxi variants, total lifetime cost per mile will drop even further below high-volume internal combustion competitors.
Turning to corporate operational efficiency, Kirkhorn addresses operating expenses (OpEx), noting that while quarterly earnings calls typically obsess over vehicle gross margins, Tesla’s OpEx trajectory represents an underappreciated competitive moat. Since 2018, Tesla has reduced non-GAAP OpEx by 60% as a percentage of total revenue and by 75% on a per-car-delivered basis.
Tesla accomplishes this in Research & Development (R&D) by deliberately constraining the number of active programs running in parallel, maximizing the revenue and cash generated per project, and heavily sharing underlying technology across divisions—such as utilizing shared power electronics between the energy storage and automotive business units. On the Selling, General, and Administrative (SG&A) front, Tesla estimates its SG&A per car delivered is 60% to 70% lower than traditional legacy automakers when combining OEM corporate overhead with independent franchise dealership markups. As Kirkhorn notes regarding margin philosophy: “This is part of the reason why on the last earnings call, I made the comment that we think about operating margin more so than gross margin as a company because the the integration that we have here on the SG&A side, particularly with vertical integration into our dealership network, this provides efficiencies that give us a competitive advantage from a cost structure standpoint relative to other companies.”
- Five-year total cost of ownership per mile for a base Model 3 is lower than that of a Toyota Corolla in the US market.
- Non-GAAP OpEx down 60% as a percentage of revenue since 2018.
- Non-GAAP OpEx down 75% on a per-car-delivered basis since 2018.
- SG&A per car delivered is 60% to 70% below traditional automotive OEM and dealer network baselines, equating to an advantage of thousands of dollars per car.
---
[02:44:55](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9895s) Surviving "Back Office Hell" with the Tesla Operating System
Kirkhorn reflects that Tesla’s structural cost advantage was not an overnight achievement; years prior, the SG&A cost comparison was completely inverted. When Tesla launched the Model 3 ramp at Fremont, the factory jumped from 2,000 Model S and X vehicles per week to 7,000 total vehicles per week (around 350,000 cars per year). While outside observers frequently heard about “production hell,” Tesla simultaneously confronted an internal crisis that Kirkhorn terms “back office operations hell.” The company realized that if it was struggling to support a run-rate of 7,000 cars a week, scaling to its long-term ambition of 20 million cars per year with conventional tools would be impossible without fundamentally overhauling its administrative infrastructure.
To solve this, Tesla developed what it calls the "Tesla Operating System." Paralleling its vertical integration into vehicle software, Tesla brought corporate software development in-house. Rather than relying on a tangled web of generic, expensive third-party enterprise tools that require extensive custom integration, an internal applications engineering team builds lightweight, dedicated solutions tailored specifically to Tesla's operational workflows. Most recently, Tesla replaced its third-party recruiting software with a custom in-house platform.
This in-house engineering team partners directly with embedded business process improvement teams, applying the exact principles Tom Zhu outlined for vehicle manufacturing: delete unnecessary process steps, simplify existing procedures, and automate what remains. The financial and productivity results have exceeded expectations:
```mermaid
flowchart TD
accTitle: Tesla Business Process Optimization Loop
accDescr: Three-step continuous improvement workflow applied to internal company operations.
A["Audit Workflow / Business Need"] --> B["1. Delete Unnecessary Steps"]
B --> C["2. Simplify Existing Steps"]
C --> D["3. Automate Remaining Work"]
D -->|"Deploy to Tesla Operating System"| E["Custom Lightweight Software"]
E -->|"Monitor Real-Time Granular Telemetry"| A
```
Operationally, the North American sales team is now four times more productive than in 2018, with comparable efficiency leaps across order operations and financial services. Paralleling the vehicle philosophy of adding features while lowering cost, the Tesla Operating System has expanded customer self-service capabilities and underpinned the vertical integration of Tesla Insurance—which operates its own proprietary claims software, adjusters, and customer service staff. Because Tesla owns the full end-to-end software stack, the business extracts highly granular, real-time telemetry across all operational functions, allowing leadership to spot bottlenecks immediately and adapt workflows dynamically.
[02:48:38](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=10118s) Internal Feedback Loops, Reporting Speed, and Capex Requirements
Zach Kirkhorn explains how tightly coupled Tesla's internal systems are, pointing out that when teams need specific data or workflow updates, “it's a Teams meeting away from the applications engineering team to make a request, and then we can start tracking that data and make those changes.” This immediate internal software integration makes the feedback loop for business process improvements remarkably fast. From a corporate governance and accounting perspective, Kirkhorn highlights that the finance department now files its Form 10-Q and Form 10-K filings among the fastest of the top 20 largest market-cap public companies.
Turning to long-term capital requirements, Kirkhorn contextualizes the capital expenditures needed to fulfill Master Plan 3. Reaching an annual output of 20 million electric vehicles and 1 terawatt-hour of energy storage—along with proportionate expansions in battery cell manufacturing, service locations, and Supercharging infrastructure—requires massive outlays. Historically, Tesla has invested approximately $28 billion to reach its current scale. While the future investment figure sounds daunting in absolute terms, Kirkhorn argues it is modest compared to the company’s operating cash flow trajectory and operational targets.
- Tesla's target vehicle output: 20 million cars annually.
- Tesla's target stationary storage output: 1 terawatt-hour (TWh) per year.
- Cumulative historical capital spent: ~$28 billion.
- Kirkhorn asserts: “I actually think it's quite small relative to our ambitions.”
[02:50:21](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=10221s) Capital Allocation Strategy: Operations, Growth, and Shareholder Returns
Kirkhorn lays out Tesla’s explicit hierarchy for capital allocation. The primary requirement is securing daily operating cash and working capital, which fluctuates significantly. Depending on production timing and delivery schedules toward the end of quarters, working capital can swing by several billion dollars. The second tier is maintaining a fortress balance sheet for downside protection against macroeconomic downturns and unexpected black swan events, as demonstrated during the COVID-19 pandemic.
```mermaid
flowchart TD
accTitle: Tesla Capital Allocation Hierarchy
accDescr: Four-tier prioritization flow of Tesla cash deployment from daily operations to capital returns.
subgraph CoreOps["1. Operational Stability"]
A["Day-to-Day Working Capital\nBuffer for multi-billion dollar swings"]
B["Downside Protection Reserve\nBuffer against macro shocks & disruptions"]
end
subgraph Shareholder["3. Capital Return"]
E["Excess Cash Evaluation\nBoard reviews for buybacks or dividends"]
end
A --> B
B --> C
C --> D
D --> E
```
Cash generated by core manufacturing operations directly funds the company's aggressive growth initiatives. Kirkhorn notes that even through global supply disruptions and macroeconomic instability over the preceding years, Tesla never paused or cut back capital investments in expansion. Once organic growth needs are satisfied, Tesla evaluates opportunistic expenditures based on rate-of-return hurdles—such as targeted efficiency retrofits at the Fremont factory. Any remaining surplus is reviewed by the board of directors to determine whether to distribute capital via share buybacks or dividends. Kirkhorn concludes the finance presentation by stressing that heavy engineering innovation lowers baseline manufacturing costs, unlocking consumer affordability and high-volume demand, which in turn fuels the cash generation required to achieve manufacturing scale and accelerate sustainable energy.
- Working capital variance: Multi-billion-dollar intra-quarter swings driven by delivery cycles.
- Reinvestment discipline: Zero pullback on expansion projects throughout the pandemic.
- Kirkhorn emphasizes: “preserving daily operations is the engine that then generates cash that allows us to reinvest into the growth of the business.”
[03:03:07](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=10987s) Announcement of Gigafactory Mexico in Monterrey
Following an intermission, Elon Musk returns to the stage to deliver what he terms the most significant announcement of Investor Day: Tesla will build its next Gigafactory in Mexico, located near Monterrey in the state of Nuevo León. Musk acknowledges the presence of local dignitaries in the audience, welcoming the Governor of Nuevo León (Samuel García) and the state's Secretary of Economy.
Musk clarifies that the establishment of Gigafactory Mexico is purely supplemental to Tesla's existing manufacturing footprint rather than a reallocation of existing capacity. Tesla intends to expand production across every existing site, including Fremont (California), Giga Nevada, Giga Texas (Austin), Giga Berlin, and Giga Shanghai. As Musk puts it: “This is not, to be clear, moving output from anywhere to anywhere. It is simply about expanding total global output.” A formal groundbreaking ceremony and eventual grand opening will follow as construction gets underway.
[03:05:20](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11120s) Q&A: Next-Gen Vehicle Architecture and Manufacturing Paradigms
Musk opens the Q&A session by inviting the broader executive engineering bench to join him on stage, encouraging questions that address long-term structural value rather than short-term quarterly guidance. The first question comes from Rod Lache of Wolfe Research, who asks about the deployment timeline for the next-generation vehicle platform and whether the radical packaging and assembly methods introduced earlier will be retrofitted into existing lines building the Model Y.
Musk explains that retrofitting an active manufacturing plant requires taking production offline for an extended duration, which Tesla avoids whenever possible: “retooling a factory means bringing the factory down for an extended period of time and that's that would prefer not to do that, I think.” Instead, foundational architectural shifts—such as fully unboxed assembly paradigms—are prioritized for entirely new vehicle programs. However, evolutionary upgrades are routinely phased into current lines; Musk cites how Model Y manufacturing has iteratively adopted single-piece rear gigacastings, front gigacastings, and structural battery packs.
VP of Vehicle Engineering Lars Moravy agrees, noting that while innovations eventually migrate backward into older vehicle architectures over the long term, avoiding downtime on high-volume lines is critical. Moravy confirms that Gigafactory Mexico will be configured to manufacture the next-generation platform, with other facilities slated to build it as well, calling it “a huge volume product” that the team will stand up as fast as possible.
- Facility downtime avoidance: Radical assembly retooling is held for clean-sheet vehicle factories.
- Incremental Model Y updates deployed: Rear mega-castings, front mega-castings, and structural pack integration.
- Next-gen vehicle footprint: Production will be anchored by Giga Mexico and subsequently added to other Tesla plants.
- Lars Moravy on rollout speed: “As far as a timeline goes, um you know, we're going to go as fast as we can left to right.”
[03:08:19](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11299s) Lithium Mining, Clay Extraction, and Refining Bottlenecks
The Q&A session opens with Adam from the audience congratulating the Tesla team on the newly announced Gigafactory in Mexico ("enhorabuena, Mexico vamos"). He asks Elon Musk about applying first principles thinking to the mining and extraction sector—specifically referencing a patent Tesla filed regarding using sodium chloride to extract lithium from clays and spodumene. Adam inquires how this fits into their strategy to modernize an industry whose slow pace and high costs could become a critical gating factor.
Musk clarifies that Tesla addresses whatever the limiting factor is at any given moment, though their preference is to do as little of that direct work as possible. As Elon Musk emphasizes: “So we don’t want to get into the mining or refining sector. We will do that if we have to.” He identifies refining capacity as the genuine bottleneck across the entire supply chain, rather than raw mining: “There’s it’s really the refining capacity that is the biggest choke point.” Drew Baglino joins in, confirming that this bottleneck is precisely why Tesla is building its own lithium refinery in Corpus Christi, Texas. Baglino explains that while Tesla has large lithium suppliers, the company is openly showing those suppliers the scalable chemical engineering methods being piloted in Corpus Christi and on clay extraction trials. The intention is not to hoard the technology or replace the industry, but to prove out faster, cheaper techniques and transfer that knowledge back to suppliers so the entire world can extract ore more efficiently.
Musk and Baglino also point out the adjacent construction of a cathode processing facility next to Giga Texas, noting that there is essentially no large-scale cathode production in the United States. While Tesla would prefer established vendors handle it, the lack of domestic scale forced Tesla to apply first-principles thinking to invent new cathode manufacturing methods, which they likewise intend to share with industry partners once validated.
```mermaid
flowchart LR
accTitle: Tesla Refining Strategy Flow
accDescr: Diagram showing Tesla developing refining and cathode processes and transferring technology to suppliers.
subgraph Internal ["Tesla First-Principles R&D"]
A["Corpus Christi Lithium Refinery"]
B["Texas Cathode Processing Facility"]
end
subgraph Industry ["Supply Chain Scaling"]
C["Lithium Mining & Refining Suppliers"]
D["Domestic Cathode Production Base"]
end
A -->|"Transfer Scalable Process Knowledge"| C
B -->|"Deploy Low-CapEx Techniques"| D
```
- **Lithium refining location:** Corpus Christi, Texas.
- **Cathode manufacturing:** A dedicated cathode processing plant is under construction adjacent to Gigafactory Texas.
- **Technology transfer model:** Tesla tests scalable clay extraction and refining techniques internally to validate them, then shares the intellectual property and playbooks with existing commercial suppliers.
---
[03:12:00](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11520s) Accelerating Renewables and Grid-Scale Energy Storage
Ben Kallo from Baird asks whether Tesla can do more to nudge the rest of the renewable energy sector to speed up deployment, given its central role as a pillar of Master Plan 3. Musk responds with a direct call to aspiring business leaders, highlighting the immense commercial opportunity in raw material processing: “I can say that if there’s like entrepreneurs out there that want to have like a guaranteed chance of success, it would be refining lithium or anodes and cathodes or any any materials whatsoever for lithium-ion cells. It is a no-brainer.”
Drew Baglino expands on how Tesla accelerates renewables indirectly through its stationary storage and grid integration products. He explains that renewable energy deployment is often throttled by grid instability and price cannibalization. In Texas, for instance, there is immense wind generation at night—so much that grid operators face curtailment issues because baseload nuclear plants cannot simply be cycled off. By aggressively driving down the cost of stationary storage (such as the Megapack) and turning electric vehicles into flexible loads that charge when green energy is abundant, renewables become significantly more economically viable. Deploying ultra-low-cost grid-scale batteries directly unlocks faster wind and solar adoption worldwide.
---
[03:13:36](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11616s) Vehicle Model Strategy and Bidirectional Charging
Philippe Houchois from Jefferies poses two distinct questions: first, how Tesla plans to hit its 2030 target of 20 million vehicles per year with a minimal portfolio—avoiding consumer fatigue over seeing the same car on every corner—and second, what Tesla's latest thinking is on bidirectional charging (vehicle-to-home or vehicle-to-grid), given past reluctance.
Baglino takes the bidirectional charging question first, explaining that the historic lack of support was simply an issue of engineering priorities rather than an ideological refusal. As Tesla redesigned its onboard power electronics, engineers discovered a way to implement bidirectional power flows while simultaneously reducing the net cost of the vehicle's power electronics. Tesla is undergoing a complete power electronics retool across its vehicle lineup that will introduce bidirectional capability to all vehicles over roughly the next two years.
However, Musk injects a dose of practical reality regarding consumer behavior, arguing: “I don’t think very many people are going to use bidirectional charging unless you have a Powerwall, because if you unplug your car, your house goes dark. And this is extremely inconvenient.” Baglino adds that the vast majority of grid value comes from smart unidirectional charging—pulling power when energy is abundant and cheap—rather than pushing power back out. Musk concludes that vehicle-to-home will primarily serve as a supplemental reserve for homes that already have a stationary Powerwall, ensuring seamless backup power when the car is driven away.
Turning to the portfolio question, Musk rejects the automotive industry's traditional habit of producing hundreds of superficial trim and styling variants:
> “I mean, what's happened with conventional cars is people have run out of things to do, so they run out of things to do, they just end up reshuffling the deck and you have pretty much the same—I mean, how many variants of car are there on the road? There's like hundreds. But they're—are they good variants? No, mostly not. They're just variants for the sake of variants.”
Musk predicts automotive manufacturing will follow the path of the smartphone market, where hundreds of unique flip-phone form factors converged into a handful of standardized, highly functional designs. When asked how many vehicle models Tesla might need to reach 20 million units annually, Musk estimates: “I don't know. Not that many. Um, really 10? I don't know. Not that many.”
- **Bidirectional rollout:** Power electronics redesign underway across the fleet; rollout anticipated across all models over the next two years.
- **Use case focus:** Primarily designed to supplement a Tesla Powerwall, rather than act as a standalone home power source.
- **Model portfolio estimate:** Roughly 10 distinct models globally to deliver 20 million vehicles per year.
---
[03:17:32](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11852s) China Market Expansion and Demand Elasticity
George Gianarikas from Canaccord Genuity asks about Tesla's strategy for expanding market share in China, and whether rising geopolitical tensions between the United States and China pose a threat to the company's long-term ambitions there.
Musk jokes to Tom Zhu, who oversees Tesla’s global manufacturing and operations: “Hey Tom, don’t sweat too hard,” sparking laughter from the audience. Zhu addresses the question by emphasizing that Tesla’s market share in China continues to grow strongly. He points to the sharp demand surge following price adjustments made in early 2023, noting that the price cuts generated far more orders than the Shanghai factory could supply. Zhu underscores that Tesla’s core strategy in China is grounded in simple market fundamentals: offering exceptional product value at an accessible price point completely eliminates demand risk, allowing the team to concentrate entirely on relentless cost reductions.
[03:18:40](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11920s) China Market Share and Geopolitical Positioning
Tom Zhu concludes his response regarding Tesla’s operations in China, emphasizing that competitiveness comes from efficiency improvements across the supply chain and factory floor, which are then passed on to customers in the form of lower prices. As long as Tesla continues executing on this strategy, he remains confident about holding market share. Addressing geopolitical tensions, he points out that Tesla creates substantial employment not just internally, but across its extensive domestic supplier ecosystem, providing a notable lift to the local economy. As Tom Zhu puts it: “I think you know, as long as we're needed um in this country, I don't see there much of a risk of that.”
[03:19:25](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11965s) Operational Efficiency, Measuring Sticks, and Price Elasticity
Colin Rush from Oppenheimer steps up to ask about Tesla’s operational efficiency metrics and how the leadership team approaches and tracks learning velocity across different parts of the organization. An executive answers bluntly: “I mean, I'll I'll say one very high-level thing, which is you can't improve something you don't measure, and we're like ruthless measurers at Tesla.” Once an organization tracks every metric feeding into operational efficiency, it opens up a clear path to fix problems.
```mermaid
flowchart TD
accTitle: Ruthless Measurement to Demand Scaling
accDescr: Visual flowchart outlining Tesla's operational and demand logic from tracking metrics to fleet autonomy.
subgraph Operational_Discipline["Operational Discipline"]
M["Ruthless Measurement of Metrics"] --> O["Operational Efficiency Gains"]
O --> C["Cost Reductions Passed to Buyer"]
end
subgraph Demand_Dynamics["Demand Dynamics"]
C --> A["Increased Affordability"]
A --> D["Exponential Demand Elasticity"]
end
subgraph Asset_Utility["Fleet Asset Multiplier"]
D --> F["Mass Global Vehicle Fleet"]
F --> AU["Autonomy Enabled (50–60 hrs/wk usage)"]
AU --> V["Historical Asset Value Increase"]
end
```
Elon Musk pivots from operational metrics to demand, reiterating a core truth that he feels investors often overlook. Musk explains that the desire to own a Tesla is near universal; the only bottleneck is whether a customer has the money to pay for it. He remarks that inside an investor room where incomes dwarf vehicle purchase prices, it is easy to view purchases through the lens of value-for-money rather than raw affordability, but ordinary consumers are strictly constrained by their budget. For that reason, one cannot simply double prices without destroying demand—in the theoretical limit, an infinitely desirable vehicle priced at $10 million sells nothing because almost no one can buy it. In assessing recent price changes, Tesla discovered that price elasticity is exceptionally pronounced: even tiny price drops unleash enormous demand.
Turning to the longer term, Musk describes autonomy as the ultimate utility and valuation inflection:
- A typical passenger car sits idle most of the time, seeing only about 10 to 12 hours of usage per week while racking up high parking costs.
- Fully autonomous vehicles could see 50 to 60 hours per week of active utility, eliminating significant parking overhead.
- Because of this 5x increase in operational utility, Musk asserts: “if this is true, then as autonomy is effectively turned on for the fleet, uh it may be the probably will be the biggest uh asset value increase in history overnight.”
[03:22:50](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=12170s) Next-Gen Platforms, Product Cannibalization, and Supply Chain Realities
Emmanuel Rosner from Deutsche Bank asks about the ramp of Tesla’s next-generation platform, referencing a presentation slide showing two mystery silhouettes—one resembling a van and another resembling a smaller vehicle or compact car. Rosner asks what vehicle segment Tesla will prioritize first and how Tesla plans to prevent a 50% cheaper next-generation vehicle from cannibalizing its existing lineup.
Elon Musk brushes aside concerns about cannibalization, asserting that product desire is practically limitless: “Demand for our vehicles in terms of desire to own them may as well be infinite.” When affordability improves, demand surges; the true bottleneck is manufacturing at scale:
> “The hard part is building the cars and the entire supply chain that goes with the cars. Uh this is a logistics challenge of extraordinary difficulty.”
Musk warns that during an exponential production ramp, all parts and subcomponents must scale concurrently. A failure in a single upstream link—whether triggered by an earthquake, flood, fire, or revolution—halts the entire assembly line.
Tesla’s supply chain leaders, Roshan Thomas and Karn Budhiraj, expand on the day-to-day discipline of maintaining production. They emphasize that in Tesla manufacturing, “perfection is a passing grade.” Supply chain teams must possess granular subject-matter expertise across every tier. Over recent years, the company has had to engineer bespoke solutions to navigate:
- Shifting international trade tariffs
- Global maritime shipping bottlenecks and container port congestion
- Severe semiconductor shortages
- COVID-19 lock-downs and regional infection spikes (such as in Malaysia's chip testing and packaging facilities)
- Natural disasters and industrial fires (including a factory fire at a major semiconductor fabrication plant in Japan)
To insulate against these risks, Tesla has established direct relationships deep into the lower tiers of the supply chain rather than relying purely on tier-1 suppliers, utilizing dual- and triple-sourcing strategies wherever possible.
Lars Moravy adds that just as the Model S/X and Model 3/Y share architectural DNA, the next-generation platform is engineered to support multiple vehicle segments rather than a single standalone model. It is designed alongside suppliers to expand into segments Tesla has not yet penetrated. Moravy reminds the audience that skeptics originally claimed the sedan market was dead when Model 3 debuted, yet it became a massive global seller because affordability and desirability transcend rigid segment lines.
Musk closes the section by invoking a classic military axiom: “battles are won with tactics, wars are won with logistics.” When an automaker produces millions of units, excessive over-capacitization becomes impossible because it inflates capital expenditures, builds wasteful supplier idle capacity, and inflates warehouse inventories. Triple-sourcing critical, large components is also deceptive; Musk compares it to flying an airplane with three engines where the failure of any single engine brings down the plane. Instead, Tesla engineers manageable safety buffers, pays expedite freight when inevitable bottlenecks arise, and treats production scaling as solving “10,000 logistics problems” simultaneously. Among those, battery cell supply is paramount, leading Tesla to intentionally build cell capacity well beyond vehicle manufacturing projections to avoid bottlenecks.
[03:29:02](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=12542s) Stationary Storage Buffer and Quasi-Infinite Megapack Demand
Elon Musk addresses the strategic dilemma of what to do if battery cell output exceeds vehicle production needs or if car factories stall. The natural buffer, he explains, is stationary energy storage: Powerwall and Megapack. It is far simpler to scale stationary storage output up and down than it is to adjust complex vehicle manufacturing assembly lines. By intentionally over-capacitizing on cell and pack production, Tesla creates a safety net where excess volume can immediately be funneled into stationary storage products.
Furthermore, the capital expenditure needed for Megapack production is minimal compared to the massive investments required for vehicle assembly. Musk emphasizes that demand for utility-scale stationary storage is virtually limitless as long as Tesla remains cost-competitive against traditional utility solutions:
> “It basically as long as we are competitive with utilities, we can sell as many mega—quasi-infinite demand for that, really. Or multi-ter—many terawatt-hours, and we got a long way to go to get to many terawatt-hours per year.”
```mermaid
flowchart LR
accTitle: Battery Cell Allocation Flow
accDescr: Flowchart showing cell production feeding vehicle lines or buffering into stationary storage.
subgraph Supply["Cell Supply"]
A["Battery Cell & Pack Production"]
end
A -->|"Primary Demand"| B
A -->|"Excess Overcapacity Buffer<br/>(Quasi-Infinite Demand)"| C
```
[03:30:13](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=12613s) Regional Cost Localization vs. Global Process Standardization
Dan Levy of Barclays asks whether Tesla's cost-reduction strategies differ significantly across regions—pointing out how Giga Shanghai achieved a far superior cost structure compared to Fremont—or if Tesla applies a uniform global approach. Tom Zhu steps up to explain that Tesla’s core manufacturing philosophy is deeply rooted in localizing supply chains as heavily as possible. In China, over 95% of the supply chain is localized across Tier 1, Tier 2, and Tier 3 suppliers, unlocking enormous cost reductions. Giga Shanghai also benefits from the skilled workforce in the Yangtze River Delta region; out of more than 30,000 employees in China, fewer than 20 are expatriates. While labor markets and supplier ecosystems inevitably vary around the world, Tesla strives to replicate this intensive localization model across every Gigafactory.
Tom Zhu notes that other structural cost advantages remain universally consistent, such as the direct-to-consumer sales model without franchised dealers, strict operational expenditure controls, and spending zero dollars on traditional marketing or advertising. Lars Moravy adds that core powertrain and electrical architecture engineering concepts apply universally regardless of geography. Zach Kirkhorn expands on this, highlighting a strategic shift toward full factory standardization:
> “And as we move towards the next-generation platform, I think the term you used, Lars, is copy-paste. So, to get it right from the first time... try to have as much standardization and commonization as possible, which allows us to go as quickly as possible with expansion.”
- In China, more than 95% of Tesla's supply chain is localized from Tier 1 through Tier 3.
- Giga Shanghai operates with over 30,000 local workers and fewer than 20 expatriates.
- Whereas previous Model Y factories involved bespoke incremental engineering adjustments, future Gigafactories building the next-gen platform are planned around a standardized "copy-paste" footprint for rapid global scaling.
[03:34:19](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=12859s) Long-Term Operating Margins, CapEx Context, and FSD Auto-Labeling
Omer Rokah raises questions covering both financial forecasting and autonomy tech. For Zach Kirkhorn, he asks about operating margins, long-term growth, and whether the previously shown $150–$175 billion CapEx figure represents explicit guidance through 2030. For Ashok Elluswamy, he asks how Tesla’s multi-trip reconstruction differs from Mobileye’s Road Experience Management (REM) mapping and inquires about Hardware 4 (AP4).
Kirkhorn clarifies that the CapEx figures should not be interpreted as formal multi-year financial guidance. Instead, they were shared transparently to demonstrate internal modeling showing that scaling to 20 million vehicles per year and 1 TWh of annual energy storage capacity is fully achievable within Tesla's expected cash generation capabilities. On margins, Kirkhorn states that Tesla aims to continuously reduce operating expenses as a percentage of revenue and strip out product costs to keep automotive gross margins healthy. He highlights that autonomous software represents massive upside beyond the core hardware:
> “And then there's a software portion that's added on top of that. And so, Elon has commented on this many times about the impact that full self-driving software can have on the economics of the business. And in that space, you know, profitability, operating margins would be impacted dramatically.”
Addressing multi-trip reconstruction, Ashok Elluswamy explains that unlike HD map architectures, Tesla's method is strictly an auto-labeling pipeline. The fleet collects real-world video clips across repeated trips to reconstruct detailed 3D scenes offline. This offline reconstruction provides ground-truth geometric supervisory labels for training the neural nets, preserving a fully scalable vision system that operates without relying on pre-compiled HD maps. Elon Musk chimes in to emphasize the computing infrastructure powering these models, noting that Tesla operates one of the largest neural network training clusters in the world. Musk states that Tesla expects its AI compute capability to increase by an order of magnitude by the end of 2023, followed by another order of magnitude increase by late 2024, powered by a mix of Nvidia GPUs and custom Dojo hardware.
[03:37:40](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=13060s) Progress and Scaling of the Dry Battery Electrode Process
Alex Potter of Piper Sandler asks Drew Baglino for an update on the 4680 dry battery electrode (DBE) process. Referencing the factory tour where investors could see the automated lines through the observation windows, Potter observes that the technology is clearly no longer just a "science project," and asks about yield trends, tool speeds, and execution relative to initial internal targets.
Drew Baglino confirms that the facility is a fully operational, automated factory churning out significant volumes of dry electrode. He acknowledges that reaching commercial maturity is a continuous spectrum and that the team maintains exacting standards:
> “Every week that goes by making progress towards those end goals, whether it's speed of the tool, yield of the—of that process or the downstream process. We—we haven't stalled out yet on the rate of progress either. And that's both on the anode and the cathode side.”
Baglino remarks that steady weekly advancements across tool speeds and yields on both the anode and cathode lines ensure that cell manufacturing continues to scale toward the overcapacity buffer strategy outlined earlier by Musk.
[03:39:14](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=13154s) Scaling the Dry Electrode Process and 4680 Manufacturing
The engineering leadership explains that Tesla’s internal manufacturing strategy prioritizes rapid experimentation over rigid adherence to decisions made quarters prior. Senior Vice President Drew Baglino notes that having active production lines in the Texas factory allows them to test competing methodologies in real time. For instance, the factory operates two separate anode lines testing slightly different mechanical implementations for feeding dry powder into the tool, pitting the two approaches against each other to see which yields higher output and superior performance. As Drew Baglino recaps, Elon Musk has repeatedly drilled a core operating principle into the team: “It's okay to scrap equipment or money. It's not okay to scrap time.”
```mermaid
flowchart LR
accTitle: Dry Electrode Scaling and Validation Pipeline
accDescr: Diagram showing the development path from Maxwell lab scale to competing multi-ton factory lines.
subgraph LabScale["Laboratory Scale"]
A["Maxwell Acquisition<br/>(Proof of Concept)"]
end
subgraph FactoryCompetition["Giga Texas Parallel Lines"]
B1["Anode Line Version A<br/>(Powder entry variant 1)"]
B2["Anode Line Version B<br/>(Powder entry variant 2)"]
end
subgraph HighVolume["Mass Production Target"]
C["High Yield &<br/>Continuous Output"]
end
A --> B1
A --> B2
B1 -->|"Yield & uptime competition"| C
B2 -->|"Yield & uptime competition"| C
```
Elon Musk elaborates on why dry electrode technology represents such a punishing engineering hurdle. While Tesla acquired Maxwell Technologies specifically to capture its proprietary dry electrode intellectual property, the gulf between bench-scale demonstration and multi-ton industrial throughput proved colossal. Musk remarks, “It seems likely that we will be able to scale it to volume this year.” Baglino adds that the factory is expanding output week over week, tracking toward an internal ramp trajectory of approximately 1,000 vehicle sets per week per quarter. At current rates, individual tools process roughly one ton of powder per hour, bringing unique mechanical challenges into play that never exist in a lab environment. Even an escape of 0.1% dust fines into an enclosure can contaminate air systems and short out sensitive electronic controls, requiring the team to systematically discover and eliminate brand-new physical failure modes.
- **Pacing Rule:** Time is treated as the strictly non-renewable resource, making the financial write-off of experimental tooling acceptable if it buys speed.
- **Factory Competition:** Parallel anode lines are running head-to-head on the Giga Texas floor to evaluate competing powder-feed tooling geometries.
- **Tooling Throughput:** Production tooling handles approximately one ton of raw electrode material per hour per machine.
- **Micro-Failures at Scale:** Escapes as low as 0.1% of airborne powder fines into equipment bays generate dust buildup that causes unexpected electrical shorts.
---
[03:41:56](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=13316s) Next-Generation Vehicle Inquiries and Product Event Plans
During the final portion of the Q&A session, an analyst named Chris poses a three-part question centered on Tesla’s upcoming next-generation platform. Chris asks when the public might see an initial prototype, requests specific hardware details regarding dimensions, content, and powertrain performance, and asks whether vehicle production could begin inside existing Gigafactories before construction on the newly announced Gigafactory Mexico facility is finished.
Musk shuts down the inquiry to protect product confidentiality and marketing timing, declining to provide specs or milestone dates. Rather than reveal the vehicle's architecture or manufacturing rollout cadence prematurely from the Investor Day stage, he indicates that the company will reveal the vehicle on its own terms at an unannounced future date. Addressing the question, Elon Musk states, “We will have a proper sort of product event, but that we would be jumping the gun if we were to answer your questions.”
---
[03:42:54](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=13374s) AI in Manufacturing, Existential AI Safety, and FSD Neural Networks
An investor asks Musk two closing questions: how his leadership philosophy has shifted as Tesla doubled and prepares to double again, and whether recent breakthroughs in generative artificial intelligence could make manufacturing vehicles substantially easier. Musk quickly dismisses the immediate utility of generative AI on factory shop floors, stating, “I don't see AI helping us make cars anytime soon.” He jokes that if artificial intelligence reaches the sophistication required to handle complex physical assembly lines, human labor would be largely moot and everyone might as well relax.
The conversation turns somber as Musk reiterates his long-standing anxieties regarding rapid advancements in artificial intelligence. He argues that artificial intelligence presents severe existential risks and calls for public-interest regulatory oversight over frontier model development. Reflecting on his role in the tech ecosystem, Musk admits, “I fear I may have done some things to accelerate it, which is I don't know.” Turning to practical applications, he distinguishes broader AGI concerns from Tesla’s vision-based Full Self-Driving stack. Because modern road infrastructure was purpose-built for biological eyes and biological neural networks, Musk maintains that vision cameras coupled to artificial digital neural networks are the only logical architectural solution to autonomy. Team members remark that the multi-layered neural net model within the vehicle has grown so dense and stacked with interconnected subnetworks that internal diagnostic tools crash simply attempting to visualize the graph. Closing the event as the clock reaches 7:00 PM, Musk confesses that thinking deeply about the trajectory of AI stresses him out, before thanking the audience and concluding Investor Day.
- **Regulatory Oversight:** Musk calls for proactive government regulatory bodies to monitor leading AI labs and safeguard public safety.
- **FSD Visual Architecture:** The vehicle software relies on digital neural nets processing camera streams to mirror human biological perception.
- **Graph Complexity:** The onboard driving network consists of extensive "nets upon nets upon nets," pushing visualization software past its memory and rendering limits.
## Video description
We plan to host Tesla's 2023 Investor Day on March 1, 2023. Our investors will be able to see our most advanced production line as well as discuss long term expansion plans, generation 3 platform, capital allocation and other subjects with our leadership team.
Part 2 | Executing Master Plan 3
0:40:21 - Vehicle Design with Franz Von Holzhausen & Lars Moravy
0:52:43 - Powertrain with Colin Campbell
1:01:17 - Electronic Architecture with Pete Bannon
1:09:53 - Software with David Lau
1:20:18 - Full Self-Driving with Ashok Elluswamy
1:30:12 - Bot Update with Elon Musk and Ashok Elluswamy
1:35:35 - Charging with Rebecca Tinucci
1:44:04 - Supply Chain with Karn Budhiraj & Roshan Thomas
2:05:05 - Manufacturing with Tom Zhu & Drew Baglino
2:18:15 - Energy with Drew Baglino & Mike Snyder
Part 3 | Impact at Tesla
2:31:15 - Impact with Laurie Shelby and Brandon Ehrheart
2:37:50 - Financials with Zach Kirkhorn
Q&A
3:03:24 - New Gigafactory Location Announcement
3:05:20 - Q&A
特斯拉2023年投资者日
在特斯拉2023年投资者日上，高管团队概述了“宏伟蓝图第三篇章”（Master Plan Part 3），强调向可持续能源经济转型完全具备可行性，无需严苛的紧缩生活。该规划主张建立在240太瓦时储能、30太瓦可再生能源及约10万亿美元资本投资的基础之上，通过电网改造、车队全面电动化与自动驾驶化、普及热泵、工业高温储热以及航运航空脱碳五大支柱实现减排。此外，公司设计与工程团队还回顾了从早期车型到Cybertruck的开发历程，展示了特斯拉如何通过颠覆传统造车逻辑与制造工艺来重塑生产效率。
特斯拉发布“宏伟蓝图第三篇章”，提出全球向可持续能源经济转型需要240太瓦时电池储能、30太瓦可再生电力以及10万亿美元的长期资本开支。
规划列出五大脱碳支柱：重构电网、交通工具全面电动化与自动驾驶、部署建筑及工业热泵、工业高温储热与绿氢应用，以及航运与航空的清洁化。
设计与工程主管回顾了从Model 3“生产地狱”中吸取的经验，并介绍Cybertruck如何以特种不锈钢材质为核心，通过简化冲压和取消喷漆颠覆传统制造流程。
Reading summary
Tesla framed its long-term roadmap around Master Plan 3, arguing that a fully sustainable global energy economy is physically and economically viable through electrified transport, high-volume stationary storage, and radical manufacturing simplifications. Executives presented multi-disciplinary initiatives across powertrain design, electrical architecture, vehicle assembly, and artificial intelligence to support long-term targets of 20 million vehicles and 1 terawatt-hour of stationary storage annually.
Macro Energy Blueprint: Master Plan 3 (15:23)
Drew Baglino and Elon Musk argued that the global fossil fuel economy is fundamentally inefficient, with over 80% of primary energy sourced from fossil fuels but less than one-third converted to useful work. In internal combustion engine vehicles, less than 20% of extracted crude oil energy pushes the vehicle forward once upstream extraction, refining, and thermal losses are factored in. Tesla outlined five core pillars to decarbonize the planet: repowering the existing grid with renewables and utility storage, transitioning passenger and commercial fleets to electric vehicles, switching heating to heat pumps, deploying high-temperature thermal storage and green hydrogen for industrial processes, and powering aviation and maritime transport with synthetic clean fuels and high-density structural battery packs (targeting 450 Wh/kg).
According to Tesla's modeling, the physical footprint required for solar and wind infrastructure to support this transition occupies less than 0.2% of Earth's land surface—an order of magnitude smaller than the 12.5% currently dedicated to intensive agriculture. Musk emphasized that closed-loop battery recycling will eventually reduce new mining extraction demand by an order of magnitude, projecting that a fully sustainable energy economy is achievable within a human lifetime based strictly on thermodynamic and engineering realities.
Powertrain and Low-Voltage Architecture Redesign (58:34)
Colin Campbell revealed Tesla’s next-generation powertrain, engineered to reduce silicon carbide transistor usage by 75% without compromising efficiency, eliminate rare earth elements entirely from its permanent magnet motor, and cut drive unit manufacturing costs to roughly $1,000 while requiring 50% less factory floor space than Giga Texas. The drive unit is designed to accommodate any cell form factor and chemistry, insulating vehicle production from specific battery supply bottlenecks.
Pete Bannon outlined a parallel overhaul of the low-voltage vehicle architecture, confirming that starting with Cybertruck, all future vehicle platforms will transition from a 12-volt to a 48-volt bus. Quadrupling the voltage reduces current requirements fourfold, slashing resistive power dissipation (I²R) sixteenfold. This thermal reduction allows engineers to shrink wire harness diameters, minimize controller dimensions, and downsize eFuses. Next-generation platforms will feature 100% in-house controllers communicating over high-speed Ethernet. Additionally, Baglino stated that a comprehensive retool of vehicle power electronics will introduce bidirectional charging capability across the lineup within two years, though Musk noted widespread consumer utility depends on pairing vehicles with stationary storage like Powerwalls to prevent home blackouts when unplugged.
Unboxed Manufacturing and Factory Scaling (45:55)
Lars Moravy challenged the century-old automotive manufacturing paradigm established by Henry Ford, which sequentially stamps, welds, paints, and partially disassembles car shells before interior parts are manually installed in cramped quarters. Tesla is designing an 'unboxed' assembly model that divides the car into discrete, pre-tested subassemblies before final assembly, significantly decreasing factory footprint and non-value-added material handling. For manufacturing throughput, Tom Zhu established target operational benchmarks of 90% Overall Equipment Effectiveness (OEE) and a 45-second vehicle completion cycle time, while Musk announced a new Gigafactory in Monterrey, Mexico, to add incremental production volume alongside existing plants in Fremont, Nevada, Texas, Berlin, and Shanghai.
In cell production, Baglino reported that dry electrode 4680 equipment at Giga Texas operates without manual intervention, achieving more than 20 times the peak tool productivity demonstrated during Battery Day pilot trials. When fully scaled across facilities, the 4680 manufacturing architecture is projected to cut factory footprint and volume by 10x relative to traditional cell lines. Addressing upstream mineral processing choke points, Tesla began building its own lithium refinery in Corpus Christi, Texas, to demonstrate lower-cost chemical engineering processes to commercial suppliers.
Autonomous Software, Fleet Telemetry, and Robotics (80:33)
Ashok Elluswamy argued that vehicle autonomy is essential to planetary decarbonization because continuous fleet utilization reduces the absolute number of vehicles required to deliver global passenger miles. To address data annotation bottlenecks in Full Self-Driving (FSD), Tesla built an automated labeling engine that aggregates multi-trip fleet camera footage into 3D reconstructions, which are then imported into physics simulation engines to generate adverse weather and lighting edge cases. David Lau added that Tesla uses real-world telemetry from 123 million miles driven daily to reconstruct collisions and distribute safety improvements over the air, including algorithmic adjustments to seatbelt tensioning.
Musk highlighted that the humanoid robot Optimus directly inherits the computer hardware, vision architecture, and neural network foundations of FSD, with robotic actuators custom-designed by the engineers behind Tesla’s vehicle drive units. While Musk expressed ongoing anxiety regarding unchecked artificial general intelligence—calling for public-interest regulatory oversight—he dismissed suggestions that generative AI could assist vehicle manufacturing in the near term, asserting that assembly lines remain far beyond the capabilities of contemporary foundation models.
Stationary Storage Buffering and Long-Term Capitalization (138:14)
Mike Snyder reported that Tesla's dedicated Megafactory in Lathrop, California, transformed a former retail warehouse into an active Megapack XL manufacturing facility in under 12 months, while plug-and-play field designs tripled site installation speed. Drew Baglino noted that Tesla has shipped more than 1.4 terawatts of power electronics across its vehicle and storage fleets, deploying 'Virtual Machine Mode' firmware to provide synthetic grid inertia that stabilizes AC grid frequencies without traditional mechanical turbines. Musk characterized Megapack utility demand as quasi-infinite at competitive costs, emphasizing that stationary storage functions as a strategic shock absorber that can rapidly consume battery cells whenever vehicle production lines fluctuate.
CFO Zach Kirkhorn addressed the corporate scale required to deliver 20 million cars per year and 1 terawatt-hour of annual stationary storage. He clarified that an internal estimate of $150 to $175 billion in cumulative capital expenditure is an illustration of funding feasibility via expected operating cash generation rather than formal multi-year guidance. On operating leverage, Kirkhorn noted that non-GAAP operating expenses have fallen 60% as a percentage of revenue since 2018, while the base Model 3 already achieves a lower five-year total cost of ownership per mile than a Toyota Corolla.
Tesla's presentation framed its expanding technological and manufacturing footprint as an integrated operational model designed to deliver the thermodynamic and scale requirements of Master Plan 3. However, fulfilling these targets relies on unproven execution milestones, including the industrial-scale validation of unboxed vehicle assembly, the ramp of dry-electrode 4680 cell manufacturing, and resolving technical and regulatory hurdles surrounding unsupervised autonomous driving.
## Detailed walkthrough
### [11:18](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=678s) 投资者日开场与整体议程介绍
特斯拉首席财务官 Zach Kirkhorn 在得克萨斯州奥斯汀全球总部登台致辞，向现场及线上观众表示欢迎。Zach 回顾了特斯拉的发展历程，指出公司经历了产品进步、技术创新以及产量快速增长的多个不同阶段，其中最近的一个阶段便是 Model Y 车型的全球扩张与本土化生产。然而，他强调本次投资者日的核心并不是讨论本季度或下季度的短期财务状况，而是放眼更长远的未来。
Zach 将整场演示分为三个核心部分：首先从宏观层面切入，探讨将整个地球转化为可持续能源生产和使用所需的条件；其次介绍特斯拉为满足这一全球需求所做出的贡献，各业务职能部门负责人将登台深入讲解团队正在开展的具体工作；最后一部分则汇总提炼，阐明这些技术与战略对特斯拉整家公司的深远意义。在演讲正式开始前，Zach 宣读了前瞻性陈述的常规免责声明，随后邀请首席执行官 Elon Musk 和动力总成与能源工程高级副总裁 Drew Baglino 上台开启第一部分——秘密宏图第三篇章（Master Plan Part 3）。
### [13:31](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=811s) 宏观论点：通往完全可持续地球的清晰路径
Elon Musk 开门见山地指出，他们今天最希望传达的核心信息，就是地球走向可持续能源存在一条极其明确的路径。这一转变既不需要摧毁自然栖息地，也绝不需要人类去过苦行僧式的紧缩生活、停止用电甚至挨冻受饿。相反，特斯拉经过缜密计算发现，地球完全能够在实现丰足的同时走向彻底的可持续化，甚至能够可持续地承载远超过现有80亿人口的更大规模文明。为了佐证这一观点，特斯拉后续将公开发布一份包含所有底层假设与具体测算过程的详尽白皮书。
Musk 坦言，让他感到震惊和意外的是，身边的聪明人几乎很少有人意识到这一点。许多专家误以为不存在一条通往可持续能源未来的现实路径，或者认为当前人口规模无法实现可持续，甚至觉得必须诉诸极端残酷的倒退手段。Musk 强调这些悲观论调全部脱离了事实，并随后引用最坚实的一揽子测算来拆解如何建立可持续能源文明。
> “The thing that I think is we wanted to convey probably more importantly than anything else that we talked about here is that there is a clear path to a sustainable energy Earth.” —— Elon Musk
### [15:23](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=923s) 化石能源的低效与电气化带来的“需求减半”效应
Drew Baglino 结合能源现状定调：当今由化石能源主导的能源经济既肮脏又极其浪费。全球超过 80% 的一次能源（Primary Energy）来自化石燃料，但其中最终仅有约三分之一转化为了有用的功或热量，其余绝大部分都在开采、提炼、输送和燃烧中白白耗散。全球一次能源年消耗量高达约 165 PWh（即 165,000 TWh）。
```mermaid
flowchart TD
accTitle: 化石能源与电气化经济的能源流动对比
accDescr: 对比化石能源高浪费率与全面电气化后能源需求减半的模型
subgraph Fossil["当前化石能源体系 (约 165 PWh/年)"]
A["一次能源输入: 80%+ 化石燃料"] --> B["开采 / 提炼 / 运输损耗"]
B --> C["终端燃烧低效率 (如内燃机仅 20%-25% 转化为动能)"]
C --> D["绝大部分转化为无用废热 (仅 1/3 做有用功/热)"]
end
subgraph Electric["全面电气化可持续体系"]
E["高效率电能生成 (光伏 / 风电)"] --> F["全流程高转换效率"]
F --> G["终端高效利用 (电机 / 热泵)"]
G --> H["总能源需求直接减半"]
end
```
Elon Musk 进一步以燃油车为例做了直观拆解：
- 内燃机通常只能将汽油中不到三分之一（甚至仅约 25%）的化学能转化为机械动能，其余全变成了毫无用处的废热；
- 算上原油的地下开采、炼油厂精炼、油罐车长途运输到加油站的全生命周期能耗，一辆典型燃油车最终转化为实际行驶动能的能量甚至不足开采原油能量的 20%；
- 许多分析机构在测算可持续未来时犯了根本性错误：他们机械地认为电气化文明需要与当前燃烧文明完全等量的能量供给。实际上，由于电机和热泵等电气设备的终端效率极高，完全电气化后，全人类所需的总能量直接缩减了一半；
- Musk 特别说明，特斯拉在建模时采用了偏悲观、保守的假设；实际上电气化能节省的能量比例很可能远超一半。
> “Over 80% of global energy, primary energy comes from fossil fuels and only 1/3 of that global energy actually ends up delivering useful work or heat.” —— Drew Baglino
### [18:08](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1088s) 秘密宏图第三篇章的宏观参数与投资估算
为了支撑这一完全可持续的经济体，核心缺口在于大规模的储能电池。Elon Musk 与 Drew Baglino 给出了宏观维度的关键资源与资金测算：
- **储能容量需求**：全球共需要约 **240 TWh**（即 240,000 GWh）的电池储能系统，涵盖纯电动车车载动力电池和应对风光间歇性的固定式储能电站（电网储能）。
- **发电功率配置**：按照储能能量与功率约 8:1 的比例测算，需要匹配约 **30 TW** 的可再生能源发电功率（以太阳能和风能为主）。
- **资本支出（CapEx）规模**：在全产业链（涵盖矿端开采、原材料精炼、电池工厂、回收体系、整车工厂等）的实际建造成本测算约为 6 万亿美元；为保持保守口径，特斯拉将其放宽设定为 **10 万亿美元**。
- **全球经济占比对比**：全球当前年 GDP 略低于 100 万亿美元。如果这 10 万亿美元的投资平摊在 10 年内完成，仅占全球经济年总量的 1%；若平摊在 20 年内完成，每年仅占全球 GDP 的 0.5%，在宏观经济上完全可承受。
- **土地占用极小**：支撑全球可持续运行所需的风能与太阳能电站占用的陆地面积不足地球总陆地面积的 **0.2%**。太阳辐射到达地表的峰值功率约为 1 GW/km²，折算全天候连续功率（除以 4 到 5 的系数）后所需占地面积依然极小。况且风电与光伏可以共用土地，风电和光伏还可部署在覆盖地球 70% 面积的海洋（海上风电与水上光伏）。
- **采矿量不增反降**：可持续电气化体系所需要的开采矿石总量，在终态下实际上显著**少于**当前化石能源开采消耗的规模，不存在无法逾越的资源瓶颈。
> “The electrified economy will require less mining than the current economy does.” —— Elon Musk
### [21:31](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1291s) 实现可持续文明的五大核心工作支柱
Drew Baglino 与 Elon Musk 总结了秘密宏图第三篇章落地执行的五大支柱领域：
1. **现有电网的清洁化改造（Re-power the existing grid）**：大规模部署太阳能、风能并匹配固定式储能电站；
2. **交通工具转向全面电动化（Switch to electric vehicles）**：乘用车、商用车全品类转向纯电动；
3. **建筑与工业采暖切换为热泵（Switch homes, businesses, and industry heating to heat pumps）**：家庭、商业和工业的供暖制冷全面采用高效热泵替代天然气等化石能源燃烧；
4. **工业高温供热与绿色氢能（High-temp heat delivery and storage）**：利用电热储能解决高温工业与化学制造流程的热源需求，并在确需氢气参与的特定化工合成中应用少量绿氢；
5. **飞机与船只的可持续燃料化/电动化（Sustainably fuel planes and boats）**：为海运与航空开发可持续燃料方案。Musk 特别表达了他的个人判断：随着动力电池能量密度的持续攀升，几乎所有的交通工具最终都将走向完全纯电动化。
### [22:20](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1340s) 火箭制造中的合成燃料与可再生电网重建
Elon Musk 与 Drew Baglino 探讨了电能向化学能转换的潜力，即便是火箭这种极端依赖化学燃料的场景，最终也能融入电气化版图。Musk 笑着自嘲火箭是唯一的直接例外，但他随即指出，利用水和二氧化碳，单靠电力就能合成甲烷（$CH_4$）和氧气（$O_2$）。他强调这一原理在火星上同样适用——火星大气几乎全由二氧化碳构成，且地下蕴藏丰富水冰，完全可以通过电力原位制备推进剂。正如 Musk 所说：
> “So, ultimately, even rockets can be electrified.”
回归地球现存电网的绿色重塑，Drew Baglino 阐释了用可再生能源为既有电网重新供电的进展。2022 年，美国电网新增发电装机容量中有 60% 为太阳能，且全美太阳能装机量正以 50% 的年复合增长率激增。Baglino 指出，这场转型已非停留在规划层面的愿景，而是在现实中高速推进；如果保持当前势头，电网重塑的目标在人们意识到之前就将全面达成。
- **火箭燃料原位合成路线**：$CO_2 + H_2O \xrightarrow{\text{电力}} CH_4 + O_2$（萨巴蒂尔反应等路线）。
- **美国电网清洁增量**：2022 年新增电网装机中太阳能占比达 60%。
- **光伏部署增速**：截至 2022 年，太阳能装机年增长率达到 50%。
---
### [23:32](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1412s) 全面转向电动汽车与自动驾驶带来的车队效率重构
Drew Baglino 继续分析全面转向电动汽车（EV）的宏观效益。仅淘汰燃油乘用车并全面切换为纯电动车，就能直接削减全球 21% 的化石燃料消耗。全球电动车制造呈现爆炸式扩张，2022 年全球电动车产量同比增长 59%，市场占有率首次突破 10% 的里程碑。Elon Musk 对此展望道：
> “And so, riding a non-autonomous gasoline car is going to be analogous to riding a horse and using a flip phone. That's basically the going to be the situation.”
两人的计算模型在车队规模测算上采取了相当保守的假设，并没有激进计入完全自动驾驶（FSD）普及后可能带来的共享化乘数效应。Musk 指出，当车队具备自主行驶能力时，车辆利用率将产生数量级的跃升，全球所需乘用车总量将大幅缩减。目前全球运行中的各类乘用车与卡车总量约为 20 亿辆，而该规划中的电动车存量模型仅设定在约 14 亿辆左右，对应年产量目标约为 8500 万辆，特斯拉自身的中长期目标则是实现年产 2000 万辆。
```mermaid
flowchart TD
accTitle: 汽车产业全面电气化与自动化转型逻辑
accDescr: 展示从传统燃油车向自动驾驶电动车转型的效能与车队规模变化
subgraph Current["现状"]
A1["全球燃油车队规模: 约 20 亿辆"]
A2["端到端能源损耗巨大"]
end
subgraph Transition["驱动要素"]
B1["全球电动车普及<br/>削减 21% 化石燃料"]
B2["全自动驾驶 (FSD)<br/>资产运营效用倍增"]
end
subgraph Target["宏观目标"]
C1["全球优化后电动车队: 约 14 亿辆"]
C2["全球电动车年产量: 约 8,500 万辆"]
C3["特斯拉远期年产能: 2,000 万辆"]
end
A1 --> B2 --> C1
A2 --> B1 --> C2
C2 -.-> C3
```
- **化石燃料消耗削减幅度**：乘用车全面电动化可减少 21% 的化石燃料用量。
- **2022 年全球 EV 增速与份额**：产量同比增长 59%，全球新车市占率达 10%。
- **全球在册车辆存量**：现阶段全球运行的汽车与卡车约 20 亿辆；预测电气化+适度自治后车队总需求缩减至约 14 亿辆。
- **全球年化产量需求**：全球年产量约 8500 万辆；特斯拉终极产能目标为每年 2000 万辆。
---
### [25:48](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1548s) 终端能效对比：从意面烹饪看 Model 3 的极度节能
Drew Baglino 深入拆解了电动车取代内燃机的核心底层优势：超高的“油井/电网到车轮”（Well-to-Wheel）终端能效。在可持续能源结构下，Tesla Model 3 的全链路综合效率是传统丰田卡罗拉（Toyota Corolla）的整整 4 倍。内燃机不仅受限于卡诺循环本身的低热效率，还额外背负了原油勘探、开采、长途海运、炼化裂解以及槽车逐级分销运送到加油站所付出的巨额能源损耗。
为了形象地展示这种差异，Baglino 列举了一个生活化的参照：Model 3 仅需烧开一锅煮意面所需的水所耗费的能量，就能在公路上行驶超过一英里；如果算上随后把那磅意面彻底煮熟所用的热能，它还能再开上一英里。Musk 笑着补充强调，那包意面只有区区 1 磅重，而 Model 3 是一台重达 4000 磅的实体工业机器；人们日常烧水做饭毫无感知，但其实相较于物理移动，产生热量需要消耗大得多的能量。
- **全链路能效倍率**：Model 3 的 Well-to-Wheel 综合效率达丰田卡罗拉的 4 倍。
- **烧水做饭能耗换算**：烧开一锅煮面用水的电能可驱动 4000 磅重的 Model 3 巡航 1 英里以上；后续将 1 磅意面完全煮熟的热能还可再开 1 英里。
---
### [26:56](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1616s) 全面普及热泵：能耗减至三分之一的“低垂果实”
在讨论建筑与工商业供暖领域时，Drew Baglino 将普及热泵定义为替代化石燃料最具经济效益、投资门槛最低的“低垂果实”（lowest hanging fruit）。目前热泵仅覆盖全球建筑物 10% 的供暖需求，部署安装量正以每年 10% 的速度递增，但要满足气候转型必须成倍加速。热泵不仅能满足居民住宅供暖，还能直接输出高达 200°C 的中低温工商业级供热。
针对观众可能存在的疑惑，Baglino 解释了热泵的底层热力学逻辑：热泵本身不“生产”热量，而是消耗少量电能机械做功，把室外环境中的热能“搬运”进室内，本质上就是反向运行的空调或冰箱。与直接燃烧天然气供暖的锅炉相比，热泵使建筑物供暖所需的总输入能量骤降至原先的三分之一。Musk 提到特斯拉全系在售车辆目前已将热泵作为出厂标配，并半开玩笑地透露，未来特斯拉或许真会考虑制造一款家用智能热泵系统。
- **建筑供暖现状与增速**：热泵目前满足约 10% 建筑供热，年装机增速约 10%。
- **工业温区覆盖**：商用/工业热泵可稳定提供高达 200°C 的中温工艺热源。
- **能效提升比率**：相比天然气与燃油直接燃烧供暖，热泵系统带来 3 倍的总体能耗缩减（即能耗降至原来的 1/3）。
- **业务横向扩展暗示**：特斯拉全系车型已标配整车热泵，未来不排除推出家用级别热泵产品。
---
### [28:18](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1698s) 高温工业热能电气化与绿氢在冶金化工中的定位
转向更难减排的高温工业与化学加工过程，Drew Baglino 指出，超过 50% 的工业热能需求温区高于 400°C，而在水泥熟料、粗钢冶炼、氮肥合成、特种化工、塑料制造以及有色金属精炼等核心工业门类中，更是普遍要求 1500°C 甚至更高的连续高温。解决路径在于使用碳石墨（耐温高达 2800°C）或碳化硅（可耐受 1500°C 左右）等专用耐高温电气化加热元件与储热介质。其运行模型是将非连续、波动性的光伏和风电在白天过剩时转化为高温热能进行绝热储存，随后由储热介质 24 小时不间断释放热能供应工业生产。
在工业用氢层面，全球目前每年消耗约 1.2 亿吨源自化石燃料重整的灰氢，主要用于合成氨及基础化学精炼。转型方案是通过绿电电解水生产绿氢，并在高炉炼铁环节推广“直接还原铁”（Direct Reduced Iron, DRI）工艺，直接用绿氢取代传统焦炭炼铁，从化学机理上切断二氧化碳排放。关于氢能源在交通运输中的炒作，Elon Musk 给出了极其明确的定论：
> “My personal opinion is that hydrogen will not be used meaningfully in transport, but and shouldn't be. If you're going to use a chemical fuel, you should use CH4 not H2.”
Musk 强调，氢气的角色应该严格限制在不可替代的工业化学还原和精炼环节中，且通过电解水制取纯氢在工业界已有数十年的成熟操作经验，根本不需要高深莫测的技术突破。
- **高温热能结构**：全球工业供热中超过 50% 运行在 400°C 以上；水泥、钢铁、化肥冶炼等工序需要 1500°C 级极端高温。
- **电热储能介质**：碳石墨结构在惰性环境下耐温可达 2800°C，碳化硅材料耐温覆盖 1500°C 工业温区。
- **氢能消耗现状与替代**：全球每年消耗约 1.2 亿吨由化石能源提炼的灰氢；需切换为电解水绿氢，并推行氢基直接还原铁（DRI）完全取缔焦炭高炉。
---
### [30:23](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1823s) 航运与航空的脱碳工程：结构化电池与高能量密度电芯
在整体能源版图中，远洋航运与航空占比较小，但属于实现全面脱碳不可或缺的一环。Drew Baglino 指出，全球远洋海运产生的二氧化碳占全球总排放量的 3%，这一领域已经完全具备纯电替代的可行性。得益于现代船舶对重量不敏感的特性，即便采用成本极低、循环寿命极长的磷酸铁锂电池（LFP），也能完全支撑大型远洋货轮完成长途巡航。航空领域对能量密度的要求虽然严苛得多，但短途客运纯电飞行在现有技术下已可实现；在长途纯电客机彻底成型前，可借助多余可再生电力合成的可持续航空燃料（SAF）进行过渡。
Elon Musk 针对长途纯电动船舶与飞机的工程实现提出了第一性原理分析：绝不能采取简单的“油改电”思路，把笨重的电池包硬生生塞进传统机身或船体架构中。正如特斯拉所验证的，最优的方案是推行“结构化电池”（Structural Battery）设计，让电池包直接充当机身、机翼或船体本身的主承力结构，从而实现极致的质量效率。Musk 指出，当电芯单体能量密度达到约 450 Wh/kg 时，结合专为电池量身定制的气动结构，长途干线客机的纯电动化在物理上便完全成立：
> “The same if if that's done with the aircraft, then you can get long-range aircraft at around with with cells at around 450 watt hours per kilogram, which you can buy right now, actually.”
- **海运排放与技术路线**：海运占全球 CO2 排放量的 3%；依靠现有的磷酸铁锂（LFP）电池即可实现远洋船舶的全电动化。
- **纯电长途客机临界指标**：需要匹配电芯单体能量密度约为 450 Wh/kg，且必须采用电池-机身一体化的结构化设计。
- **过渡期航空能源**：长途航空在电池密度完全铺开前，可采用基于绿电制备的可持续航空燃料（SAF）。
### [32:06](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=1926s) 实现可持续能源经济的可行性与投资规模
主讲人回顾了开场时给出的宏观数据指标：30 太瓦（TW）发电装机、240 太瓦时（TWh）储能容量以及 10 万亿美元的总投资额。针对观众对这些数字是否具备现实可行性的疑问，发言者明确打消顾虑：“剧透一下，这完全是可行的（Spoiler alert, it's entirely feasible）。”从产能部署的复合增速来看，风能与太阳能部署只需维持 3 倍的增长率，而两者的装机容量目前正以惊人的极速狂飙，缺口将迅速弥合；电动汽车（EV）的保有规模需要增长 11 倍，而去年全球电动车同比增长已达 60%，该差距同样会极快缩小；在储能方面，自 2016 年以来特斯拉自身储能业务的复合年均增长率（CAGR）就高达 65%，全球储能产业同样在全速加速。特斯拉确立的长期目标是实现每年生产 2000 万辆电动车，并尽早达成每年 1 太瓦时（TWh）的固定式储能电池部署。
在资本支出方面，10 万亿美元看似庞大，但放置于全球经济维度来看，它仅相当于全球一年国内生产总值（GDP）的 10% 左右。若换个对比角度，将该数额与全球现有化石能源基础设施的年化投资相比，建设完全绿色的可持续能源经济所需的年均投入，仅为继续维持与扩建传统化石能源基础设施年投资的 60%。因此，从逐年资本分配的角度衡量，这不仅绝对可控，而且在经济账上比维系传统化石能源更为划算。
- 目标技术增长倍数：太阳能与风能需增长 3 倍；电动汽车保有量需增长 11 倍（去年全球同比已增长 60%）。
- 特斯拉业务与长期目标：储能业务自 2016 年以来 CAGR 达 65%；长期产能规划为 2000 万辆 EV/年及 1 TWh 固定式储能/年。
- 投资体量对比：10 万亿美元总投资约为全球单年 GDP 的 10%；构建可持续能源经济的年投资额，仅占继续维持化石燃料体系年投资的 60%。
---
### [33:57](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2037s) 极度克制的土地占用与微乎其微的生态足迹
在土地资源占用上，主讲团队强调可持续能源基础设施所需的全部土地仅占地球总陆地面积的不到 0.2%。相比之下，全球目前进行高强度密集农业种植（intensively farmed）的土地面积占到了陆地总量的 12.5%，二者相差一个数量级以上。这意味着清洁能源的部署完全不需要侵占宝贵的基本农田、原始森林、热带雨林或各类自然生态保护区。
马斯克（Elon Musk）在台上补充道，风光发电阵列完全可以安装在人口极为稀少的大片荒漠、贫瘠的不毛之地，或是原本就不适宜做任何商业开发与居住的边远地区。马斯克明确断言：“转向可持续能源经济不仅基本不会产生有意义的生态负面冲击，相反，它将大幅缩减人类当前活动对生态系统造成的破坏（In fact, transition to a sustainable energy economy would result in a substantial reduction in current ecological impact）。”
---
### [35:02](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2102s) 全球矿产资源储量、铁基电池主导与回收闭环
关于矿物开采与原材料可持续性，团队展示了一张图表以对比物料吞吐规模：目前全球每年从地壳中开采出的各类矿石与化石燃料约为 680 亿吨（68 Gigatons，示意图中以每辆卡车代表 10 亿吨）。但在完全切换至清洁能源经济后，庞大沉重的化石燃料开采将彻底归零，取而代之的是规模小得多的新能源金属与材料开采，全球每年的总开采物质通量实际上是不增反降的。
针对截至 2050 年全球向可持续能源转型所需的全部累计材料需求，主讲人指出将其与美国地质调查局（USGS）当前探明的地质储量进行对比，没有一种关键金属会“击穿资源储备上限”。此外，人类勘探史证明了一个铁律：“我们勘探得越多，发现的储量就越多”。自 2000 年以来，伴随特斯拉及全球可再生能源供应链的爆炸式扩张，关键矿产的探明可用储量非但没有耗竭，反而持续大幅增加。
```mermaid
flowchart TD
accTitle: Sustainable Energy Raw Material Flow
accDescr: Diagram contrasting fossil fuel extraction volume with renewable materials, battery use, and recycling loop.
subgraph MiningComparison ["年度地壳开采质量对比"]
A["传统化石能源驱动时代<br/>年开采量约 680 亿吨"]
B["可持续能源经济时代<br/>开采总量大幅缩减"]
end
subgraph BatteryCycle ["清洁能源材料全生命周期"]
C["精炼加工<br/>电池级氢氧化锂/碳酸锂"]
D["电池制造<br/>磷酸铁锂占主力 / 高镍用于长续航"]
E["规模化梯次利用与闭环回收"]
end
A -.->|"彻底淘汰化石燃料提取"| B
B --> C
C --> D
D --> E
E -->|"回收料重新输入造电池<br/>长期矿石采掘需求再降一个数量级"| D
```
针对外界长期对锂资源的误解，马斯克进行了深入澄清，指出锂在地壳中含量极其丰富，是地球上最常见的元素之一，没有任何单一国家垄断或接近垄断锂供应。马斯克直言：“如果全美是世界上唯一生产锂的地方，仅凭美国本土的锂矿石，就足以让整个地球实现全面电气化（There's enough lithium ore in the United States to electrify all of Earth if the United States was the only place producing lithium）。”真正的工业瓶颈不是地下蕴藏的锂矿石，而是将粗矿冶炼加工为电池级氢氧化锂或碳酸锂的高端精炼产能；相关技术早已成熟，所缺的只是充足的行业资本投入。
对于电池材料路线，马斯克与主讲人指出：
- 镍虽然在所有关键金属中相对最棘手，但全面转型大概也仅需动用全球目前探明镍储量的约 30%，且自 2000 年以来全球探明镍储量同样处于持续净增长中。
- 高能量密度的镍基材料仅需定向供应给大型远程航空器、远洋大型船舶以及超长续航的特殊乘用车或重卡。
- 全球电气化绝大部分繁重的主力重担（heavy lifting）将由铁基电池（Iron-based cells）承担。从物理构成而言，地球按质量计算最多的成分就是铁，其次是氧。马斯克幽默地戏称地球本质上就是个“泥泞的铁锈球（muddy rust ball）”，因此人类制造铁电池绝不会有枯竭之虞。
- 伴随未来电池闭环回收体系（Recycling）的大规模成熟，维持可持续能源体系运转所需的年采矿量将在现有基础上再降一个数量级。
---
### [39:22](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2362s) 宏大叙事基调：基于物理定律的乐观主义与汽车设计团队登场
在总结宏观愿景时，马斯克强调本次投资者日的主旨不仅是向特斯拉的持股人汇报，更是面向所有“地球的投资者（investor in Earth）”。他明确表态：“我们试图传达的是一份充满希望与乐观的信息；这种乐观并非一厢情愿的空想，而是完全建立在实际物理法则与坚实数学计算之上的（What we're trying to convey is a message of hope and optimism, and optimism that is based on actual physics and real calculations. It's not wishful thinking）。”他向全场坚信，地球完全有能力且必将迈向可持续能源体系，并且这一切必定会在当代人的有生之年内彻底实现。
在全场热烈的掌声中，马斯克将舞台移交给负责整车设计与制造架构的核心骨干：首席设计师弗朗茨·冯·霍兹豪森（Franz von Holzhausen）以及车辆工程副总裁拉尔斯·莫拉维（Lars Moravy）。
弗朗茨回顾道，自己于 2008 年加入特斯拉，任务是将汽车工业中原本分离的外包设计垂直整合进公司内部。彼时整个团队规模极小，在尚无成熟电动车工业范本的环境下起步。早期的设计理念确立为围绕成本、效率与极致创新展开，同时追求前所未有的产品美感。拉尔斯则补充，在 2008 年设计初代旗舰 Model S 时，特斯拉甚至连一个属于自己的整车制造工厂都没有。
但极其精简的团队反而让设计在早期能够主导整车定义，催生出了一系列打破常规的产品创举：例如在一台四门轿跑中塞入 7 座布局（尾厢反向儿童座椅）、将车门把手完全平整收缩进车门板内，以及在座舱中控直接塞入一块颠覆行业的超大中控触摸大屏。正是这种激进突破，使 Model S 一举斩获 2013 年度《Motor Trend》年度风云车（Car of the Year）大奖，为品牌奠定了首发全垒打。
然而拉尔斯与弗朗茨也坦诚反思，在开发 Model S 时特斯拉依然沿袭了传统汽车工业非常线性的老旧流水开发范式——即“先出设计造型，再交由工程团队结构研发，最后再去摸索如何把车制造出来”。当时两个人坐在一起勾勒图纸时，甚至完全不知道这台车未来到底会在哪座工厂落地产出。这一粗放的线性开发逻辑，直接催生了特斯拉后续对全新一代“解构式（Unboxed）”生产制造范式的大胆重构。
### [42:27](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2547s) 从弗里蒙特的“边飞边造”到 Model 3 的“产能地狱”
首席设计师 Franz von Holzhausen 与车辆工程副总裁 Lars Moravy 共同回顾了特斯拉艰难的制造演进之路。在拿下弗里蒙特工厂初期，团队的制造摸索就像“一边开着飞机，一边在空中装机翼和造引擎”。为了实现规模化扩张，特斯拉在宏图计划指导下研发了 Model 3，目标是打造一款尺寸更小、效率更高且价格更亲民的 Model S 替代版本，同时保留用户喜爱的一切核心体验。为了达成这一目标，特斯拉打破传统，首次将设计、工程与制造团队整合在一起协同推进。
然而在推进过程中，团队中途做出了全面转向高度自动化的决定。由于车辆最初是按照人工装配逻辑工程设计的，强行导入全自动化导致现场乱套，最终陷入了众所周知的“生产地狱”（Production Hell）。Lars Moravy 直言：“亲身经历过那段日子的人，身上都留着战斗伤疤。”团队在弗里蒙特工厂不得不拆除了大量失败的自动化设备，但也有部分方案最终跑通并沿用至今。尽管经历了这场磨难，Model 3 依旧成为史上最畅销的纯电动车，而基于其衍生的 Model Y 销量也即将实现超越。
> “automating something that we designed to be built manually is super hard.” —— Lars Moravy
- Model 3 的设计理念：更小、更高效、更实惠，但品质与用户喜好程度不逊于 Model S 与 Model X。
- 制造瓶颈成因：过早将原本按手工装配设计的车型强行进行全自动化改造，造成设计、工程与制造脱节。
- 最终成果：虽然初期经历了痛苦的试错和产线重构，Model 3 与 Model Y 仍先后奠定了全球最畅销电动车的地位。
---
### [44:21](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2661s) Cybertruck：由材料与制造约束倒逼出的全新设计语言
面对 Cybertruck，特斯拉彻底颠覆了以往的产品开发逻辑，直接从制造工艺和底层材料出发来反推整车形态。由于选用了超硬不锈钢材料，传统的冲压机床根本无法成型，研发团队被迫采用一种汽车工业历史上极少见到的线性折弯工艺。这种看似极端的工艺限制，反而让车辆外观具备了极强的视觉冲击力与辨识度，同时大幅精简了工厂占地面积——车身不再需要传统的大型冲压车间，甚至完全省去了喷漆车间。
Franz von Holzhausen 强调，这台车绝非展厅里的概念模型，而是真正量产在即的硬派皮卡。严苛的材料约束并没有牺牲整车的实用功能，Cybertruck 具备竞品皮卡拥有的所有核心性能。最重要的是，这款车确认将在年内正式交付。这种以工艺为起点的设计闭环，得益于特斯拉独特的组织架构：设计、工程、制造和自动化团队全在一个组织架构下向同一位负责人汇报，彻底杜绝了传统车企各部门之间的“踢皮球”，把限制转化为创新的源动力。
> “Forming full hard stainless steel isn't rocket science, but it sure isn't easy, and it limited the way we could do it.” —— Lars Moravy
- 材料特性：超硬不锈钢无法采用传统模具冲压，促成了全新的线性折弯成型工艺。
- 工厂精简：完全省略冲压件制造和整车喷漆流程，大幅降低工厂占地面积（Footprint）。
- 组织优势：设计、工程、制造及自动化团队高度集成，向单一负责人汇报，实现设计与工艺无缝融合。
- 交付节点：现场展示的均为可开动的真实工程样车，确认将于 2023 年内交付。
---
### [46:32](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=2792s) 拆解百年总装流程：从 Model Y 结构化电池到“解构式”平行制造
Lars Moravy 尖锐地指出了百年汽车装配线的荒谬之处：亨利·福特在 1922 年发明流水线，一个世纪以来汽车工业几乎没有本质变革。传统车厂将冲压件焊接成白车身、装上车门送去喷漆，喷好漆后再把车门拆下来，工人或机器人再把笨重的内饰件塞进狭窄的车门开孔，车身被反复抬起放下，最后才把在工厂里“旅游了一圈”的车门装回去。在这漫长过程中，绝大部分时间都是在挪动一个巨大的空壳子，却没有产生任何实际价值。即便丰田近期将 Model Y 的制造称赞为“工程艺术品”，特斯拉依然认为这远远不够，必须彻底革新制造逻辑。
在 Model Y 上，特斯拉通过一体化压铸（Giga Castings）消除了数百个零部件，并引入了结构化电池包（Structural Battery Pack），让电池直接充当座舱底板。座椅与内饰直接在电池包上方装配好，再整体从下方开口塞入车身，使总装线缩短了约 10%。以此为基础，特斯拉提出了更激进的“解构式”下一代制造理念：将整车完全拆解为独立模块进行平行（Parallel）组装。前后车体、侧围、车顶与底板独立加工并测试，只涂装必要外板，所有内饰件均从上方或下方无遮挡直接安装，最后像“拼积木”一样仅进行一次性的最终总装合体。
```mermaid
flowchart TD
accTitle: 下一代解构式平行总装流程
accDescr: 展示特斯拉下一代平台如何将车辆各模块独立平行总装并最终一次性合体的流程图
subgraph SUB_FRONT_REAR ["前后车体与动力总成"]
A1["前车身压铸件与底盘"]
A2["后车身压铸件与电驱"]
end
subgraph SUB_BODY_PANELS ["外围车身覆盖件（独立喷漆）"]
B1["左侧围模块"]
B2["右侧围模块"]
B3["车门及玻璃组件"]
end
subgraph SUB_STRUCTURAL_FLOOR ["结构化电池底板模块"]
C1["结构化电池包（充当座舱底板）"] --> C2["座椅及中控内饰直接安装在底板上"]
end
A1 & A2 --> D["模块化核心总装工位"]
B1 & B2 --> D
B3 --> D
C2 --> D
D --> E["整车单次完成最终合体 (One-Time Final Assembly)"]
```
> “If we're going to scale the way we want to do, we have to rethink manufacturing again. It's part of the master plan. We have to make another step change in cost.” —— Lars Moravy
- 传统制造死穴：车身喷漆后拆门、内饰从车窗/车门挤入、频繁升降搬运，大量工时浪费在搬运未完工的整车壳体上。
- 一体化压铸与结构化电池：Model Y 减少数百个零件，将座椅直接装在电池包顶面后整体托举入壳，缩减约 10% 最终总装线。
- 空间-时间效率（Space-Time Efficiency）：通过平行模块化装配，操作工密度提升 44%，时空效率提升 30%，机器人和人员无需进出狭窄车厢即可作业。
- 制造降本与缩减占地：下一代制造工艺可使工厂占地面积减少超过 40%，以更少的资本支出（CapEx）换取更高产出，推动整车制造成本降低最高达 50%。
### [51:58](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3118s) 动力总成环节开场与效率的全新定义
Lars Moravy 与 Franz von Holzhausen 结束了上一环节的整车设计制造理念分享。Lars 调侃道，外界总能听到他和埃隆·马斯克在财报电话会议上反复探讨制造约束与创新，而过往记录充分证明，特斯拉不仅能够顶着这些约束造出最好的车，甚至正是在约束中才诞生了最极致的工程突破。尽管全场翘首以盼，Lars 坦承大家还需等待一段时间才能看到下一代汽车的正式亮相，但他郑重保证：“我保证我们会一如既往地带来令人兴奋、引人注目且极具吸引力的汽车。”随后，Lars 引出特斯拉的另一大标志性特征——“速度”，并将舞台交给了特斯拉动力总成工程副总裁 Colin Campbell。
Colin Campbell 登台后直入主题，自豪地表示特斯拉制造着无论在燃油车还是电动车领域同等价位下速度最快的汽车。以在场展示的 Model S Plaid 为例，它拥有超过 1,000 马力的强悍动力，按功率重量比计算，其电动机的推力堪比喷气发动机。除了极致性能，特斯拉动力总成的核心竞争力还在于效率——在同等工况下，特斯拉的续航里程要比同级别竞争对手高出 25% 至 30%。然而，在特斯拉的工程哲学中，“效率”绝不仅限于让车辆消耗更少的电能，它涵盖了动力总成研发、制造、迭代和规模化扩张的全生命周期过程。
- Model S Plaid 输出功率突破 1,000 马力，电机功重比与喷气发动机相当。
- 特斯拉电动车在相同能量利用效率下，续航里程比同级竞品高出 25% 至 30%。
- Model 3 与 Model Y 自 2017 年发布以来持续迭代，其驱动总成（Drive Unit）在保持输出功率不变的前提下减重 20%。
- 重稀土材料用量相比最初减少了 25%。
- 制造该驱动单元的动力总成产线工厂面积缩小了 75%，建造成本降低了 65%，且产能产出丝毫不减。
---
### [54:21](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3261s) 全栈自研协同：功率电子与磁场仿真软件
Colin 深入剖析了特斯拉能够实现性能提升与成本骤降的底层原因：打破传统汽车工程割裂的外包模式，将整车设计与工厂产线制造紧密融合。在传统车企（例如购买高端德国电动车）中，逆变器或关键零部件的工程师往往隶属于外部供应商或承包商，而特斯拉拥有高度精干且跨学科的自研团队。电池化学家、机械工程师、制造工程师、供应链专家、自动化工程师和软件开发者汇聚在同一间办公室实时协作。谈及组织优势，Colin 强调：“我们通过将整车与制造工厂作为一个整体公司来共同设计，正是这一点让特斯拉与众不同。”这种组织架构让团队能在全局最优的前提下以惊人速度做决策。
在功率电子领域，车载充电机内部晶体管的封装直接关系到电能转换效率。特斯拉自主研发了专有定制封装，导热效率比市售通用元件翻倍，使核心的碳化硅（Silicon Carbide, SiC）晶圆尺寸得以大幅缩减。碳化硅性能优越但成本高昂且扩产困难，用量减少直接缓解了供应链约束。此外，协调众多功率晶体管高频精准开关是一项计算密集型任务，过去需要多达 4 颗微处理器；特斯拉自行研发了专用于高功率电子系统的微处理器，仅凭 1 颗芯片就完成了过去的全部工作，成本直接减半。自 2012 年 Model S 推出以来，特斯拉车载充电机的成本与重量均已降低了 50%。在软件方面，特斯拉自研了转子与定子旋转磁场仿真工具，其计算速度和精度远超商用软件，让工程师能在数百万种电机设计方案中快速筛选出成本、尺寸、重量和 NVH 噪音控制最优的解法。
```mermaid
flowchart LR
accTitle: 特斯拉跨学科全栈自研动力总成闭环
accDescr: 展示特斯拉跨领域团队在同一屋檐下协同，同时推进定制硬件、微处理器与仿真软件自研并直接反哺产线装备设计。
subgraph Collab["跨学科一体化团队"]
A["电芯化学 / 机械 / 制造"]
B["自动化 / 供应链 / 软件"]
end
subgraph CoreTech["高功率电子与核心软件自研"]
C["定制晶体管封装<br/>(散热翻倍 / 缩减SiC用量)"]
D["定制单微处理器<br/>(1颗替代原先4颗 / 成本减半)"]
E["自研磁场高精度仿真软件<br/>(数百万设计快速迭代)"]
end
subgraph ScaleProd["制造端协同落地"]
F["电机与产线装配设备协同设计"]
G["高集成驱动总成量产"]
end
Collab --> CoreTech
CoreTech --> ScaleProd
```
---
### [58:34](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3514s) 下一代动力总成平台：零稀土、降本 75% 碳化硅与千元级电驱
当动力总成设计与产线装备研发在同一屋檐下开展时，工程师在画出电机图纸的第一天就在考虑如何让自动化设备最容易地将其装配起来。Colin 宣布，这些深厚的技术积累将在特斯拉下一代车辆平台上带来颠覆性突破。下一代驱动系统将碳化硅晶体管用量大幅减少 75%，但汽车的性能与能量效率不打任何折扣。在电池兼容性上，新动力总成可无缝适配任意电芯化学体系（包括磷酸铁锂、高镍三元等），赋予供应链极高的采购韧性。
在成本与制造规模方面，特斯拉将下一代驱动单元的制造成本压缩至约 1,000 美元。Colin 断言：“我们认为没有任何其他汽车制造商能够接近这个数字。”与此同时，下一代动力总成工厂的占地面积在保持现有工厂同等产能的前提下再度缩减 50%，极大缩短了建厂与投产周期。针对清洁能源转型中需求激增、开采伴随环境与健康风险的稀土资源，Colin 带来了一项重磅技术突破：特斯拉下一代永磁电机将实现 100% 零稀土材料使用。这些技术创新共同构成了“宏图计划第三篇章”（Master Plan Part 3）的核心基石，通过使用更少受限资源、建设更紧凑高效的工厂，支撑起千万级年产能的大规模扩张。
- 下一代动力总成碳化硅（SiC）用量减少 75%，性能与效率零妥协。
- 动力系统全面兼容所有电芯化学体系，保障供应链柔性。
- 驱动单元整机成本降至约 1,000 美元。
- 下一代动力总成工厂在产能持平的情况下，厂房占地面积减少 50%。
- 下一代永磁同步驱动电机完全摒弃稀土元素，实现 0 稀土用量。
---
### [01:01:16](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3676s) 低压电子电气架构的演化挑战
Colin Campbell 在掌声中结束了演讲，并邀请负责电子与软件工程的同事 Pete Bannon 与 David Lau 登台分享电子架构与底层软件体系。
Pete Bannon 走上台向全场问好，大屏幕上展示出 2012 年初代特斯拉 Model S 的低压电气系统图。Pete 回顾道，当年的 Model S 很大程度上依赖供应商提供的现成控制器，再由特斯拉在车端进行系统集成。这套架构包含 300 多个低压电气设备，涵盖了极其繁复的门类：从手套箱照明灯等平凡的小部件，到复杂的信息娱乐车载电脑，再到安全攸关的关键系统，如安全气囊、线控转向（Steer-by-wire）以及线控刹车（Brake-by-wire）。然而，这种早期架构采用的是传统线束，每一根线缆都需要人工切断到特定长度、端子压接并手工插入接插件。Pete 直言这种传统工艺高度依赖人工，“这是一个繁琐、容易出错且难以规模化扩展的过程”，而特斯拉未来的战略目标正是大幅缩减线束的尺寸与复杂度，实现底层架构的自动化与革命性简化。
### [01:02:14](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3734s) 线束复杂性痛点与控制器自研演进之路
传统汽车制造中，线束对自动化制造构成了极大阻碍。错综复杂的线束在整车早期开发与系统调试（bring up）阶段会带来超乎寻常的复杂度：当整个系统无法正常运转时，工程师很难判断问题究竟出在软件、控制器、处理器，还是这团线网连接的某一个终端上，导致必须全面排查、同时调试所有节点。Pete Bannon 详细拆解了特斯拉自 Model S 以来在线束与控制器上的设计变革。
从 Model S 跨越到 Model 3，特斯拉开始将多个控制器合并，大幅简化了布线设计。仅从 Model S 到 Model 3 的迭代中，整车线束就成功减重 17 千克。Pete 透露了工程团队内部的一个有趣典故：车辆工程副总裁 Lars Moravy 曾立下规矩，只要工程师能帮车辆减掉 1 千克的重量，他就会亲自往对方桌上送一瓶对方最喜欢的烈酒；因此，单单 Model 3 这次减重就让 Lars“破费不少”。随后，Model 3 的控制器经过优化被应用在 Model Y 上；当车辆引入热泵系统时，这些升级版控制器又被反哺回 Model 3，实现了跨车型通用，大幅简化了供应链体系。这一系列架构还在 2021 年版 Model S 和 Model X 上继续迭代演进。
在垂直整合的深度上，特斯拉不仅止步于部分整合。Pete Bannon 宣布：“在 Cybertruck 上，我们自己设计了车内 85% 的控制器。而在下一代平台上，我们将彻底完成这项工作，自己设计 100% 的控制器，以在组件级别实现对设计和供应链的完全掌控。”（“For Cybertruck, we are now designing 85% of the controllers in the car. For the next-gen platform, we're going to finish the job and be designing 100% of the controllers to give us full control of the design and the supply chain at the component level.”）这种在组件层面的自主权，帮助特斯拉在过去几年的全球供应链危机中从容应对、持续产车。伴随硬件完全自主掌控的是软件控制权，这让团队能够开发出许多在硬件设计之初甚至“连做梦都没想过”的功能，例如哨兵模式（Sentry Mode）就是软件团队在最后关头利用现有硬件巧妙实现的最爱功能之一。
- 从 Model S 到 Model 3，整车线束总重量削减了 17 千克。
- 控制器自研比例：Cybertruck 达到 85%，下一代车型平台将达到 100%。
```mermaid
flowchart TD
accTitle: Controller Integration Evolution
accDescr: Evolution of Tesla controller design and in-house percentage from Model S to Next-Gen Platform.
subgraph Iteration["控制器设计演进与跨车型复用"]
S["Model S (分立控制器)"] -->|"整合合并减重 17kg"| M3["Model 3 控制器"]
M3 -->|"优化增强"| MY["Model Y 控制器"]
MY -->|"随热泵系统反哺"| M3_Heat["Model 3/Y 共享控制器"]
M3_Heat -->|"再次迭代增强"| S_X_2021["2021款 Model S / X"]
end
subgraph InHouse["自研比例跃升"]
CT["Cybertruck: 85% 控制器自研"]
NextGen["下一代平台: 100% 控制器自研"]
end
Iteration --> CT --> NextGen
```
---
### [01:04:33](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=3873s) 软件定义硬件：电子熔丝、锂离子低压电池与元器件极致优化
2017 年随 Model 3 问世，特斯拉彻底移除了传统的机械式继电器和保险丝，全盘采用电子熔丝（eFuse）。电子熔丝利用固态晶体管取代机械运动部件，将整个配电系统的主导权交给了软件。软件由此能够实现细粒度控制，在极端工况下进行动态切载（load shedding），还能够在瞬态故障发生时通过软件尝试复位重试，并对整车电气状态进行毫秒级的细致监控。
David Lau 阐述了特斯拉“软件控制硬件”的核心哲学。David Lau 指出：“当我们团队思考这一点时，软件控制硬件从根本上说就是能够避开硬件原本必须在某一种属性与另一种属性之间所做的静态取舍。通过软件，我们能够将智能、情境感知和情境特定行为注入到一块原本必须针对单一场景进行妥协优化的硬件中。我们由此获得了方方面面的全面提升。”（“When I think about that in my team, software-controlled hardware is fundamentally about being able to sidestep what otherwise would have been static trade-offs between one attribute or another in a piece of hardware. We are able with software to instill intelligence, context awareness, and context-specific behavior into what otherwise would have been a piece of hardware that had to get optimized for one type of scenario. We get more of everything.”）
低压供电和零配件领域的微观改进同样在持续进行：
- **低压锂电池与免工具接头**：2022 年特斯拉完成了从传统铅酸小电瓶向 16V 锂离子低压电池的全面过渡，并配以全新的免工具接头。该电池寿命预计贯穿整车生命周期，彻底解决了车主早晨醒来遭遇“电瓶没电抛锚”的行业老毛病。免工具接头简化了售后维护，并在软件中内置了安装验证逻辑，服务结束时软件会自动核对插头是否装配到位，杜绝二次返修。
- **15 英寸中控屏持续降本提效**：自 2017 年在 Model 3 搭载以来，该屏幕成本降低了 24%，重量减轻了 12%，功耗下降了 33%；与此同时，屏幕亮度反而提升了 50%，色彩准确度也得到进一步提升。Pete Bannon 将特斯拉的效率优化比作“削胡萝卜”：每次削下一小片看似无足轻重，但日积月累就会汇聚成惊人的质变。
---
### [01:07:03](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4023s) 48V 电压架构与以太网区域控制器网络
汽车工业沿用 12V 低压电气系统已有整整 60 年。但随着车内功耗急剧上升，12V 系统需要通过极其粗大的导线来传输超过 200 安培的大电流，导致重量与成本居高不下。特斯拉宣布：Cybertruck 及未来所有全新平台将全面转向 48V 电气架构。
从物理学原理来看，电压提升至 48V 意味着在相同功率下电流缩小为原来的四分之一（4x reduction）。根据焦耳定律，线束上的功率损失与电流的平方成正比（$P = I^2R$），因此 4 倍的电流降低直接带来配电损耗降低至原来的十六分之一（16x reduction）。损耗的骤降带来了链式轻量化红利：导线线径可大幅缩减，eFuse 和控制器尺寸变小，原本为了散发大电流热量而设置的沉重散热片得以大幅缩小甚至彻底移除。Pete Bannon 强调，48V 是特斯拉低压设计的未来，也终将成为整个汽车工业的必经之路，特斯拉公开欢迎并鼓励其他车企及整个供应商网络共同拥抱这一进化。
配合 48V 架构的还有网络通信拓扑的颠覆。以往中心化的控制架构需要电线纵贯全车，导致线束数量暴增。Cybertruck 转向了“区域控制器（Local Controller）+ 车载以太网（Ethernet）”架构：
- 传感器与执行器等所有终端只需要连接至物理距离最近的区域控制器。
- 区域控制器在本地将电信号直接封装为网络数据包，通过高速以太网总线传输至车内目标节点。
- 该网络具备极高的可靠性、超低延迟和超低抖动，彻底消除了 Cybertruck 上绝大部分横跨车身的线缆；而在下一代车型平台上，特斯拉将把所有横跨车身的杂乱连线彻底消灭。
- 相比传统分散在车身各处的独立 CAN 总线硬件死结，统一的以太网网络支持动态在线重组通信逻辑，而且工程人员只需通过一个单一物理接口即可纵览整车所有节点，让排错调试效率发生质的飞跃。
---
### [01:09:50](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4190s) 软件驱动：全车 OTA、车队遥测与数据驱动的设计决策
软件副总裁 David Lau 回顾道，自己于 2012 年加入特斯拉，早在 Model S 面世的极早期阶段，特斯拉就具备了对整车所有控制器进行全面 OTA（空中下载技术）空中升级的能力；而当时传统汽车行业才刚刚起步，且普遍仅能升级车载信息娱乐系统的车机固件。
伴随全车 OTA 的，是特斯拉高度精密且匿名的车辆数据遥测（Telemetry）系统。这套数据底座能实时洞察庞大车队的运行状态、客户的真实驾驶使用习惯，以及硬件在真实复杂道路工况下的物理表现。OTA 迭代能力与来自真实车队的海量数据洞察相结合，构成了特斯拉极其迅速的产品迭代闭环，并在软硬件设计层面为团队提供了无数关键决策依据。
在分享具体实例时，Pete Bannon 笑着提到，工程团队曾利用数据系统持续监测车主对全景天窗（sunroof）的开启频率，结果发现绝大多数用户在提车后几乎从未使用过打开天窗的功能，这让团队毫不犹豫地做出了直接取消可开启天窗的工程决策。此外，车队收集的真实碰撞传感器数据，也为特斯拉碰撞安全系统的定制化迭代提供了无法从实验室模拟中获得的精准现实依据。
### [01:11:09](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4269s) 现实碰撞数据、仿真模拟与庞大的车队数据闭环
特斯拉在车辆安全性上的研发已远远超出法规和消费者碰撞测试机构所规定的测试范围。通过从实际道路行驶中收集海量真实碰撞数据，特斯拉能够在每次对车辆设计或控制安全气囊等约束系统（restraint systems）的软件进行调整时，都在仿真环境中完全重现这些碰撞事故。在大屏幕展示的图表上，灰色圆点代表全球监管与消费者评价机构规定的特定碰撞测试工况，而周围和重叠覆盖的丰富彩色数据点，则代表特斯拉车队在现实世界中所经历的各种真实碰撞场景——这些才是特斯拉研发所面向并进行验证的基准。去年，特斯拉正是基于这些真实世界的数据，优化了安全带预紧算法，并通过 OTA 空中升级推送到所有在役车辆中，有效降低了真实事故中的乘员伤害。
依托全球庞大的在役车队，特斯拉正以惊人的速度从每日接收的匿名数据中获取新洞见。例如，通过分析车队的日常驾驶和充电行为模式，团队可以为下一代车型优化匹配最佳的电池包容量。这种由真实数据驱动的决策方式，贯穿于特斯拉的产品开发和生产制造等方方面面。
- **每日行驶里程**：车队每天累计行驶 1.23 亿英里（123 million miles）。
- **每日充电频次**：车队每天完成 190 万次充电会话（1.9 million charge sessions）。
- **OTA 安全带更新**：基于真实碰撞数据迭代安全带预紧算法，通过 OTA 全量下发以降低乘员受伤风险。
- **下一代电池尺寸设计**：依托大规模日常驾驶与充电特征分析，科学测算并确定下一代车型的最佳电池容量。
---
### [01:12:45](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4365s) 软件纵向整合与跨组件联动：以路面颠簸预测调节为例
传统汽车供应链的软件模式高度割裂，不同控制器由不同一级供应商（Tier 1）提供，一级供应商将软件外包给二级供应商（Tier 2），二级供应商再分包给三级供应商（Tier 3），导致跨组件的软件修改在正式动工前往往需要数月的商务与流程协调。相比之下，特斯拉在软件层面坚持彻底的纵向整合（Vertical Integration）。对于特斯拉软件团队而言，系统并不仅指孤立的车载计算机，而是包含了车端、后端服务器端应用与基础设施，以及连接整个车队的闭环反馈生态。
```mermaid
flowchart TD
accTitle: 颠簸路面自适应空气悬架工作流程
accDescr: 展示从车端传感器感知到云端聚合，再到前向预测与执行调整的跨组件全流程
subgraph Sense ["1. 数据感知与定位"]
RCM["约束控制模块 (RCM)<br/>IMU 惯性传感器感知颠簸"]
AP["Autopilot 计算机<br/>GPS 定位与局部化解算"]
end
subgraph Cloud ["2. 云端地图聚合"]
NavServer["云端导航服务器<br/>聚合匿名遥测数据并标注路面颠簸度"]
end
subgraph Actuate ["3. 车端决策与执行"]
NavEngine["车端车载导航引擎<br/>前瞻预测前方行进路线路况"]
Suspension["空气悬架控制器<br/>综合路况决策并提前调高底盘高度"]
end
RCM --> NavServer
AP --> NavServer
NavServer --> NavEngine
NavEngine --> Suspension
```
以近期在 Model S 和 Model X 上推出的“颠簸路面自适应调节悬架”功能为例，车辆在行驶经过坑洼颠簸路段前，能够自动且具有预测性地抬高空气悬架以保证乘坐舒适性。该功能高度依赖车端多个核心模块与云端服务器的无缝协同：
1. **约束控制模块（Restraint Control Module, RCM）**：利用内置的惯性测量单元（IMU）实时检测路面的实际颠簸程度。
2. **Autopilot 自动辅助驾驶计算机**：利用 GPS 模块和本地化定位系统，精确锁定车辆在全球地图中的地理坐标。
3. **云端导航服务器**：汇总并聚合来自整个车队的匿名遥测数据，向高精导航地图上动态标注实时限速、车道拓扑以及新增的路面粗糙度信息。
4. **车端车载导航引擎**：实时检索车辆规划路线的前方路况，预判车辆是否即将驶入颠簸路段。
5. **空气悬架控制器**：综合上述全部前瞻信息，实时、平滑地决策并主动调整空气悬架高度以提升驾乘品质。
在这一功能全量推送给用户前，特斯拉向整个车队静默推送了原型算法，让其在后台影子模式（shadow mode）中被动运行，记录并在满足触发条件时匿名回传数据，从而使团队能够充分掌握该算法在各种极端环境与场景下的实际表现。该后台验证能力对紧急制动（AEB）等安全攸关功能同样至关重要，早在 2017 年推出 Model 3 时，早期车身稳定控制（ESP）算法也是通过这种影子测试完成了极其高置信度的验证。
David Lau 强调了特斯拉软硬件全栈自研与车云一体视角的独特性：
> “But in the software team, we think about the system as including the vehicle as well as all of our back-end server-side applications and infrastructure and the resultant feedback loop from the entire fleet that informs all the decisions we make as an engineering team.”
---
### [01:15:55](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4555s) 软件思想重构制造与售后：产线自动化测试与智能诊断
除了提升面向用户的产品功能，特斯拉将软件工程中“尽早测试、频繁测试（test early and test often）”的理念深度融入内部生产制造与售后体系中。在早年生产 Model S 时，人工装配容易出现线束漏插、接插件未插紧、从错误料箱取料等隐患。因此在 Model 3 及后续车型的产线设计中，整车装配线的每个工序都深度植入了软件自动化自检能力。
当装配工人在流水线上将任一零部件线束插入车辆时，车载中央电脑会立即识别该连接事件，自动比对该零部件型号是否与工单目标车型一致，必要时自动刷写固件更新，校准传感器与配置，并对该部件及关联电气链路运行一系列自动化测试。一旦发现纯软件手段无法修复的物理异常，系统会立即在车内中控屏上弹出告警，并同步将故障推送到后端中央指挥调度系统，通知技术人员在车辆进入下道工序前立即介入修复，防止带有缺陷的车辆在流水线上继续组装。
此外，在售后服务端，特斯拉通过车载诊断系统（Onboard Diagnostics）、客户 App 报修文本的自然语言处理（NLP）分析以及自研内部诊断工具链，实现了超过三分之一（33%以上）的服务预约自动完成故障诊断、工时预约及所需备件自动预订。配合使用自研控制器开发的预装配测试工装，分总成在合装到底盘之前就已经过独立验证，为下一代平台（Lars Moravy 主导的“开箱工艺（Unboxed Process）”）奠定了坚实的技术基础。
- **产线动态侦测机制**：零件一旦接入，中央电脑自动校验物料匹配度、刷写固件并运行电气闭环自检。
- **装配缺陷前置拦截**：发现物理故障即刻锁止告警，派工检修，杜绝隐性缺陷随装配线流转。
- **售后智能化指标**：超过 33% 的售后报修诉求通过 NLP 和远程诊断工具链，实现无人介入下的自动诊断、排期与备件下单。
---
### [01:18:06](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4686s) 面向无人驾驶出租车（Robotaxi）车队管理的基础软件布局
随着全自动驾驶技术的演进，特斯拉已经在底层平台能力上为未来管理大规模自动化无人出租车网络（Robotaxi Fleet）进行了数年的基础设施铺垫。这些能力包括跨设备同步、基于权限的数字车钥匙、集中化车队运营调度与端到端高强度加密。
许多底层架构已经在现有的量产功能中接受了检验：
- **手机钥匙分享（2021 年上线）**：车主可以通过短信或电子邮件直接向任何人分享车辆的数字钥匙控制权限。
- **云端个人偏好同步（Cloud Profile Sync，去年上线）**：座椅记忆、方向盘位置、后视镜偏好、车机多媒体收藏夹以及常用导航收藏夹实现账号级云端同步，用户登上账号名下的任意特斯拉车辆均可即刻还原个人设置。
- **内部车队运营 App**：工程、生产、交付与物流团队通过内部专用应用，可以全局查看、精准定位并直接开走所属厂区/场站内的任何一辆特斯拉所有权资产。
- **售后代步车无缝上车**：在部分门店试点中，当车主将车送修并领取售后代步车时，代步车权限会自动绑定到其手机 App 中；配合云端 Profile，代步车在感应解锁瞬间便自动调整为车主熟悉的座椅与设置，实现与开私家车完全一致的体验。
这一切指令流与控制链路均建立在端到端加密（End-to-end Encryption）和密码学签名（Cryptographically Signed Commands）基础之上，确保车主隐私数据的绝对安全，同时保障在役车队仅严格响应经过双向鉴权的合法指令。
在结束软件架构与底层系统平台的串讲后，演讲者总结了特斯拉在效率、降本与执行速度上的后台核心技术，并正式将讲台交接给自动驾驶负责人 Ashok Elluswamy，开启 Full Self-Driving（完全自动驾驶）核心算法的深度分享：
> “And so now, to tell us about the latest in full self-driving, I'd like to introduce Ashok.”
### [01:20:33](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4833s) 自动驾驶与可持续未来的关联：规模化 AI 的三大基石
特斯拉 Autopilot 与完全自动驾驶（Full Self-Driving, FSD）负责人 Ashok Elluswamy 登台演讲。他回顾道，自己早于 2014 年就加入特斯拉，至今已在这项难题上深耕近 9 年。针对外界“自动驾驶与可持续未来规划到底有什么关系”的疑问，Ashok 强调自动驾驶绝非边缘功能，而是实现宏伟蓝图不可或缺的一环：目前私家车在绝大多数时间内都闲置在停车场；一旦真正实现完全自动驾驶，车辆便可以在闲暇时服务其他用户，从而成倍提高单车利用率。这种利用效率的飞跃，将从根本上削减将整车制造规模推向极端庞大水平的迫切需求。
然而，打造一套可规模化部署的自动驾驶系统绝非易事。Ashok 指出：“我认为构建一个可扩展的自动驾驶系统，是目前世界上最难的现实世界人工智能难题之一。”（“Building a scalable self-driving system is, I think, one of the hardest real-world AI problems out there right now.”）为此，特斯拉将该难题的攻坚路径清晰拆解为必须全部做对的三大支柱：
1. **系统架构（AI Architecture）**
2. **数据闭环（Data）**
3. **计算能力（Compute）**
```mermaid
flowchart LR
accTitle: 可规模化自动驾驶系统的三大核心支柱
accDescr: 展示特斯拉解决规模化自动驾驶所依赖的架构、数据与算力三大支撑
subgraph Pillars ["规模化自动驾驶的核心支柱"]
A["先进 AI 架构<br/>(Transformer / 占用网络)"]
B["车队海量数据<br/>(自动标注 / 数据引擎)"]
C["庞大计算集群<br/>(车载芯片 / 1.4万GPU / Dojo)"]
end
Pillars --> D["实现完全自动驾驶 (FSD)"]
D --> E["车辆利用率大幅提升"]
E --> F["降低过度制造需求，助力可持续未来"]
```
---
### [01:22:04](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=4924s) 架构革新：从单帧拼接转向多相机端到端视频网络与 AI 规划
在架构演进上，特斯拉全面押注机器学习与神经网络。早期系统依赖单摄像头、单帧图像神经网络生成初步检测结果，再由下游复杂的后处理代码强行拼接给规划模块，这种架构非常脆弱且极易失效。过去几年间，特斯拉将整个底层软件栈彻底重构为多相机端到端视频神经网络。该网络能够实时接收全车 8 颗摄像头的原生视频流，并直接在统一的三维向量空间（Unified 3D Output Space）中完成对障碍物存在性、运动状态、车道线、道路走向及交通信号灯的联合预测。现场展示的占用网络（Occupancy Network）画面显示，系统能够精准捕捉旁边一辆大货车剧烈的摇摆与偏移姿态，从而为下游路径规划提供关键输入，提前规避潜在碰撞。
在车道拓扑连接这类复杂的非线性关系建模上，传统计算机视觉算法往往捉襟见肘。特斯拉直接跳出传统 CV 框架，跨界融合了大型语言模型（LLM）与强化学习前沿技术，采用类似于 ChatGPT 的 Transformer、注意力机制（Attention Modules）以及自回归 Token 建模，消除了所有生硬的规则后处理。与此同时，规划系统（Planner）也摒弃了旧时代的几何规划逻辑，全面接入神经网络。在复杂的无保护左转十字路口，自车需要同时避让对向直行车流、横穿行人并遵守路权优先级，如果采用传统算法计算，推演每一种组合大约耗时 10 毫秒，数以千计的潜在组合将使算力瞬间崩溃。特斯拉通过将规划逻辑模型化，把所有推演打包进严苛的 50 毫秒算力预算内，从而在车端流畅实现实时运算与毫秒级决策。
---
### [01:24:54](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5094s) 数据飞轮：全自动标注流水线与 14,000 例幽灵刹车攻坚
自动驾驶的第二大基石是数据。特斯拉拥有其他竞争对手难以企及的车队规模优势，能够精准按需调取边缘样本。然而原始视频本身并无监督信号，若完全依赖人工逐帧标注，其数据产出速率根本无法喂饱多相机视频大模型。为此，特斯拉打造了一套全自动标注（Auto-labeling）流水线：车载传感器将特定片段回传至数据中心后，算法集群通过多视角几何与神经重构，将来自不同车辆、不同时段的视频片段融合为高度精细的真实世界统一 3D 重建场景，无论是车道标线、路沿、人行横道还是路面印刷文字都能被精准复原。基于这一 3D 底座，仿真平台能够随意合成狂风骤雨、极端光照以及极端交通参与者交互，以近乎无限的合成数据穷举测试极端边缘用例（Corner Cases）。
```mermaid
flowchart TD
accTitle: 特斯拉自动驾驶数据引擎闭环流程
accDescr: 展示识别问题、车队数据挖掘、自动标注、模型重训到部署更新的完整数据闭环
A["识别系统瓶颈与边缘案例<br/>(例如：停放车辆引发幽灵刹车)"] --> B["向全球车队下发定向挖掘触发器"]
B --> C["回传特定场景多视角视频片段"]
C --> D["数据中心集群 3D 重建与自动标注"]
D --> E["合成仿真强化与训练集扩充<br/>(新增14,000条样本)"]
E --> F["端到端大模型重新训练"]
F --> G["OTA 固件部署回车载系统验证"]
G --> A
```
为了直观证明数据驱动的威力，Ashok 剖析了一个曾让系统产生误判的“幽灵刹车”典型案例：此前系统曾将路边一辆完全停稳熄火的车辆误判为即将侵入自车车道，从而引发防御性急刹。针对这一问题，团队并未编写打补丁式的逻辑代码，而是向全球车队下发定向抓取指令，从车队中精准挖掘出 14,000 段类似的静止停靠车辆视频并纳入训练集。完成重训后的神经网络学会了精准判断驾驶位是否有人，从而彻底消除了对真实驻车车辆的误减速。Ashok 总结道，通过反复运转“发现边缘案例—车队挖掘数据—自动生成标签—重训更新模型—固件推回车队”的数据引擎闭环，系统在各项细分任务上的性能曲线呈现出持续、稳定的线性提升。
- **标注吞吐突破**：摆脱人工标注产能瓶颈，利用数据中心大算力实现多车多轨迹时空拼接与全自动 3D 几何标注。
- **幽灵刹车实战**：仅针对停放车辆误判单一问题，一次性定向注入 14,000 个真实车队视频片段重训解决。
- **仿真极限工况**：基于 3D 真实重建场景无缝合成恶劣气候、眩光反光及物理碰撞边缘轨迹。
---
### [01:28:10](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5290s) 计算集群与 FSD 现状：40 万用户、超国家均值 5-6 倍的安全性
在计算这一支柱上，训练超大规模视觉与时序模型以及跑通自动标注均需要巨额算力支撑。在车端，特斯拉自研车载计算机具备高达 150 TOPS 的实时推理能力；在云端后台，特斯拉部署了包含 14,000 块 GPU 的超大规模超级计算集群。在算力配比上，大约 30% 的 GPU 算力专门分配给离线全自动标注流水线，其余 70% 则全部投入神经网络的重训与迭代。此外，云端视频缓存已达 30 PB，并正迅速扩容至 200 PB 规模，仅占用网络单个模块就吞吐了多达 14 亿帧图像进行训练。随着自研训练专用超级计算机 Dojo 的全面介入，这一数据与算力规模将再次迎来指数级爆发。
在用户落地进展方面，特斯拉已向美国和加拿大所有购买了 FSD 的车主全量推送了 FSD Beta 软件，覆盖约 40 万名车主。车辆不仅能够自主完成无保护转弯、识别红绿灯停启、处理复杂交叉博弈，还能顺畅导航至目的地。更关键的是真实道路安全数据：Ashok 指出，根据后台统计分析，“我们观察到，使用 FSD 的车主，其安全性已经达到美国国家平均水平的 5 到 6 倍。”（“We have observed that people who use FSD are already five to six x safer than the US national average.”）
随后，Elon Musk 重新登台并播放了人形机器人 Optimus 的最新研发进展视频。画面中不仅展现了实验室里密布的机械臂与机械腿测试台架，还展示了人形机器人自主平稳行走的最新状态。Musk 边看画面边打趣实验室里散落的人形四肢略显怪异，随后严肃指出：回顾上一届特斯拉 AI Day 时，这一版本的 Optimus 根本还无法独立站立行走，而在极短时间内其机械本体与动力学控制便实现了突飞猛进的迭代——尽管它目前显然还无法像特技演员那样做出跑酷等高难度动作，但整体步态控制与环境探索已取得突破性进展。
### [01:31:18](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5478s) Optimus 机器人：现实世界 AI 与定制执行器
在展示了多台 Optimus 原型机后，Elon Musk 强调了特斯拉在人形机器人领域拥有的不可替代优势。他指出，特斯拉拥有行业内最顶尖的现实世界人工智能（real-world AI）：自动驾驶汽车本质上就是“装在轮子上的机器人”，而 Optimus 则是“装在双腿上的机器人”。推动汽车行驶的计算平台和软件可以直接无缝移植到 Optimus 身上。Musk 强调，如果每个单独动作都需要逐一编程，人形机器人就失去了实用价值；它必须能够完全自主地四处行走、完成任务，用户只需直观地向它展示该做什么，或者直接通过口令下达指令。
除了 AI 软件层，特斯拉的核心制造与工程能力同样是关键壁垒。Musk 透露，团队在研发初期惊讶地发现，全球市面上虽然有海量的电机和减速箱现货，但竟然没有一样能够直接用在人形机器人上，“真的是一件可用的都没有”（literally none）。因此，Optimus 内部的所有执行器（actuator）——包括电机、齿轮箱、功率电子器件以及电池包——全部由特斯拉内部自研。为 Model S Plaid 研发突破性驱动电机的原班工程团队操刀设计了这些机器人执行器，使特斯拉有能力以远超行业对手的速度将量产级产品推向市场。
谈及人形机器人的长远未来，Musk 提出了一些颇具科幻色彩但影响深远的构想。他预测未来人形机器人与人类的比例可能会超过 1:1，机器人将广泛渗透到家庭看护和工业生产中。如果经济规模等于“人均产出乘以人口总量”，而在机器人的支持下产出大幅提升且劳动力数量不再受限，那么传统经济学的边界将被彻底改写。Musk 直言：“这可能是我们在特斯拉所做的业务中最不被理解、也最被低估的部分，但长远来看，它的价值可能会显著超过汽车业务。”
- **AI 架构迁移**：自动驾驶的视觉感知、规划决策与车端 FSD 计算平台直接复用到人形机器人。
- **全栈自研执行器**：市面现成零件无法满足高功率密度与动态响应需求，电机、齿轮箱、功率电子均由 Model S Plaid 动力总成团队自主定制。
- **人机比例预测**：Musk 预估人形机器人对人类的比例可能突破 1:1，引发生产力与经济体量的本质飞跃。
> “The thing that's I think Tesla brings to the table that others don't have is that we have the real-world AI.” —— Elon Musk
> “We were quite surprised to find how little was available off the shelf... we found none of them were useful in a humanoid robot, literally none.” —— Elon Musk
> “It's probably the least understood or appreciated part of what we're doing at Tesla, but will probably be worth significantly more than the car side of things long term.” —— Elon Musk
---
### [01:35:36](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5736s) 特斯拉充电网络体系与预制化快速部署
负责特斯拉全球充电基础设施团队的 Rebecca Tinucci 登台发表演讲。她明确指出，特斯拉从第一天起就深知，出色的充电体验是推动电动汽车普及的绝对关键（linchpin）。特斯拉对充电采取了全面的系统化考量（holistic approach），覆盖了从公寓、独立住宅的夜间家充，到日常通勤及长途自驾的各类场景。2022 年，特斯拉通过旗下各类充电方式累计提供了 9 TWh（太瓦时）的电量，其中超过 50% 来自便利的交流家充桩；而在公共充电领域，车主可使用全球超过 80,000 个特斯拉充电桩，其中包括 40,000 个标志性的超级充电桩（Superchargers）。
特斯拉历经 10 年自主铺设充电网络，在行业几乎无人跟进的时期积累了巨大的工程与运营优势。Tinucci 表示，特斯拉拥有全行业最低的部署成本，其硬件和建造成本通常比行业替代方案低 20% 甚至 70%。这种成本控制源自端到端的垂直整合：特斯拉自主研发制造所有充电设备，不同产品线之间高度共享核心零部件，并且在超充端实现自主安装与运营。
为了进一步颠覆传统基建施工流程，特斯拉将先进制造理念引入了充电站建设。位于纽约超级工厂（Giga New York）的产线会将 4 个超充桩预先固定在混凝土底座上并完成内部布线，形成一体化预制单元（pre-building four-post supercharger units）。整套预制单元由卡车直接运至现场，再用吊车精准吊装就位。这项创新不仅直接削减了 15% 的部署成本，还能将站点的实际现场施工周期缩短至短短数天。
```mermaid
flowchart LR
accTitle: 超级充电站预制化模块部署流程
accDescr: 展示从纽约工厂预组装到现场吊装的四桩一体化快速建站流程
subgraph Factory["纽约超级工厂 (Giga New York)"]
A["4联装超充桩模块预装"] --> B["集成式底座与电气总装"]
end
subgraph Logistics["运输与吊装"]
B --> C["平板卡车整装运输至站点"]
C --> D["现场吊装精准就位"]
end
subgraph Outcome["降本增效成果"]
D --> E["削减 15% 部署成本"]
D --> F["安装周期缩短至数天"]
end
```
- **2022 年供电规模**：全网络供电 9 TWh，其中 50% 以上由家用交流充电承载。
- **全球桩网规模**：公共充电接入点突破 80,000 个，其中直流超充桩达 40,000 个。
- **建设成本优势**：全栈垂直整合使建设成本比竞品低 20% 至 70%；纽约工厂预制 4 桩一体单元进一步降本 15%。
> “And at Tesla charging, we have understood since day one that a great charging experience is the linchpin to electric vehicle adoption.” —— Rebecca Tinucci
> “We pre-building four-post supercharger units at our factory in New York. We load them on a truck, we truck them to site, and then we crane them into position. This method saves us 15% on our deployment costs and we can install a site in a matter of days.” —— Rebecca Tinucci
---
### [01:38:02](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=5882s) 行程规划器智能化调度与超充网络全面开放
在站点建成后，超充网络的高效运营成为拉开差距的另一重关键。过去几年中，特斯拉成功将每千瓦时（kWh）的运营成本降低了 40%，其核心杠杆是提升站点利用率（site utilization）。通过在每个充电桩上提供更多充电频次并输出更多电量，固定成本被有效摊薄。然而，盲目拉高利用率极易引发排队拥堵，进而摧毁用户体验。特斯拉凭借独有的车机导航系统——行程规划器（Trip Planner）完美化解了这一矛盾。
其他主机厂可能掌握车辆端数据，第三方充电桩运营企业掌握场站端数据，但只有特斯拉同时掌握车辆动态与站点负荷的全部实时数据。行程规划器就像充当全球充电体系的“空中交通管制员”（air traffic controller），智能地将沿途车辆引导至空闲站点，避开高负荷站点。这一软硬件闭环带来了极具说服力的成果：在整体利用率提升 30% 的同时，车主的平均排队等待时间直接缩短了一半，平均单次超充停靠时间也缩减了 30%。未来，Trip Planner 将从“根据当前在站车辆进行静态反馈”，进阶为“基于正在前往该站点的车辆路线提前预测站位占用情况”。
最后，Tinucci 谈及特斯拉支持全球车队完全电动化的宏大蓝图。据测算，全球电动车队每年将需要 9 PWh（拍瓦时，即 90 亿度电）的电能支持。为了让所有电动车都能享受到优质、可靠的补能体系，特斯拉正在全球范围内稳步开放超级充电网络：欧洲超过 50% 的超充站已对非特斯拉车辆开放；亚太地区的首批试点已在澳大利亚落地；就在本次活动前一天，美国本土的前 10 个超级充电站也正式向非特斯拉车型开放。为了抹平充电接口协议的物理差异，特斯拉在特定区域的超充桩上加装了自主研发的“Magic Dock”转换装置，非特斯拉车主只需下载并登录 Tesla App，即可一键解锁充电枪，无缝开启高功率充电。
- **降本与提效双赢**：超充单度电运营成本降低 40%；站点利用率提高 30% 的同时，车主等待时间减少 50%，平均停靠充电时间压缩 30%。
- **全球用电远景**：全球汽车全面电动化后，行业年充电需求预计达到 9 PWh。
- **网络对外开放进展**：欧洲超 50% 超充桩向非特斯拉开放，亚太区拓展至澳大利亚，北美首批 10 个站点刚完成开放；通过集成 Magic Dock 硬件模块兼容异构接口标准。
> “Trip Planner is our in-vehicle routing or navigation system... And ultimately, our vision for Trip Planner is that it's the air traffic controller for electric vehicle charging across all infrastructure on a global basis.” —— Rebecca Tinucci
> “Over 50% of our superchargers in Europe are currently open to other electric vehicles... And just yesterday, we opened our first 10 supercharger sites here in North America in the US to other electric vehicles.” —— Rebecca Tinucci
### [01:41:30](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6090s) 第四代超级充电桩与日间交流充电布局
特斯拉正在欧洲率先安装第四代超级充电桩（V4 Supercharger）。与上一代相比，V4 配备了更长的充电线，旨在让不同品牌、不同充电口位置的电动汽车都能更轻松地接入充电网络，从而免去车主自行携带转接头的麻烦。
除了硬件接口的通用化，特斯拉在充电网络扩展中重点关注能源来源的清洁度。过去两年中，特斯拉采购的可再生能源已足以抵消为客户提供的全部充电电量。然而面向未来的全球全电动车队，特斯拉希望引导充电负荷曲线与可再生能源的发电周期更加契合。在美国整体电网模型中，风能和太阳能的发电高峰主要集中在白天，因此特斯拉力推“日间充电”模式。为实现这一目标，特斯拉正全力在车辆日间停放场景（如工作场所和公共停车场）大规模铺设便捷且低成本的交流（AC）充电桩，形成超级充电、家庭充电与日间交流充电互补的生态格局。
结合宏伟蓝图第三篇章（Master Plan Part 3），特斯拉充电团队设定了全行业每年 9 PWh（拍瓦时）的产能支撑目标，并坚持向所有品牌开放并提供极致的用户体验。
```mermaid
flowchart TD
accTitle: 特斯拉充电基础设施扩展与能源匹配架构
accDescr: 展示V4超充与交流充电如何结合可再生能源发电曲线，服务全行业车辆。
subgraph Hardware["充电硬件升级"]
V4["V4 超级充电桩<br/>(更长线缆 / 适配非特斯拉车型)"]
AC["低成本日间交流充电桩<br/>(布局于日常停放场所)"]
end
subgraph Energy["能源匹配 (Master Plan 3)"]
SolarWind["日间太阳能与风能发电高峰"]
Offset["100% 可再生能源抵消与消纳"]
end
subgraph Fleet["服务生态"]
TargetFleet["支持全行业车辆并网充电<br/>(目标支撑全球 9 PWh/年需求)"]
end
SolarWind --> AC
Offset --> V4
V4 --> TargetFleet
AC --> TargetFleet
```
- 欧洲作为首发区域已开始部署第四代超级充电桩（V4），其线缆长度显著增加，大幅改善跨品牌车型的插拔适配性。
- 过去 2 年内，特斯拉已采购足够的可再生能源来完全抵消车主充电消耗的电量。
- 行业目标要求支撑全行业每年 9 PWh 的充电容量规模。
---
### [01:44:05](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6245s) 特斯拉独特的供应链模式：供应链工业化工程
特斯拉供应链负责人 Karn Budhiraj 与 Roshan Thomas 登台分享。Karn 负责电子元件、动力总成、电池供应链，并分管间接采购、工程采购及仓储物流；在特斯拉任职超 14 年的 Roshan 则掌管整车大宗物料、光伏、物流规划及资本设备。两位负责人指出，特斯拉与传统汽车制造商依赖采购现成（off-the-shelf）零部件的商业模式截然不同——特斯拉高度自主设计底层子组件，使得供应链管理超越了单纯的商务合同关系。
特斯拉在供应链体系内设立了专属的“供应链工业化工程”（Supply Industrialization Engineering）部门。该团队由机械、电气及工业工程师组成，其职责是将研发团队的设计图纸直接转化为具有成本优势、良品率达标的大规模量产实体。在电动车许多细分供应链尚不存在的背景下，特斯拉工程师会深入供应商一线，协助选型设备、搭建产线，并驻厂驻守数周乃至数月，直至产线速度和良率稳定，随后继续推进自动化降本和良率改进。这种深度绑定的驻厂与数据监控体系，让特斯拉能跨越全球上万家零部件工厂，在问题恶化前就做到实时预警。
Karn 强调了供应链工作的极端容错率以及特斯拉自主制造所带来的挑战：“在供应链领域，完美才算及格（Perfection is a passing grade）。只要一个零件没送到，产线就会停摆，我们的 CEO 在不到 20 分钟内就会知道。”与苹果等外包制造的科技巨头不同，特斯拉坚持自建工厂制造，供应链始终处在核心中枢位置。
- Karn 团队掌管电子架构、动力总成、电池核心供应链，并覆盖间接采购、施工采购和仓储物流配送；Roshan 团队掌管整车大宗商品、太阳能、物流规划与资本设备。
- 供应链工业化工程师亲自参与从图纸到量产全周期，包括产线工艺设计、制造设备选型与现场产线调试（Line Bring-up）。
- 特斯拉通过数字仪表板实时监控全球约 10,000 家供应商制造工厂的运行健康状况。
---
### [01:47:21](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6441s) 从 Model S/X 到 Model 3/Y：供应商精简与二级复杂性控制
在平台迭代过程中，特斯拉对零部件和供应商网络实施了大规模精简。从早期的 Model S/X 平台发展到 Model 3/Y 平台，车辆整体零部件数量大幅削减，这得益于工程设计团队推行的极简与易制造化设计。
```mermaid
flowchart LR
accTitle: 平台迭代与供应商过滤演进路径
accDescr: 展示从Model S/X的大范围供应商筛选，收敛到Model 3/Y的高效率核心供应商网络。
subgraph Gen1["Model S / X 平台"]
direction TB
SX_Parts["零部件基数庞大"]
SX_Net["广撒网式接触供应商<br/>(能力与意愿考察筛选)"]
SX_Parts --- SX_Net
end
subgraph Gen2["Model 3 / Y 平台"]
direction TB
ThreeY_Parts["零件数量显著减少<br/>(更易制造的架构设计)"]
ThreeY_Net["核心优质供应商网络<br/>(具备高技术/资金/文化匹配度)"]
ThreeY_Parts --- ThreeY_Net
end
Gen1 -->|"设计精炼 & 供应商筛选"| Gen2
```
在供应链策略上，特斯拉在开发 Model S/X 时广泛引入外部供应商，将该平台作为严苛的筛选漏斗，以此检验供应商在技术能力、财务稳定性以及企业文化匹配度上是否能跟上特斯拉的高速节奏。Karn 指出：“你必须兼备意愿与能力（You need both desire and capability）。”基于此，特斯拉在 Model 3/Y 上大幅收紧并裁减了供应商数量，这种打法将继续复用到下一代更高产量级别的平台中。
虽然直接供应特斯拉整车工厂的一级（Tier 1）供应商零部件总计约为 8,000 个，看似数量有限，但供应链深层的二级（Tier 2）乃至多级复杂性呈指数级剧增。二级供应商涵盖大量的硅芯片、电阻、电容、二极管和各类底层子总成；若追溯到电池单元层面，特斯拉的介入深度甚至直接穿透至六级（Tier 6）采购环节，与锂矿、钴矿等原材料开采矿场直接对接谈判。为了说明多级供应链穿透管理的必要性，Karn 随后引入了车载电脑（Car Computer）这一高集成的典型案例。
- 供应层级定义：一级供应商（Tier 1）直接供货给特斯拉的 5 座超级工厂；二级供应商（Tier 2）向一级供应商提供元器件；极端情况下（如动力电池原材料）供应链穿透可达六级（Tier 6）。
- 一级零部件总数约维持在 8,000 种，供应链管理的真正壁垒在于向下穿透并精细管理二级元器件。
- 下一代更高吞吐量的汽车平台将继续沿用“极简零件设计 + 严选核心供应商”的高效供应链模型。
### [01:50:32](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6632s) 车载电脑的高集成制造与自动化降本
在散热器的另一侧装配着多媒体集群电路板（MCU）。这块板子同样极为精密，采用双面布局、八层板设计。由于这些车载计算机在峰值运行时发热量极大，必须通过散热器进行液冷散热。装配流程需要将这两块极其复杂的电路板贴合并气密密封在散热器总成上，完成软件烧录及各项测试后，作为单一零件号送达特斯拉。这正是“一个一级供应商（Tier 1）零件号在二级供应商（Tier 2）层面对应极高管理复杂度”的典型案例。
整个车载电脑包含超过 7,000 个零部件；台上演讲的同时，平均每 1.4 毫秒就有一个元件被焊接到车载电脑上。生产该车载电脑的产线足足有一个足球场那么长。最初制造该板卡时，受制于极高的复杂度，产线严重依赖人工。但在将质量、节拍和良品率彻底调试稳定后，团队开始将重心转向降本增效。“控制成本的唯一途径就是剔除人工。第一步是减少人工，第二步是完全自动化，第三步是关灯让工厂全自主运转，这是最理想的状态。”Karn Budhiraj 强调，目前自动化率虽已达到 95%，但工作远未结束，未来还会走得更远。
这个案例证明，“一个零件并不仅仅是一个零件”，其背后隐藏着庞大的复杂性。供应链本质上是一场跨越多个层级的博弈。特斯拉之所以能取得成功，关键在于让受聘于特斯拉的供应商工程师深入所有细节，亲自驻场帮助供应商提升产能并管理每一个工艺属性。正是因为掌握了底层的全部细节，团队才得以在经历芯片短缺、疫情冲击以及各种突发状况时，依然稳定应对每天 7,000 个零部件的庞大运转。
```mermaid
flowchart LR
accTitle: 车载电脑制造自动化演进
accDescr: 展示特斯拉车载电脑从初期劳动密集型到未来黑灯工厂的三步自动化降本路径
subgraph Phase1["初级阶段"]
A["工艺调试<br/>(保证质量与良率)"] --> B["剔除人工<br/>(减少工位依赖)"]
end
subgraph Phase2["进阶阶段"]
B --> C["完全自动化<br/>(目前达 95%)"]
end
subgraph Phase3["终极目标"]
C --> D["黑灯工厂<br/>(无人工自主运转)"]
end
```
- 车载电脑板卡规格：双面贴片、8 层板设计，采用气密密封液冷散热总成。
- 组装速率：每 1.4 毫秒组装一个电子元件。
- 单体部件复杂度：总计包含 7,000 多个独立元器件。
- 制造产线规模：生产线长度达一个足球场。
- 自动化现状与路线：目前已达 95% 自动化，目标是全面迈向“关灯工厂（Lights-out factory）”。
---
### [01:52:38](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6758s) 全球物流规模与就近制造策略
Roshan Thomas 随后分享了特斯拉入厂物流维度的惊人数据。仅在过去一年中，特斯拉就从全球供应商向各超级工厂运送了约 1,600 万个托盘。如果将这些托盘并排铺开，足以绕地球半圈。而在二级供应商层面，为了支撑生产运转，特斯拉每周调运的电子元器件超过 10 亿颗，折合每年高达 520 亿颗，且这一体量仍在持续上升。与此同时，物流网络还必须保证将正确的零部件准时交付到全球约 685 家服务中心，以支持不断扩大的保有车队。
在应对疫情冲击及更早的布局中，特斯拉始终坚定推行关键零部件的双重乃至三重供应链体系，目前供应商已遍布全球 45 个国家。Karn Budhiraj 指出，其核心战略是将子部件的生产尽可能搬迁到靠近消费地的地方。这不仅因为减少柴油运输而使供应链更为环保，更重要的是，一旦某家工厂停产，多产地备份能提供极高的供货安全垫——这正是特斯拉平稳度过疫情危机的关键手段。
- 入厂托盘吞吐量：年运送约 1,600 万个托盘，连起来可达半个地球周长。
- 电子元器件调拨量：当前每周超过 10 亿颗，年吞吐量逾 520 亿颗。
- 售后服务网点：支撑全球约 685 个服务中心的精准即时（Just-in-Time）配送。
- 供应多元化：零部件采用双源或三源采购，供应商网络覆盖全球 45 个国家。
---
### [01:54:10](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6850s) 疫情封控、港口瘫痪与极限保供
回顾过去三年，Roshan 开玩笑说他和 Karn 在疫情前看起来要年轻得多。疫情彻底颠覆了供应链管理的传统规则，特斯拉的供应链团队不再只是坐在办公室里下采购单，而是“挽起袖子，直接与州政府、市政经理以及市长谈判，全力协助供应商复工复产”。面对不同国家在不同时间点爆发的疫情，团队必须与政府事务团队紧密配合，吃透各地防疫法规，给出合理的争辩依据，帮助工厂保全开工许可。因为制造一辆汽车缺一不可，任何一家工厂或辖区因疫情停工，整车装配线就会面临停摆。
正当团队逐渐适应疫情常态化运作时，西海岸又爆发了严重的港口拥堵危机。海运物流的前置时间直接翻倍，物流追踪系统的准确度跌破 35%，这意味着在任何特定时刻，货船的预计到达时间（ETA）有高达 75% 的偏差。特斯拉团队不得不在系统数据极度失真、船期完全不可控的混乱环境中持续摸索，维持工厂的物料供给。
“控制成本的唯一途径就是剔除人工。第一步是减少人工，第二步是完全自动化，第三步是关灯让工厂全自主运转，这是最理想的状态。”Karn Budhiraj 谈到供应链韧性时总结道，正是前期对零部件和供应链架构的大幅简化，加上打下的多源供应基础，才让特斯拉能以远胜传统车企的姿态渡过难关。
- 跨部门政企协作：供应链团队直接下沉与地方政府沟通，为供应商争取复工特批。
- 港口拥堵冲击：西海岸物流交期拉长一倍。
- 物流能见度恶化：追踪系统准确率低于 35%，船期 ETA 误差率高达 75%。
---
### [01:56:34](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=6994s) 芯片荒突围与进军 2000 万产能的半导体测算
港口拥堵尚未平息，全行业又遭遇了灾难性的“缺芯潮”。在整个汽车行业产出停滞甚至整体萎缩的背景下，特斯拉不仅扛住了危机，更实现了产量的强劲增长。Karn Budhiraj 解释了其背后的三点核心原因：
1. **自主设计与深耕二级供应商**：由于特斯拉自主设计整车电子架构，早在 10 年前就跨过一级供应商直接与半导体厂商建立商务关系，因此在出现短缺时能迅速找准负责人升级问题、请求支持，并在产能瓶颈点集中攻坚。
2. **前瞻性阶梯产能规划**：二级供应商的产能扩充必须早于一级供应商，三级供应商则需更加提前，特斯拉团队与战略伙伴始终保持高度步调一致。
3. **清晰稳定的需求信号**：半导体行业重资产运营，投资数十亿美元建立的晶圆厂耗时数年。在全行业因恐慌砍单或摇摆不定时，特斯拉作为使命驱动型公司，始终保持“全油门踩到底（Full throttle, pedal to the metal）”的高速扩张态势，这种无可匹敌的需求确定性赢得了半导体厂商的高度重视。
针对外界“特斯拉如果年产达到 2,000 万辆，是否会击垮全球芯片产业”的质疑，Karn Budhiraj 明确给出了否定的回答。目前配备全自动驾驶（FSD）电脑的特斯拉汽车，其芯片含量大约是普通车辆的 4 倍。在年产约 200 万辆的运行速率下，特斯拉年消耗量折合 12 英寸等效晶圆约 70 万片；而全球 12 英寸晶圆的总年产能约为 1.35 亿片，特斯拉目前的占比仅为 0.5% 左右。未来迈向 2,000 万辆时，伴随着电子电气架构的深度精简与集成，特斯拉完全不需要占用不可承受的半导体产能。
- 芯片含量倍数：搭载 FSD 电脑的特斯拉车辆，芯片半导体用量约为传统汽车的 4 倍。
- 当前晶圆消耗量：按年化 200 万辆推算，年消耗约 70 万片 12 英寸等效晶圆。
- 全球半导体产能占比：在全球约 1.35 亿片 12 英寸等效晶圆年产能中，特斯拉当前仅占约 0.5%。
### [01:59:19](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7159s) 芯片供应链保障与热管理系统的极致集成
特斯拉在评估供应链扩张时，对关键的半导体晶圆需求进行了测算。即便在最悲观的假设情景下，特斯拉达到年产 2000 万辆目标时需要约 800 万片晶圆。而届时整个半导体产业的年产能预计将扩大至 2 亿片左右，特斯拉的需求占比依然低于行业总规模的 5%。“如果我们采取一直以来的进攻策略，继续与合作伙伴协作并投资晶圆厂与产能，我们就会在需要时准备好充足的产能，”供应链团队胸有成竹地指出，“从供应链的角度来看，我真的不认为这是实现 2000 万辆目标的障碍。”
随后，演讲者将话题转向特斯拉如何将电子工程的设计哲学应用到机械架构中。以往特斯拉车辆的热管理系统采用分散式布局，依靠错综复杂的水管互相连接；而随着热泵（Heat Pump）系统的引入，工程团队借鉴了电路板的高度集成理念，打造出被称为“超歧管”（Super Manifold）的紧凑模块。该系统将原先 60% 的热管理组件浓缩到仅相当于一颗篮球大小的空间内，内部集成了约 100 个精密零件、50 个密封界面，并融合了多种截然不同的制造工艺。
```mermaid
flowchart TD
accTitle: 歧管集成化与产线自动化演进
accDescr: 热管理系统从分散式演进为篮球大小的超级歧管，制造流程经历手工、半自动到全自动仿真产线的三级跳。
subgraph Design["系统架构演进"]
A["早期分散式热管理系统<br>(依赖大量管道连接)"] -->|"借鉴电路板理念"| B["超级歧管 (Super Manifold)<br>(篮球大小，集成100+零件)"]
end
subgraph Manufacturing["制造产线迭代"]
B --> C["手动产线<br>(质量问题多/吞吐量受限)"]
C -->|"快速迭代"| D["半自动产线<br>(吞吐量提升/质量缺陷未根除)"]
D -->|"3D产线仿真开发"| E["全自动化生产线<br>(8个月上线，占地相当于两个足球场)"]
end
```
- **晶圆需求预测**：在最悲观估算下，生产 2000 万辆车需 800 万片晶圆，仅占全球预计 2 亿片产能的不到 5%。
- **超歧管（Super Manifold）集成度**：将整车 60% 的热管理系统集成在篮球大小的体积内，包含约 100 个组件与 50 个密封界面。
---
### [02:00:40](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7240s) 疫情下的自动化突破：3D 产线仿真与超歧管制造
超歧管的投产恰逢新冠疫情爆发初期。面对跨国旅行受阻的现实，团队不得不远程管理全球各地由数百家供应商生产的零配件。起初搭建的手动生产线暴露出大量质量问题与产能瓶颈；团队随即转向半自动产线，虽然产量有所突破，但质量缺陷依然居高不下。最终团队得出结论：全自动化是唯一的出路。
在建造全自动产线前，特斯拉构建了一套极其复杂的 3D 数字仿真系统。当某个零件完成规格设计后，特斯拉可以在虚拟环境中精确模拟包含复杂子装配的所有环节，明确指导供应商采购何种设备、设定何种测试流程，并结合供应商的实际厂房布局进行产线优化。历经多轮迭代，这套全自动化产线在不到 8 个月内便完成落地，取得了惊人的成效：
- **人力节约 99%**：在全负荷产能下，原本需要 1000 名工人的超级歧管组装线，最终仅需 10 名员工维持运转。
- **质量实现数量级跃升**：系统缺陷率骤降至 0.005% 以下。
- **超高节拍**：遍布厂区（占地面积达两个足球场大小）的多条自动化产线，平均每 7 秒即可下线一套超级歧管。
演讲者强调，在当下劳动力成本持续攀升、劳动参与率走低、人员流动频繁的大环境下，而自动化成本预计在未来十年将下降三分之二（降至原先的 1/3）。“这是我们思考复杂组装的方式，我们将把这种思维越来越深入地推进到为我们制造复杂总成的供应链基地中，”他断言道，“我们认为这是自信地迈向 2000 万辆汽车目标的唯一途径。”
在为供应链环节做总结时，团队指出，从每周 0 辆爬坡到 4 万辆极其艰难，过程中留下了埃隆所说的“精神伤疤”（Mental Scar Tissue）；但如今地基已经夯实，从每周 4 万辆迈向 40 万辆不再令人畏惧。
---
### [02:05:06](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7506s) 朱晓彤登台：超级工厂的建设极速与全球第 400 万辆达成
特斯拉汽车业务全球副总裁朱晓彤（Tom Zhu）随后登台，主讲整车制造与产能扩张。他回顾自己于 2014 年加入特斯拉，长期负责大中华区及亚太区业务，如今统管全球生产、销售、交付和服务。目前，特斯拉在三大洲拥有四座整车工厂，制造团队员工超过 6.5 万人（包括在内华达州斯帕克斯/里诺制造三电系统的优秀团队），全球安装年产能已达约 200 万辆，并且仍在挖掘现有厂房的挖潜空间。
回顾 2019 年上海超级工厂的建设，朱晓彤展示了平地起高楼的前后对比照片：从一片农田的绿地项目到整车工厂落成，特斯拉仅用了 9 个半月，再过 3 个月便开始向客户交付首批 Model 3。“人们总是问我，特斯拉怎么能建厂这么快？”朱晓彤直言，“说实话，我们从弗里蒙特工厂学到了很多。我们与‘生产地狱’中的幸存者交流，竭尽全力避免我们曾经犯过的所有错误。”为此，工厂被设计成直线流线，大幅减少上下搬运和弯道，简化施工与制造；同时剔除了传统汽车工厂中所有冗余的设计与缓冲库存。此外，特斯拉自建了一支强大的工程建设团队，全面掌控从规划设计、现场施工到项目管理的全流程。这支内部团队不仅打造了上海工厂，还主导了柏林、得克萨斯及内华达扩建工程。
就在演讲当天的早晨，特斯拉全球累计产量正式突破了 400 万辆大关，而第 400 万辆整车恰好就诞生在当天举办活动的得州超级工厂内。特斯拉产量的指数级增长节奏呈现出惊人的加速度：
- **第 1 个 100 万辆**：耗时 12 年
- **第 2 个 100 万辆**：耗时 18 个月
- **第 3 个 100 万辆**：耗时 11 个月
- **第 4 个 100 万辆**：仅耗时不到 7 个月
朱晓彤指着现场观众笑道，大家在厂区参观时很可能就已经从那辆具有里程碑意义的第 400 万辆车旁走过了。
---
### [02:08:48](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7728s) 工厂产能爬坡的核心考核：OEE 与 45 秒节拍
谈到超级工厂的产能爬坡，朱晓彤幽默地表示：“如果你有 600 台机器人、10000 名受过良好培训的员工——或者是 5000 名人类外加 5000 台 Optimus 人形机器人——以及数百道工序，你就能做到。这听起来很简单，但做起来极难。”在日常生产管理中，特斯拉严密聚焦于两大核心指标：
1. **设备综合效率（OEE, Overall Equipment Effectiveness）**：综合衡量设备的开机可用时间、运行性能与产品良率。简单来说，就是用生产合格品实际消耗的有效时间对比计划生产时间，合格线设定在 90% 以上。
2. **生产节拍（Cycle Time）**：总装线终端的产出节拍合格线被设定为 45 秒，即平均每 45 秒就有一辆新车从总装线上驶下。
朱晓彤阐明，工厂产能爬坡的速度越快，规模经济效益就释放得越彻底。以上海超级工厂为例，伴随自动化与节拍提速，单车所需的人力工时和制造劳动力成本实现了断崖式下降。
### [02:10:01](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7801s) 制造效率优化与第一性原理哲学
朱晓彤（Tom Zhu）继续剖析特斯拉工厂的生产效率提升历程。屏幕展示了单车工时（hours per car）在产能爬坡过程中的显著下降曲线。尽管由于 2022 年第二季度的疫情停工（COVID shutdown）导致数据出现短暂波动，但整体效率保持强劲增长。右侧展示的是加州弗里蒙特（Fremont）工厂的 Model Y 车间。即便该厂房已有 6 年历史，现场团队依然通过优化物料流动、消除所有单点故障（single point of failure）并提升产出率，持续大幅压低单车工时。朱晓彤现场祝贺弗里蒙特团队在活动前一天刚刚刷新了单日产量历史纪录。
特斯拉降低节拍时间（cycle time）并持续优化效率的底层逻辑，严格遵循埃隆·马斯克在造火箭中推崇的第一性原理：首先是“质疑”，找出真正需要解决的问题；接着是“删除”与“简化”；随后全力“加速”并施加产线压力，验证方案是否真正起效；最后一步才考虑“自动化”。在上海超级工厂（Gigafactory Shanghai）涂装车间，团队发现 PVC 密封胶（PVC sealer）与面漆（top coat）的烘烤温区存在重叠，原本由两台独立烤箱完成的流程被整合为一道工序，不仅缩短了节拍时间，还直接削减了 9% 的能耗与 9% 的碳排放。弗里蒙特工厂则通过通用化 Model S 与 Model X 的车身结构和前大灯设计，成功精简了约 40 个零部件，节拍时间削减达 10%。
```mermaid
flowchart LR
accTitle: 产线工程优化五步法
accDescr: 展示特斯拉基于埃隆·马斯克第一性原理的产线优化流程：质疑、删除简化、产线增压验证与自动化。
A["质疑需求<br/>(Question)"] --> B["删除与简化<br/>(Delete & Simplify)"]
B --> C["加速与产线增压<br/>(Accelerate & Pressurize)"]
C --> D["验证实际效果<br/>(Verify Solution)"]
D --> E["实施自动化<br/>(Automation)"]
```
- 弗里蒙特工厂即便属于老旧设施，依然在活动前日创下全厂单日产量历史新纪录。
- 上海超级工厂涂装车间合并 PVC 密封胶与面漆烘烤流程，节约 9% 能耗与 9% 二氧化碳排放。
- Model S/X 零部件通用化削减约 40 个零件，工序节拍时间减少约 10%。
- 柏林超级工厂在总装车间（GA shop）部署 5G 专网，解决了特定工序约 90% 的超循环（over cycle）瓶颈，该技术将快速推广至全球产线。
---
### [02:12:31](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=7951s) 全球一体化组织与 2000 万辆产能目标
为消除制造与末端运营的脱节，特斯拉建立了贯穿生产制造、销售交付以及售后服务的全球一体化组织架构。这种闭环机制使客户问题与服务升级诉求能够无缝转化为车间底层的工艺改进。在过去的 6 个月内，新车首发阶段的服务需求率、在保维修停留时长以及服务预约等待时间均出现断崖式下降。
面对年产 2000 万辆汽车的宏大愿景，朱晓彤明确表示现有的四座超级工厂远远不够，特斯拉必须持续建设新工厂、新产线并导入全新车型平台。通过将全球超级工厂统一纳入同一管理体系，各工厂的最佳实践可以即时横向复刻，帮助新建工厂以更陡峭的斜率拉升产能、提升良率并摊薄制造成本。朱晓彤在总结时强调：
> “And with all gigafactories now in one organization we'll be able to replicate unified ideas across factories.”
---
### [02:13:51](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8031s) 4680 电池制造革新与干法电极突破
动力总成与能源工程高级副总裁德鲁·巴格利诺（Drew Baglino）接棒登台，将制造话题扩展至电芯领域。巴格利诺回顾了特斯拉电池日（Battery Day）上展示的实验室干法制膜视频，幽默地调侃道：“我们不妨这么说，现在已经不需要那把勺子了（there's no spoon now）。”得克萨斯超级工厂已建成完全自动化的全尺寸干法电极（dry electrode）量产设备，直接将粉末与集流体金属箔输入，即可连续产出涂覆电极，单机峰值生产率相比加州加藤路（Kato Road）试验线提升了 20 倍以上。
从传统的 2170 圆柱电芯跨越至 4680 电芯，工厂占地面积与空间体积骤降至原来的五分之一（5x reduction）。通过产品设计、制造设计、工艺设计、设备设计和厂房设施设计这五大工程团队的深度垂直整合，从弗里蒙特试验线到得州工厂、再到内华达扩建基地，电池厂房的体积利用效率实现了近 10 倍（10x）的压缩。这不仅大幅降低了每吉瓦时（GWh）的资本支出（CapEx），也为实现年产 2000 万辆电动车以及储能装机年增 1 太瓦时（TWh）的终极目标扫清了物理扩展瓶颈。
```mermaid
flowchart TD
accTitle: 4680 电芯制造系统级设计整合
accDescr: 展示产品、制造、工艺、设备与设施五维协同，驱动厂房占地与投资大幅降低的系统工程图。
subgraph Design["五维协同设计体系"]
D1["产品设计 (Product)"]
D2["制造设计 (Manufacturing)"]
D3["工艺设计 (Process)"]
D4["设备设计 (Equipment)"]
D5["厂房设计 (Facility)"]
end
subgraph Impacts["电芯制造跃迁成果"]
R1["单机生产率提升 20 倍<br/>(无人工干预全自动化)"]
R2["零部件与工艺步骤精简"]
R3["厂房体积压缩 10 倍"]
R4["单位 GWh 资本支出锐减"]
end
Design --> Impacts
```
- 得州干法电极量产设备单机峰值产能比电池日 Kato 原型工具高出 20 倍以上。
- 从 2170 到 4680，工厂占地面积与建筑体积直接缩减 5 倍，经迭代后总体积相较于传统电池厂缩小 10 倍。
- 极简化设计大幅精简了电芯零部件数量与物理生产工艺流程。
---
### [02:16:19](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8179s) 上游原材料布局：科珀斯克里斯蒂锂冶炼厂与得州正极工厂
为了破除产业供应链瓶颈并平抑扩张风险，特斯拉选择性地向电池核心原材料上游延伸。巴格利诺展示了位于得州科珀斯克里斯蒂（Corpus Christi）年产能达 50 GWh 的氢氧化锂冶炼厂渲染图。该项目已正式破土动工，目标在 10 个月内开始设备调试，12 个月内实现投产，复制上海超级工厂的建设速度。该基地距离港口仅 30 分钟车程且直连铁路运输网，工艺上舍弃了产生大量污染废料的传统酸化焙烧法，采用温和高效的纯碱直接浸出技术（direct soda ash leach），原料设计兼容主流的锂辉石精矿（spodumene）以及其他一级、二级回收料源。
与此同时，位于得州奥斯汀超级工厂主建筑后方的 60 GWh 正极材料（cathode）制造车间同样保持着 10 个月的建设节奏，首条产线设备正在进场安装，预计次季度启动调试。巴格利诺明确说明特斯拉并不打算包揽所有原材料开采冶炼，行业内有许多优秀的供应链伙伴，但特斯拉会亲自切入关键节点，通过高扩展性与极具 CapEx 优势的技术创新倒逼行业加速。巴格利诺针对产业链战略明确解释：
> “So, we we are doing this where we need to. Our plan is not to do it always. There are lots of competent companies out there, but we are also trying to sort of accelerate the pace of the industry by trying some new things that are a a bit more scalable.”
---
### [02:18:15](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8295s) 能源业务与 Megapack 规模化拓展
随着电芯制造汇报完毕，巴格利诺邀请 Megapack 业务负责人迈克·斯奈德（Mike Snyder）登台，正式开启能源板块的介绍。斯奈德在特斯拉供职近 9 年，负责领导大型工商业储能（Megapack）业务。斯奈德回顾了储能业务十余年来的成长历程，强调特斯拉不仅是在销售封装在机柜里的电池，而是从一开始就深度参与全球客户的交钥匙项目交付，打造了能够适配全球各类严苛电网环境与气候的软硬件一体化技术平台。
该储能平台具备极高的弹性与扩展能力，涵盖从偏远微电网海岛项目到吉瓦时级（GWh）的大型电网级储能阵列。特斯拉通过深入理解储能项目全流程的工程痛点与潜在风险，将系统打磨为极致的即插即用（plug-and-play）架构。目前，位于加州莱思罗普（Lathrop）Megafactory 超级储能工厂正在全力生产第六代工业储能产品——Megapack XL。斯奈德表示，特斯拉能源业务将依托高度标准化的硬件与先进的自研电网软件，承接 Master Plan 3 中支撑全球可持续能源转型所必需的太瓦时规模电网储能部署。
- 储能业务历经 10 年工程验证与项目沉淀，涵盖从微网到 GWh 级电网储能方案。
- 目前主推的第六代工业级大型储能产品 Megapack XL 正在加州莱思罗普工厂规模化量产交付。
### [02:19:52](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8392s) Megapack 的市场领先地位与极致工程设计
特斯拉能源业务高级总监迈克·斯奈德（Mike Snyder）与动力总成及能源工程高级副总裁德鲁·巴格利诺（Drew Baglino）介绍了特斯拉在大规模储能领域的快速扩张与产品工程细节。特斯拉已在 50 多个国家部署了超过 16 GWh 的工商业与户用储能产品。作为核心主力，Megapack 在效率、可靠性、能量密度以及安装便捷性与成本上均处于行业标杆。斯奈德风趣地解释了为什么给它起名为“XL”：因为它已经是在不封路、不申请极度复杂特种运输许可的前提下，全球公路上能够合法运输的最大、最重物体。他笑着调侃：“所以它是特大号（XL），可以这么说，绝不会再有什么 XXL 了。”
```mermaid
flowchart TD
accTitle: Megapack 系统级一体化设计架构
accDescr: 展示从硬件一体化设计、功率电子大脑到极高能量密度与落地部署的完整架构。
subgraph Hardware["硬件结构创新"]
Base["外壳底座集成线槽<br/>无需地下开槽埋管"]
Density["能量密度 > 300 MWh/英亩<br/>功率密度达燃气调峰电站 2 倍"]
end
subgraph Electronics["内置功率电子大脑"]
Inverter["DC 直流转 AC 交流"]
GridPlug["即插即用全球电网<br/>多台协作如单一体"]
end
subgraph Impact["规模与市场影响"]
Replace["逐一替代化石能源电厂"]
end
Base --> Electronics
Electronics --> Density
Density --> Replace
```
谈及 Megapack 市场需求，斯奈德指出 2023 年将是极为出色的一年，多座 GWh 级别的巨型项目正在建设，加州莱思罗普（Lathrop）Megafactory 超级工厂持续增产扩产，近期路线图上还有新产品推出。实现这一领先地位的关键在于对固定式储能全流程的“狂热专注”。例如，工程团队直接将走线通道（wireways）设计进 Megapack 箱体底座内部，施工时无需像传统方案那样在地下铺设大量导管与线缆，直接接入底座即可，既大幅提升了部署速度，又显著削减了工程造价。其内置集成的功率电子设备则是整套系统的“大脑”，负责直流电到交流电的转换，出厂即具备直接接入全球任何电网的能力，并能让现场成百上千台 Megapack 如同单一整体般协同运转。整个场站的能量密度高达每英亩 300 MWh 以上，功率密度达到了传统燃气调峰电站的两倍。巴格利诺强调：“这就是将化石燃料淘汰出局的产品，一座接一座地替代传统电厂。”
- 全球 50 多个国家已累计部署超过 16 GWh 工商业和户用储能系统。
- Megapack 占地能量密度超过 300 MWh/英亩，功率密度达到典型燃气调峰电站的 2 倍。
- 底座集成线槽设计替代传统地下挖沟排管，大幅加速现场施工。
### [02:22:03](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8523s) 功率电子的核心地位与虚拟充电机（Virtual Machine Mode）
巴格利诺深入阐述了功率电子技术对整个可持续能源经济的战略支撑。特斯拉在储能和车辆端累计交付的功率电子设备已突破 1.4 TW，按年化计算，特斯拉每年交付的功率电子规模比全球太阳能和风能产业加起来还要多。功率电子设备每秒进行数千次乃至数十万次的超高速开关，以极致效率实时响应电网或车辆的需求，本质上由软件核心驱动。巴格利诺形容道：“它们确实是可持续能源经济中连接发电、储能与终端用电之间的纽带。”
```mermaid
flowchart LR
accTitle: 虚拟充电机模式（Virtual Machine Mode）运行机制
accDescr: 展现软件定义的合成惯量如何像减震器一样平抑电网波动。
GridDisturb["电网波动与振荡<br/>（高比例风光并网）"] --> VMM["虚拟充电机模式<br/>（软件算法合成惯量）"]
VMM --> Dampen["如避震器般平抑振动<br/>提供电网物理级稳定性"]
Dampen --> BlackoutPrevent["替代巨型旋转水火电机组<br/>防止大停电并支撑 100% 可再生能源"]
```
在储能端，特斯拉自主研发了名为“虚拟充电机模式”（Virtual Machine Mode）的底层固件功能。斯奈德将其类比为车辆的减震器：“虚拟充电机模式对电网稳定性的贡献，就像汽车避震器吸收振动并保持行驶平顺一样；如果无法抑制振动，汽车就会失控，而在电网上则可能发生大停电。”传统电网依赖大型化石燃料机组或巨型水轮机的旋转惯量来维持电网频率稳定，而虚拟充电机模式利用软件算法控制电池功率电子，提供“合成惯量”（synthetic inertia），让电池储能电站表现得就像一台巨大的旋转机械。巴格利诺补充指出，某家电网运营商曾明确表态，如果没有该功能，他们绝不敢让电网在 100% 可再生能源工况下运行。斯奈德说道：“我们认为这是未来电池创造价值的方式，它关乎电网功率的稳定性，而不仅仅是提供能量。”
- 特斯拉在车辆与储能领域累计部署功率电子容量超 1.4 TW，年交付量超越全球光伏与风电产业总和。
- 功率电子器件开关频率达每秒数万至数十万次，全部依托软件深度控制。
- 虚拟充电机模式（Virtual Machine Mode）通过软件合成物理旋转惯量，平抑电网频率震荡，赋能 100% 可再生能源安全并网。
### [02:24:22](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8662s) 自研能源交易平台 Autobidder 与极致交付速度
针对储能资产的运营调度，斯奈德介绍了特斯拉自主开发的自主能源交易算法平台 Autobidder。不同于单一的太阳能、风能或火电机组，电池储能系统由于响应速度极快、充放电多向流动且具有多维度辅助服务价值，需要每 5 分钟做出数百甚至上千次实时决策。由于市场上缺乏能够胜任如此高维度调度的商业软件，特斯拉自主构建了 Autobidder。该系统不仅执行基础的“低买高卖”套利，还能实时参与电网各类调频与备用市场以最大化电池寿命与经济收益。Autobidder 已在澳大利亚、美国得克萨斯州和英国等地得到验证并成为行业领跑者。
```mermaid
flowchart TD
accTitle: 莱思罗普工厂改建与 Megapack 现场交付提速流程
accDescr: 展示从厂房重构到现场快速吊装并网的高速执行流程。
subgraph Factory["工厂极速落地（不到 12 个月）"]
OldDepot["前 J.C. Penney 物流仓储中心"] --> Retool["制造/工程/施工跨职能高度垂直整合"]
Retool --> Megafactory["世界级 Megafactory 超级工厂"]
end
subgraph Deployment["现场即插即用（Plug-and-Play）"]
Ship["Megapack 出厂运输"] --> DirectInstall["底座集成接线直落安装"]
DirectInstall --> GridReady["4年内安装速度提升 4 倍<br/>综合人力投入降低 3 倍"]
end
Factory --> Deployment
```
能够支撑大规模储能扩张的另一大支柱是执行速度。斯奈德展示了加州莱思罗普（Lathrop）Megafactory 超级工厂的建设延时摄影：特斯拉仅用不到 12 个月的时间，就将原 J.C. Penney 的旧物流中心改造成了世界级现代化制造基地。这种敏捷度得益于特斯拉高度的垂直整合——整车制造工程团队、建筑工程团队与储能设计团队同处一室，全员参与决策并快速拍板。在项目安装现场，凭借即插即用（plug-and-play）的模块化设计，过去 4 年间现场安装速度提升了 4 倍，将制造与现场施工所需的总人力工时削减了 3 倍，大幅缩短了从 Megapack 出厂到并网运营的时间窗口。
- Autobidder 每 5 分钟处理成百上千项自动交易与调度决策，全面释放储能多重收益。
- 加州莱思罗普超级工厂改造仅耗时不到 12 个月。
- 4 年内现场安装速度提升 4 倍，全流程制造与施工人力减少至三分之一（降低 3 倍）。
### [02:26:40](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8800s) 分布式能源网络：Tesla Electric 与虚拟电厂（VPP）
特斯拉的储能战略不仅涵盖 Megapack 这样的大型集中式项目，还深度布局户用与商业分布式储能。巴格利诺详细介绍了在得克萨斯州上线的零售电力业务 Tesla Electric，以及其如何通过将分散的 Powerwall 与光伏系统聚合，释放分布式能源的完整网络价值。斯奈德指出：“Tesla Electric 真正释放了分布式能源与储能产品的全部价值，它让我们的客户能够成为他们自己的公用电力公司。”
| 用电与储能配置阶段 | 南澳 5,000 户家庭月均服务用电成本 | 电网角色与经济模式 |
|:--- |:--- |:--- |
| **基准传统电网供电** | **$140 / 月** | 纯电网被动消费者，全额承担电价波动 |
| **仅安装光伏 + Powerwall（离线孤岛）** | **$70 / 月** | 仅供家庭自发自用，不与电网及电力市场交互 |
| **接入 Tesla Electric 虚拟电厂（VPP）** | **净收益为负（用户反向获得收益）** | 智能参与电力市场与电网互动，向电网反向售电获利 |
巴格利诺展示了南澳大利亚虚拟电厂（VPP）在 2022 年对 5,000 名客户跟踪的真实运行数据：
1. 若普通家庭完全依赖传统公用电网供电，平均每月的供电服务成本约为 140 美元；
2. 如果该家庭安装了屋顶光伏和 Powerwall，但在本地孤立运行、不参与电网调度交易，月度供电成本降至 70 美元；
3. 当接入 Tesla Electric（由 Autobidder 底层软件专为分布式能源适配升级而来）进行全自主智能调度时，系统在电网高价时段反向输送清洁电能并提供电网支撑，这 5,000 户家庭不仅不用支付电费，反而获得了特斯拉支付的电网能源服务收益。
巴格利诺强调，南澳大利亚是一个极具代表性的未来样本：2022 年南澳大利亚 70% 的电能供应来自风电和光伏，而同期得克萨斯州为 30%，加利福尼亚州为 35%。南澳大利亚向全世界印证了当高比例可再生能源并网时，通过智能软件驱动分布式储能构成的虚拟电厂，将是全球能源基础设施演进的必然方向。
- 2022 年南澳大利亚风光发电占比高达 70%（得州为 30%，加州为 35%）。
- 5,000 户接入 Tesla Electric VPP 的家庭从每月 140 美元用电支出转变为通过售电与辅助服务净赚收益。
- Tesla Electric 核心算法继承自 Autobidder，实现成千上万分布式资产的秒级自主协同。
### [02:28:23](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=8903s) 能源零售计划 Tesla Electric 与储能部署远景
储能系统的集中式与分布式资源是解锁完全可再生能源电网的关键所在，而 Tesla Electric 正是特斯拉将其落地并服务消费者的零售电力计划。目前该服务已率先在德克萨斯州面向安装了 Powerwall 的家庭推出。然而，全球有超过 10 亿人生活在具备竞争性零售电力机制的市场中，涉及的年度能源支出超过 2 万亿美元，这构成了极其巨大的商业机会。特斯拉计划像推进 Tesla Insurance（特斯拉保险）一样，逐个市场逐步扩展该业务，旨在为客户创造价值并降低产品的全生命周期拥有成本（TCO）。
紧接着，主讲人宣布了一项即将在德克萨斯州推出的计划：到当年夏季，特斯拉将面向其车主推出一项全新的零售电力套餐——每月仅需 30 美元，即可享受无限制的夜间家庭充电服务。这项优惠之所以可行，是因为德州拥有极其丰富的风力资源，且风力主要集中在夜间发电。因此，引导车主在夜间利用风能为电动车充电，不仅为用户省钱、提高电网利用率，还能直接消耗可再生能源，完全契合宏图计划的初衷。
综合来看，特斯拉正处于储能部署大规模扩张的起跑线上。结合宏图第三篇章（Master Plan Part 3）所提及的数十太瓦时（TWh）固定式储能需求，特斯拉已将目光瞄准了每年 1 太瓦时（1 TWh）的年产能目标，这一规模是加州莱斯罗普（Lathrop）Megafack 工厂产能的 25 倍。近期需求方面，Megapack 展现出强劲的增长势头，2023 年市场需求已超过 100 GWh/年，并预计在未来数年内以每年超过 100 GWh 的速度持续递增。因此，加速扩建产能并保持快速爬坡势在必行，这为特斯拉和 Megapack 业务带来了巨大的市场空间与电网转型影响力。
- 德州夜间家用充电零售套餐：每月 30 美元无限量夜间充电，直接消纳夜间风电。
- 全球竞争性零售电力市场规模：覆盖逾 10 亿人口，年支出超 2 万亿美元。
- Megapack 需求预测：2023 年需求超 100 GWh/年，未来数年年增量均超 100 GWh。
- 长期产能目标：年产能达到 1 TWh，相当于 Lathrop 储能工厂现有产能的 25 倍。
---
### [02:31:15](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9075s) 特斯拉的影响力：团队文化与全员改善机制
法务团队的布伦丹（Brandon Ehrheart）登台感谢投资者的支持，强调董事会与管理层保持着积极互动并倾听投资者心声，随后引出负责环境、健康、安全、安保与可持续发展（EHS&S）的副总裁劳里·谢尔比（Laurie Shelby）。谢尔比加入特斯拉已五年半，她指出，当天大会展示了卓越的产品、工艺与软件，但特斯拉一切成就的核心驱动力是其全球团队。无论在工厂、交付中心、服务中心、仓库还是现场一线，都能深刻感受到团队的协作与自豪感。
目前特斯拉已成为庞大的雇主，全球员工规模达到 12.9 万人，其中一半以上从事汽车制造，近 60% 的员工位于美国本土。特斯拉对顶尖人才具有极强的吸引力，在工程类求职意向排行中，SpaceX 和特斯拉稳居第一和第二名；仅在 2021 年，特斯拉就收到了高达 300 万份求职申请。在企业内部文化上，特斯拉推行名为“Take Charge”的员工参与机制，任何员工在任何时刻都可以就安全、人员、成本、生产或质量提出改进建议。正如 Laurie Shelby 强调指出的核心理念：“People who do the work, the people who do the work know how to improve the work.”（做具体工作的人，最懂得如何改进工作）。数据显示，随着“Take Charge”提案数量的上升，工伤事故率显著下降。
```mermaid
flowchart LR
accTitle: 员工参与与安全改善正向循环
accDescr: 展示一线员工通过提案降低工伤事故的良性闭环。
subgraph Culture["全员参与 (Take Charge)"]
A["一线员工提出改进构想<br/>(安全/成本/生产/质量)"]
B["管理层采纳落地"]
end
subgraph Outcome["业务成效"]
C["生产与工艺优化"]
D["工伤事故率持续下降"]
end
A --> B
B --> C
C --> D
D -.->|"增强员工自豪感与参与度"| A
```
- 员工规模：全球达 12.9 万人，超 50% 投身车辆制造，约 60% 常驻美国。
- 人才吸引力：工程学子首选雇主排名前二为 SpaceX 与 Tesla；2021 年收到 300 万份简历。
- 改善机制：推行“Take Charge”，工伤发生率与员工改善提案量成反比。
---
### [02:34:42](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9282s) 可持续发展与环境影响力数据
劳里·谢尔比强调，可持续发展是特斯拉的终极使命，特斯拉绝不搞应付差事的“表面合规”（tick boxes），而是专注评估产品和软件带来的实际减排影响。2021 年，特斯拉车主与能源客户累计避免了 840 万公吨温室气体的排放，其环境效益相当于让 170 万辆传统燃油车（ICE）彻底停驶一整年。
与此同时，特斯拉致力于在自身运营中推进可持续：工厂的水资源单耗与废弃物排放逐年下降，制造基地与超级充电网络内部也正大规模建设可再生能源发电设施。从全生命周期能源收支来看，在 2012 年至 2021 年间，特斯拉旗下清洁能源产品（如光伏等）产生的能源总量，已经超过了这期间所有特斯拉车辆行驶及工厂运转所消耗的能源总和。即使在最严苛的假设场景下——即电网 100% 依赖燃煤发电，一台 Model 3 全生命周期的排放量依然低于普通燃油车。针对投资人对透明度的诉求，谢尔比说明特斯拉正在对齐气候相关财务信息披露工作组（TCFD）的框架标准，后续将在《影响力报告》（Impact Report）中公布更多详实数据。
| 评估维度 | 特斯拉公开数据 / 表现 |
|:--- |:--- |
| **客户减排效益（2021 年）** | 避免 840 万公吨温室气体排放（折合 170 万辆燃油车停驶 1 年） |
| **能源代际收支（2012–2021 年）** | 能源产品累计发电量 > 车辆行驶总耗电 + 工厂运营总耗电 |
| **极端场景表现** | 即使电网 100% 燃煤，Model 3 排放仍低于同级燃油车平均水平 |
| **ESG 报告合规进展** | 正式对齐 TCFD（气候相关财务信息披露工作组）框架标准 |
---
### [02:37:54](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9474s) 财务总结与成本工程的核心逻辑
首席财务官扎克·柯克霍恩（Zach Kirkhorn）登台作大会预定发言环节的最后总结。他首先回顾道，这是特斯拉历史上首次举办正式的投资者日，也是第一次让整个领导层团队公开登台阐述各自的技术与业务进展。柯克霍恩代表公司对幕后整个工程及管理团队的付出表达了由衷谢意。
柯克霍恩透露，在筹备会议阶段，团队成员曾询问他演讲的重点，他给出的唯一指导原则是：“围绕你们正在做的事情谈，并把它们放在宏图计划的框架下来谈。”正因如此，当天几乎所有业务负责人都聚焦于同一个底层逻辑——千方百计地降低成本。柯克霍恩直截了当地点明了汽车与制造业的残酷本质：“Because in this industry, in this business, you survive or you die based upon the ability to manage your costs.”（因为在这个行业、这项业务中，你能否生存全看你控制成本的能力）。
作为成本控制的经典范例，柯克霍恩以特斯拉量产时间最长的代表性产品 Model 3 为例指出：自 2018 年年中 Model 3 达到每周 5,000 辆的设计产能以来，特斯拉已经在其后续的持续生产中将单车成本削减了整整 30%。这证明了特斯拉在规模化制造过程中，不仅依靠初期的产线设计，更依赖跨部门、全流程持续不断的工程优化来降低整体成本。
- 成本控制法则：把控成本是车企生死存亡的红线。
- Model 3 降本成绩：自 2018 年年中达成周产 5,000 辆设计产能至今，单车制造成本已累计下降 30%。
### [02:39:40](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9580s) 降本增配哲学与下一代平台 50% 成本削减目标
特斯拉首席财务官扎克·柯克霍恩（Zach Kirkhorn）指出，成本意识早已深深烙印在公司的文化基因中，这也是特斯拉能够走到今天的核心基石之一。在推进降本工作时，最省事的做法莫过于直接砍掉成本，但这往往会导致产品品质缩水；特斯拉的底层逻辑恰恰相反——必须在剔除冗余成本的同时持续提升产品体验。拿当前下线的 Model 3 与 2018 年最初量产的版本相比，不仅加入了海量产品升级与功能改良，整车制造成本反而显著走低。柯克霍恩强调，降本从来没有所谓的万灵丹，而是全方位协同攻坚的结果：在此期间，Model 3 项目的交付规模增长了三倍；弗里蒙特（Fremont）工厂的单位生产力相比 2018 年年中翻了一番；再加上管理费用优化、制造设计改良以及上海超级工厂所实现的深度供应链本土化，特斯拉与供应商共同榨取系统效率，最终将物料成本大幅打了下来。
面向下一代汽车平台，特斯拉设立了更加激进的目标：在现有基础上再砍掉 50% 的成本。从 Model S/X 平台迈向 Model 3/Y 平台时，特斯拉已经成功实现了 50% 的成本降幅，如今的任务就是将这一壮举再次复制。柯克霍恩阐述道，降本之所以至关重要，是因为一旦平价门槛被打破，能够买得起特斯拉的潜在客户群将呈爆发式增长，在宏大蓝图（Master Plan）的指引下，产品成本每实现一次线性递减，都将撬动销量规模的指数级飞跃。为了挤出这 50% 的成本，团队将整个降本盘子拆分成了整车工程、电池与动力总成、制造工序及其他支持板块，每一个板块所承担的削减体量大致相当，这意味着没有任何部门可以置身事外。
> “When we're working on cost reduction, it's easy just to take costs out and make our products worse. But we have to take costs out and improve our products at the same time.” —— 扎克·柯克霍恩
- **Model 3 规模效应**：项目交付规模增长 3 倍，弗里蒙特工厂当前生产效率达 2018 年年中的 2 倍。
- **下一代平台目标**：继 Model S/X 到 Model 3/Y 削减 50% 成本之后，下一代平台整车制造成本目标再降 50%。
- **成本拆解结构**：50% 降本指标由整车（Vehicle）、电池与动力总成（Battery and Powertrain）、制造费用（Manufacturing）等几个权重相当的模块共同分摊。
```mermaid
flowchart LR
accTitle: 平台成本迭代路径
accDescr: 展示从Model S和X到Model 3和Y再到下一代平台的50%成本持续削减阶梯
subgraph Gen1["第一代平台"]
SX["Model S & Model X<br/>基础成本基准"]
end
subgraph Gen2["第二代平台"]
ThreeY["Model 3 & Model Y<br/>成本降低 50%"]
end
subgraph Gen3["下一代平台"]
NextGen["Next-Gen Vehicle / Robotaxi<br/>成本再降低 50%"]
end
SX -->|"规模化与精益设计"| ThreeY
ThreeY -->|"动力总成/电池/制造一体化创新"| NextGen
```
---
### [02:42:01](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9721s) 全生命周期拥有成本（TCO）与极致运营支出（OpEx）管控
除了购车时的初始裸车价，特斯拉更加看重迈向可持续经济过程中的全生命周期总拥有成本（Total Cost of Ownership per mile）。在计算五年期每英里综合用车成本时，必须综合考量购车融资利息、车辆保险费用、电力补充开销（如德州推出的针对性电价方案）、日常磨损以及养护维保。柯克霍恩展示的数据表明，当前在美国市场上，一辆基础版 Model 3 的每英里综合持有成本，已经比全球长期销量冠军丰田卡罗拉（Toyota Corolla）还要便宜。随着下一代平台及未来 Robotaxi 无人驾驶出租车版本的落地，每英里行驶成本还将进一步断崖式下跌，展现出对全球主流内燃机经济型轿车绝对维度的降维打击。
在剖析资产负债表与利润表时，柯克霍恩话锋一转，指向了资本市场平时极少深入审视、但对特斯拉极为关键的指标——运营支出（OpEx）。每次财报电话会上投资者往往过度聚焦于毛利率与单车制造成本，但自 2018 年以来，特斯拉的 Non-GAAP 运营支出占总营收的比例整整下降了 60%，摊薄到每辆交付车辆上的运营支出更是暴跌了 75%。在研发费用（R&D）端，特斯拉的核心法则在于极力克制并行推进的研发项目数量，集中兵力确保每个启动的项目都能榨取最大的营收与现金流回报，同时将不同业务线间的底层技术复用发挥到极致，例如能源业务中 Megapack、Powerwall 所用的电力电子架构与车载电驱系统的电力电子硬件高度共享。
> “We're already at a place today in the US where a base Model 3 on a cost per mile basis is less than a Toyota Corolla, which is the highest selling car in the world.” —— 扎克·柯克霍恩
- **每英里全生命周期成本**：基础版 Model 3 的 5 年综合 TCO 已正式跨过丰田卡罗拉的门槛，成为更经济的出行选择。
- **运营支出显著压降**：2018 年至今，Non-GAAP 运营支出占营收比例累计下降 60%，单车均摊运营支出降幅达 75%。
- **研发效率最大化**：收敛研发战线，禁止盲目铺摊子，深度共享车辆与储能业务间的电力电子及软件底层架构。
---
### [02:44:08](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=9848s) 摆脱“后勤运营地狱”与特斯拉自研操作系统（Tesla Operating System）
在销售、一般及行政费用（SG&A）维度，特斯拉相较传统汽车工业呈现出碾压性的成本优势。传统主机厂受困于复杂的层级体系以及庞大的独立特许经销商网络，而特斯拉凭借直营体系和端到端垂直整合，使单车承担的 SG&A 费用较传统主机厂整整低了 60% 至 70%，折算下来单车省下的管理运营成本高达数千美元。这也是柯克霍恩反复强调公司内部更加看重营业利润率（Operating Margin）而非单纯毛利率的根本原因。然而这种成本优势并非与生俱来；在当年弗里蒙特工厂艰难拉升 Model 3 产能之际，工厂周产量从 Model S/X 的 2000 辆硬生生拔高到 Model 3 爬坡后的 7000 辆（折合年化约 35 万辆），外界只知道车间里惨烈的“生产地狱”，却极少有人知道特斯拉后台同时经历了近乎崩溃的“后勤运营地狱（back office operations hell）”。面对未来 2000 万辆的年交付远景，原本依靠通用商业软件支撑的后台流程彻底瘫痪，倒逼团队彻底重构企业运营管理体系。
为了支撑数千万级别的业务规模，特斯拉构建了与车端自研系统同等地位的“特斯拉操作系统（Tesla Operating System）”。特斯拉拒绝引入笨重、昂贵且需要无休止定制化的第三方通用 SaaS 软件网络，而是成立了专门的应用工程团队，联合内部业务流程改进小组，全面推行与整车制造完全一致的优化准则：果断删除冗余环节、极致简化既有流程、最后将留存步骤全面自动化。从全面替换第三方招聘系统、到开发支持极高并发的直营销售、自营交付管理与自营金融服务，特斯拉甚至基于该系统打造了全垂直整合的专属保险业务（Captive Insurance），拥有自己的全套理赔软件、精算系统与专属理赔团队。这种软件完全服务于业务最敏捷的需求，不仅使北美销售团队的单兵人效较 2018 年暴增 4 倍，更为管理层提供了全局实时、颗粒度极细的底层运营数据仪表盘，使整个商业帝国的运营动作能够根据瞬息万变的市场反馈进行秒级微调。
> “You've heard some folks here today talk about production hell, but I don't think we've ever talked about as a company is back office operations hell, which also happened at the same time.” —— 扎克·柯克霍恩
- **单车 SG&A 优势**：通过直营模式省去传统经销商渠道加价与中间层级，单车 SG&A 费用较传统主机厂低 60% 至 70%。
- **后勤管理突破**：弗里蒙特工厂周产突破 7000 辆引发“后勤地狱”，促成了全面自研 Tesla Operating System 的战略决策。
- **业务人效提升范例**：自研轻量化企业软件上线后，北美销售团队人均产出达 2018 年基准的 4 倍；订单处理与金融服务效能同步出现量级跃升。
- **全栈自营保险闭环**：自研底层索赔软件、雇佣专属理赔理算专员，直接打通车辆传感器与车险风控业务。
### [02:48:38](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=10118s) 内部流程改进与高效的财务结算
在流程迭代与系统支持方面，公司内部各职能团队与应用工程团队之间的沟通极为顺畅。一旦需要追踪某些新数据或调整业务流程，往往只需要开一次 Microsoft Teams 会议，工程团队就能立即跟进落地。这种紧密的反馈循环让企业内部流程的自我优化速度快得惊人。
此外，财务团队在结账与财报提交效率上也取得了显著成效。在全美市值前二十大的头部上市公司中，Tesla 提交 10-Q 季度报告与 10-K 年度报告的速度名列前茅，体现出极高的数据整合度与后台运营效率。
### [02:49:12](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=10152s) 宏大愿景下的资本开支与偿付能力
所有的降本工作都是为了给实现后续宏伟目标储备弹药。根据秘密宏图第三篇章（Master Plan 3）的规划，Tesla 树立了极其庞大的长期制造目标：实现每年 2000 万辆汽车的年产能、1 太瓦时（1 TWh）的储能产品年产能，并同步扩建电池电芯制造体系、服务网络以及超级充电基础设施。
针对实现这一系列目标所需的总资本投入，管理层给出了测算。截至本次投资者日，公司在整个发展历程中累计支出的资本约为 280 亿美元。虽然未来的总投资额表面上看起来非常庞大，但结合 2022 年强劲的经营现金流来考量，即便仅假设未来保持极其保守、温和的现金流增长，公司依靠自身业务造血能力来覆盖这一级别的投资也是完全行得通的。正如 Zach Kirkhorn 所言：“I actually think it's quite small relative to our ambitions.”（相较于我们的雄心壮志，我其实认为这个数字相当小。）
- 目标年产能：2000 万辆乘用车与商用车
- 储能生产目标：每年 1 TWh
- 历史累计资本开支：约 280 亿美元
- 资金来源依据：2022 年经营现金流基础及稳健预测
### [02:50:21](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=10221s) 资本配置的四级优先级架构
在被频繁问及的资本配置逻辑上，管理层给出了清晰直观的框架。首要任务是保障日常运营所需的流动资金。汽车与能源制造属于典型的营运资金密集型行业，由于新车生产与集中交付的时间节点差异，一个季度内账面现金流可能会出现数十亿美元的上下波动，必须保留充足的缓冲资金。第二层级是防范下行风险（downside protection），过去数年全球疫情暴发及供应链中断的危机充分证明了这一安全垫的必要性。
在守住稳健运营底线后，日常业务产生的健康现金流将全力反哺增长投资。即便面临严峻的宏观经济动荡，公司从未缩减过在业务增长上的投入。在增长投资之外，第三层级是机会主义资金配置，主要投向回报率超出既定门槛的高回报项目（例如弗里蒙特工厂内部的技改降本）。第四层级才是富余资金的处理，由董事会定期评估长期现金流预测，审慎决策是否实施股票回购或派发股息。
```mermaid
flowchart TD
accTitle: 资本配置四级优先级架构
accDescr: 展示特斯拉从日常运营到富余资金回购的四层资本分配递进逻辑。
subgraph CoreOps["核心安全层"]
A["1. 日常运营资金<br/>（应对交付波动）"]
B["2. 抵御下行风险<br/>（外部冲击安全垫）"]
end
subgraph GrowthInv["价值创造层"]
C["3. 核心增长再投资<br/>（工厂、电池与研发）"]
D["4. 机会主义高回报项目<br/>（技术降本改造）"]
end
subgraph ExcessCap["富余分配层"]
E["5. 股东回报审议<br/>（股份回购 / 股息）"]
end
A --> B
B --> C
C --> D
D --> E
```
### [02:52:01](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=10321s) 财务总结：创新、可负担性与规模效应
在总结财务演讲时，核心逻辑贯穿于技术创新对成本曲线的重塑。通过高强度的工程创新降低制造成本，其最终目的在于大幅提升终端产品的可负担性（affordability），而在下一代汽车平台上这一优势将体现得淋漓尽致。
当汽车变得更加普惠、亲民时，公司便能自信地追加投资以扩大产能体量；激增的产销量持续释放充沛的现金流，反过来支撑更深层次的先进制造投资。Zach Kirkhorn 总结指出：“And this is what's ultimately required to accelerate the world's transition to sustainable energy.”（而这正是加速世界向可持续能源转型所最终需要的。）
### [03:03:07](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=10987s) 重磅宣布：墨西哥新超级工厂落户蒙特雷
在短暂停歇后，Elon Musk 重新登台并宣布了全天最受瞩目的核心重磅消息：Tesla 确定将在墨西哥新莱昂州首府蒙特雷（Monterrey）附近建造下一座超级工厂（Giga Mexico）。墨西哥当地的新莱昂州州长及相关官员也亲临现场见证。
Elon 特别澄清并强调，建设墨西哥超级工厂完全是为了提升全球总产能，而不是把产能从既有工厂转移出去。加利福尼亚州弗里蒙特工厂、内华达超级工厂、得克萨斯超级工厂、柏林超级工厂以及上海超级工厂的产能扩张计划不仅不会停滞，反而会继续推进。随后，公司高管团队悉数登台，正式启动面向全球投资者与分析师的现场问答环节。
- 选址公布：墨西哥新莱昂州蒙特雷近郊
- 战略定调：纯增量扩张（Supplemental output），绝非转移现有工厂产能
- 协同工厂：加州、内华达、得州、柏林、上海均同步维持扩张节奏
### [03:06:05](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11165s) Q&A：下一代整车平台的部署时间表与技术落地
Wolfe Research 的分析师 Rod Lache 率先发问。他提到本次展示的下一代汽车平台、电驱系统与电池制造是一套颠覆整车装配模式的全新范式，因此希望了解具体的实施落地时间表，以及这一架构变革是否会倒推应用到现有产线（如得州正在生产的 Model Y）。
Elon Musk 直截了当地指出，重大架构调整将主要集中在未来的全新车型上，因为大规模改造现有工厂产线意味着产线必须面临长时间的全面停产，这绝非明智之举。不过，Model Y 已经在逐步吸纳渐进式创新，包括后底板一体压铸、前后一体压铸以及结构化电池包等不同配置的演进。随后车辆工程副总裁 Lars Moravy 补充表示，工程团队会在不影响现有产能稼动率的前提下，尽可能快地自左向右推进部署；墨西哥超级工厂将负责生产下一代高产量车型，而其他现有超级工厂后续也将陆续跟进导入下一代造车技术。关于新平台的导入策略，Elon Musk 坚定地阐明：“Um retooling a factory means bringing the factory down for an extended period of time and uh that's that would prefer not to do that, I think.”（对工厂进行产线重组意味着产线要在较长一段时间内彻底停工，我认为我们并不希望那样做。）
### [03:08:19](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11299s) 锂提炼瓶颈与第一性原理采矿创新
分析师 Adam 首先祝贺特斯拉在墨西哥建立新超级工厂（“Enhorabuena, Mexico vamos!”），随后提问特斯拉如何将第一性原理思维应用到传统采矿与原材料提取领域。他提到特斯拉数年前曾申请过利用氯化钠（食盐）从粘土和锂辉石中提取锂的专利，询问特斯拉是否打算以此革新整个采矿行业、降低成本以解决这一潜在的制约因素。
Elon Musk 明确指出，特斯拉的策略是随时解决当前眼前的限制性瓶颈，公司绝非渴望涉足采矿或精炼行业。Musk 直言：“我们做这些是因为不得不做，而不是因为我们想做。”（“We're doing it cuz we have to not cuz we want to.”）他强调，真正的卡脖子环节在于精炼产能（Refining capacity），例如将原材料转化为正极、负极材料以及氢氧化锂或碳酸锂的过程。因此，特斯拉正在德克萨斯州科珀斯克里斯蒂（Corpus Christi）自建一座锂精炼厂，并在当前活动现场旁边建设一座大型正极材料加工工厂。
Drew Baglino 进一步补充，特斯拉目前拥有大型锂供应商，公司正向供应商主动分享科珀斯克里斯蒂工厂所采用的技术方案，因为这些技术在底层逻辑上更具扩展性。无论是盐提锂还是黏土提锂工艺，特斯拉完成技术验证后都会把知识免费传授给现有供应商，目的是带动全行业加快建设并降低资本支出。正如 Baglino 所说，全美此前几乎没有任何大规模的正极材料生产能力，特斯拉必须用第一性原理证明可行性，再赋能整个产业链。
- 行业瓶颈：Musk 认为最关键的卡脖子环节是正极、负极、氢氧化锂和碳酸锂的提炼产能，而非原矿储量。
- 产能布局：正在 Corpus Christi 建造锂精炼厂；在得州超级工厂主楼旁同步建设大型正极材料加工制造厂。
- 知识共享策略：特斯拉无意垄断上下游制造，技术路径验证成功后将向现有大型供应商输出工艺方案以加速全球产能扩张。
---
### [03:12:00](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11520s) 可再生能源生态与储能的真正价值
Baird 分析师 Ben Kallo 询问，鉴于可再生能源是“宏图计划第三篇章”（Master Plan Part 3）的核心支柱，特斯拉还能采取哪些措施来推动整个可再生能源产业加速发展。Musk 笑着向全球创业者抛出橄榄枝，称如果有创业者想寻找成功率极高的商业机会，进军锂、正极、负极或任何锂离子电池原材料的精炼领域绝对是万无一失的选择，他直言：“这是个毋庸置疑的明智之选。”（“It is a no-brainer.”）
Drew Baglino 则从电网运营的底层机制解释了特斯拉的核心杠杆：推动可再生能源落地的最有效途径就是大幅降低固定式储能（Stationary storage）的成本，并让电动汽车在电网波谷期充电，为电网提供柔性负荷（Flexible load）。Baglino 以得克萨斯州为例指出，得州夜间风力极为充沛，但由于核电等基荷电源无法随意关停，弃风现象十分严重。只要特斯拉能够提供极具成本优势的储能方案并引导合理用电，可再生能源资产的经济价值就会倍增，进而从根本上撬动整个能源产业的扩张速度。
```mermaid
flowchart LR
accTitle: Renewable Energy Acceleration Architecture
accDescr: Shows how Tesla's cost reductions in stationary storage and EV flexible loading enhance renewable value and speed deployment.
subgraph Inputs["核心举措"]
direction TB
A["大幅降低固定式储能成本"]
B["电动车低谷充电提供柔性负荷"]
end
subgraph Grid["电网吸收能力"]
direction TB
C["消纳夜间富余风电/光电"]
D["避免传统基荷电厂受冲击"]
end
subgraph Impact["市场放大效应"]
direction TB
E["提高可再生能源资产价值"]
F["全面加速全球绿电装机部署"]
end
A --> C
B --> C
C --> D
D --> E
E --> F
```
---
### [03:13:46](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11626s) 双向充电路线图与车型阵容精简哲学
Jefferies 分析师 Philippe Houchois 提出了两个针对性问题：第一，汽车同行都在试图效仿特斯拉，但传统车企喜欢铺设繁杂的车型，而特斯拉计划在 2030 年实现 2000 万辆的年交付目标，究竟需要多少款车型？是否担心路上到处都是相同的特斯拉导致消费者审美疲劳？第二，针对 Master Plan 3 强调的能源高效利用，特斯拉过去对双向充电（Bidirectional charging，即 V2G/V2H）态度谨慎，如今有何最新规划？
关于双向充电，Drew Baglino 解释称特斯拉过去不做并非刻意排斥，仅仅是因为当时排不上优先级。目前特斯拉正在对整车电力电子架构进行改版升级，在进一步降低电力电子元器件成本的同时，直接集成双向充电功能，这项功能预计将在未来两年内逐步普及到特斯拉所有量产车型中。但 Musk 随即提醒大家不要对双向充电抱有不切实际的幻想。Musk 坦言，绝大多数人不会频繁使用车载电池给房屋反向供电，除非家里安装了 Powerwall 作为缓冲，因为“如果你拔掉车子的充电插头，你的整栋房子就会陷入黑暗，这极其不便。”（“...because if you unplug your car, your house goes dark. And this is extremely inconvenient.”）Musk 强调，双向充电最大的价值依然在于特定时段的电网充放补充，而非替代家用专用储能。
谈到车型总数与 2000 万辆交付目标时，Musk 明确表示特斯拉不会走传统车企堆砌型号的老路，未来的车型数量会非常少（“也许 10 款左右？我不知道，总之非常少”）。Musk 尖锐地批评了传统燃油车行业：“传统汽车制造所发生的情况是，大家已经想不出什么新花样了。所以一旦无事可做，他们就只能重新洗牌，造出来的东西基本千篇一律。”（“I mean what's happened with conventional cars is people have run out of things to do. So, they run out of things to do, they just end up reshuffling the deck and you have pretty much the same...”）Musk 将汽车的发展规律与智能手机类比：过去市场上充斥着数百种各色各样的翻盖手机，如今全部收敛到了形态高度统一的智能大屏手机，智能汽车产业也必然会经历相同的收敛演进。
- 双向充电落地计划：未来两年内伴随电力电子架构降本升级，双向功能将全系标配，但家庭端体验必须结合 Powerwall 才能保障不断电。
- 车型矩阵策略：拒绝为差异化而差异化，用极少数（约 10 款以内）高生产力平台支撑超大规模量产。
---
### [03:17:32](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11852s) 中国市场拓展与成本定价策略
Canaccord Genuity 的分析师 George Gianarikas 询问特斯拉在中国市场的份额拓展规划，并特别关注中美两国之间的地缘政治紧张局势是否会对特斯拉的长期雄心造成影响。Musk 听到这个敏感话题后风趣地将问题推给了负责全球生产制造的大中华区负责人朱晓彤（Tom Zhu），调侃道：“嘿，Tom，别太紧张，别流太多汗啊！”现场引来一片哄笑。
朱晓彤从容接过了话题，并未纠缠于政治假设，而是直接用市场业绩和商业逻辑作答。他指出，特斯拉在中国的市场份额仍在强劲增长。特别是在 2023 年初特斯拉进行了积极的价格调整之后，瞬间在市场上激发了巨大的需求，这种需求甚至已经超过了工厂目前的实际交付与生产能力。朱晓彤重申了特斯拉坚守的核心商业哲学：只要你能以平易近人的实惠价格提供一款极具价值的产品，就完全不必担心需求问题。特斯拉中国团队正在全力以赴做的一件事，就是利用一切工程与供应链手段极致削减成本。
- 价格调整驱动：2023 年年初的产品价格下调迅速释放出海量购车需求，订单量超过了短期产能供应。
- 底层商业信条：坚守“以实惠的价格提供高价值产品”（Offer a product with value at an affordable price），将精力聚焦于持续降低制造成本，以确定性的产品力对冲外部环境的不确定性。
### [03:18:40](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11920s) 中国市场份额与地缘政治风险回应
特斯拉高管团队在回答关于中国市场的提问时指出，只要持续从供应链端和工厂内部提升制造效率，并将由此节省下来的价值传递给消费者，特斯拉在中国市场的份额就完全无需过度担忧。面对地缘政治相关的不确定性，发言人坦言特斯拉正在尽自身最大努力做好本地化运营。特斯拉不仅在当地社区创造了大量就业岗位，还在上游供应商体系中催生了数以万计的就业机会，为当地经济做出了实质性贡献。“只要我们在这个国家仍然被需要，我认为这方面的风险就不会太大。”（“Um I think you know, as long as we're needed in this country, I don't see there much of a risk of that.”）
### [03:19:26](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=11966s) 运营效率的度量哲学与学习迭代速度
来自 Oppenheimer 的分析师 Colin Rush 提问，对发布会上展示的各项运营效率指标印象深刻，并希望了解特斯拉内部如何衡量“学习周期的迭代速度”（velocity of learning cycles），以及在跨多领域创新时组织层面如何对此进行跟踪与思考。
对此，埃隆·马斯克（Elon Musk）从高维度的管理哲学切入进行了解答。他指出，任何事情的核心都在于建立良好的衡量标准：“如果你不去衡量某件事，你就无法去改进它，而我们在特斯拉是一群无情的度量狂人。”（“I mean, I'll say one very high-level thing, which is you can't improve something you don't measure, and we're like ruthless measurers at Tesla.”）马斯克解释道，一旦团队开始严密追踪那些真正推动运营效率提升的底层数据，就自然而然找到了一条能够采取具体行动、持续改进的清晰路径。
### [03:20:17](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=12017s) 需求的核心瓶颈：可负担性与自动驾驶的资产价值跃迁
马斯克随后将话题拓展到长期被外界误读的“需求问题”上。他强调，外界往往容易产生思维盲区，尤其是收入远超普通汽车价格的人群，习惯于从“性价比”（value for money）的角度评判产品，却忽视了普通大众面临的“可负担性”（affordability）门槛。人们拥有特斯拉的意愿极高，真正的瓶颈从来不是“想不想要”，而是“买不买得起”。“压倒性的是，人们拥有特斯拉的意愿是极高的。限制因素是他们为特斯拉付款的能力，而不是他们想不想要特斯拉。”（“The overwhelmingly the desire for people to own a Tesla is extremely high. The limiting factor is their ability to pay for a Tesla, not do they want a Tesla?”）
为证明这一点，马斯克分享了特斯拉近期关于需求价格弹性的测试结果：即使价格发生微小的调整，都会对市场需求产生巨大的拉动效应。此外，自动驾驶（Autonomy）将从根本上重塑汽车的资产属性。私家车目前的平均使用时间仅为每周 10 到 12 个小时，同时还要承担高昂的停车费用；而一旦实现完全自动驾驶，单车的周使用时长将跃升至 50 到 60 个小时，并消除大部分停车成本。马斯克直言，当车队的自动驾驶功能真正落地生效时，这很可能成为“人类历史上单夜之间发生的最大规模资产价值增值”。
### [03:22:55](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=12175s) 下一代平台车型规划与供应链的“万重挑战”
来自德意志银行（Deutsche Bank）的 Emmanuel Rosner 针对下一代车型平台提问：特斯拉展示的 PPT 中似乎出现了两款覆盖幕布的车型（一款形似厢式货车 Van，另一款类似尺寸更小的 Model 2），特斯拉在爬坡阶段的近期细分市场重点是什么？在制造成本大幅削减 50% 的情况下，如何避免新车对现有产品线造成同室操戈的蚕食？
马斯克再次重申，特斯拉的需求几乎等同于无限，降价只会让需求彻底爆发；真正的硬仗在于如何把车制造出来。汽车制造以及相伴的庞大供应链是一场难度极高的物流攻坚战。“最难的部分是制造汽车，以及与之配套的整个供应链。”（“The hard part is building the cars, by far. And the supply chain that goes with it.”）在指数级爬坡的过程中，成千上万个零件必须同步按比例扩产，任何一个环节因地震、火灾、洪水甚至政局变动而中断，整条产线就会立刻陷入“休克”。
特斯拉供应链团队与车辆工程负责人 Lars Moravy 随后补充了应对细节：
- **全流程深度掌控**：供应链管理没有侥幸，“满分才算及格”（perfection is a passing grade）。特斯拉团队深入各个层级的供应商（Tier 1 至 Tier N），在经历过关税壁垒、海运危机、芯片短缺、日本芯片晶圆厂火灾以及马来西亚疫情封控等一系列突发事件后，建立了极强的端到端应变与指导控制能力。
- **下一代平台定位**：下一代平台并非单一车型，而是承载多个细分市场的通用架构。团队的设计理念是只要产品兼具极致的吸引力与亲民的可负担性，细分市场的界限就会被打破（正如当初外界不看好轿车市场，但 Model 3 依然创造了惊人销量）。
- **物流与多供应商的现实权衡**：马斯克指出，虽然小零件可以做双供或三供（dual/triple sourcing），但对于核心大件而言，多供就像“一架装了三台引擎的飞机，只要其中一台坏了整架飞机就会坠毁”，因为产能无法无限过度配置（over-capacitize）。过度的冗余会直接推高资本开支（Capex）并导致资产闲置。
- **电池产能的超前部署**：在要解决的“上万个物流难题”中，最核心的是电池电芯（battery cell）的生产。正因如此，特斯拉在规划策略上刻意让电芯供应的产能储备超越整车制造的即时需求，以确保整车制造绝对不被电芯短缺所卡脖子。
```mermaid
flowchart TD
accTitle: 汽车制造瓶颈与扩产逻辑
accDescr: 展示了从需求、可负担性到供应链管理以及核心电芯保障的逻辑推进流程。
subgraph 市场需求侧
A["高意愿需求 (近乎无限)"] --> B["价格弹性驱动"]
B --> C["提高可负担性 (Affordability)"]
end
subgraph 制造与供应核心瓶颈
C --> D["大规模制造极限挑战"]
D --> E["10,000+ 供应链节点协同"]
E --> F["极端扰动响应<br/>(火灾/疫情/物流中断)"]
end
subgraph 产能保障策略
F --> G["深度穿透多级供应商"]
G --> H["电芯供应刻意超前配置<br/>(超过整车即时需求)"]
end
```
### [03:29:02](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=12542s) 电芯过剩缓冲与储能产品的战略扩展
讨论从电池电芯的产能调配切入。如果车辆生产出现波动或停滞，多出来的电芯该如何处理？特斯拉给出的战略答案是利用固定式储能产品进行对冲。相比汽车产线极其繁复的制造流程，Powerwall 和 Megapack 的产能弹性极大，能够更容易地调高或调低生产节奏。
Elon Musk 指出，在电芯和电池包领域进行超额产能布局（over capacitize），并将多余产能导向固定式储能系统，具有极其重要的战略价值。Megapack 的资本支出（CapEx）与汽车制造相比微不足道，而其潜在市场需求几乎是无限的。只要特斯拉在经济性上能与公用事业公司（utilities）竞争，就能卖出海量的 Megapack。这一需求规模以数太瓦时（multi-terawatt hours）计，而公司目前距离每年数太瓦时的交付目标仍有漫长的路要走。
- “Yeah, I mean, the CapEx for Megapack is tiny compared to CapEx for vehicles.” —— Elon Musk
- “Quasi-infinite demand for that, really. Um or multi-ter- many terawatt hours.” —— Elon Musk
---
### [03:30:13](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=12613s) 降本战略的区域深度本地化与“复制粘贴”工厂标准化
Barclays 分析师 Dan Levy 提问：不同地区的竞争态势与成本结构差异巨大（例如上海超级工厂的成本显著优于早期 Fremont 工厂），特斯拉今天的降本方案究竟是因地制宜还是有一套放之四海皆准的全球模板？
朱晓彤（Tom Zhu）介绍了上海超级工厂的极致本地化实践。在供应链端，中国工厂的一级、二级乃至三级供应商本地化率已经超过 95%，带来了巨大的直接成本节约；在人才端，依托长三角地区的成熟高素质制造业人才储备，特斯拉中国 3 万多名员工中外籍专家（expats）不足 20 人。此外，直销模式完全省去了传统主机厂的中间商层级与大额营销广告支出，从而将节省下来的运营费用直接转化为终端产品价格优势。
Lars Moravy 补充表示，工程底层逻辑与制造创新是跨地域通用的，并不区分欧洲、亚洲还是美洲。Zach Kirkhorn 进一步阐述，面向下一代平台，特斯拉正在从过去的“每座新工厂都是上座工厂的微调迭代版”，全面转向高度统一的“复制粘贴（copy-paste）”标准化模式。过去 Model Y 工厂每建一座都要消耗海量工程设计工时，导致各工厂形态略有不同；而下一代平台追求首次即完美，通过标准化最大程度加快全球产能扩张步伐。
- “Like in China, 95% over 95% of our supply chain are localized.” —— 朱晓彤（Tom Zhu）
- “Among over 30,000 of employees in China, we have probably um less than 20 experts.” —— 朱晓彤（Tom Zhu）
- “And as we move towards the next generation platform, uh, I think the the term you used, Lars, is copy-paste.” —— Zach Kirkhorn
---
### [03:34:13](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=12853s) 长期财务利润、2000 万辆 CapEx 测算与 FSD 自动标注技术
分析师 Omer Rokah 提出两个核心问题：一是针对 Zach Kirkhorn，询问长期营业利润率（Operating Margins）走势以及展示的 1500 亿至 1750 亿美元 CapEx 是否构成截至 2030 年的正式资本支出指引；二是针对 Ashok Elluswamy，询问“多行程重建（multi-trip reconstruction）”与 Mobileye 的 REM 高精建图有何差异，并请求介绍 AP4（Hardware 4.0）硬件进展。
Zach Kirkhorn 解释，公司致力于让营业费用占收入的比重持续下降，硬件端通过不断剔除成本来维持健康的毛利率水平。若再加上全自动驾驶（FSD）软件的赋能，整体商业模式与营业利润率将产生戏剧性的跨越式提升。关于 1500 亿至 1750 亿美元的 CapEx，这并非死板的财务指引，而是向市场透明展示内部模型的测算逻辑：根据特斯拉充沛的自身造血能力，支撑年产 2000 万辆电动车与 1 TWh 储能系统的资本投入是完全可行的，该数字未来会随上下游垂直整合（例如采矿决策）和工程效率提升而动态调整。
Ashok Elluswamy 随后针对算法进行技术解耦。他澄清，特斯拉的“多行程重建”根本目的不是像 Mobileye 那样制作自动驾驶依赖的 HD Maps（高精地图），而是用于数据管线的自动化精准标注（auto labeling）。通过车队在同一区域多次行驶的视频切片，系统能够自动生成该区域的高精度 3D 几何与语义真值，从而给神经网络提供监督信号，支撑无图自适应的规模化自驾系统。
紧接着，Elon Musk 强调了特斯拉在 AI 算力底座上的断层领先。特斯拉拥有全球顶尖的神经网络训练集群，预计通过英伟达 GPU 与自主研发的 Dojo 超算相结合，其训练算力到今年底将提升整整一个数量级（10 倍），到明年度还会再跃升一个数量级。Musk 评价道，外界大多数人尚未真正意识到这个算力规模究竟有多么庞大。
```mermaid
flowchart TD
accTitle: Tesla Fleet Auto-Labeling Architecture
accDescr: Diagram showing how fleet multi-trip video feeds local 3D reconstruction to auto-label data for scalable FSD models.
subgraph Collection["车队数据采集 Fleet Data"]
A["大规模车队常规行驶多行程片段 (Multi-Trip Video)"]
end
subgraph AutoLabeling["自动化数据闭环 Auto-Labeling Pipeline"]
B["多视角/多行程局部 3D 重建 (Local Area Reconstruction)"]
C["生成高精度 3D 空间真值标签 (Precise 3D Supervision)"]
end
subgraph Training["模型迭代 Model Training"]
D["纯视觉无图自适应自动驾驶模型 (Scalable FSD System)"]
E["Dojo & Nvidia 训练算力集群 (每年跨越一个数量级)"]
end
A --> B
B --> C
C --> D
E --> D
```
- “And in that space, you know, profitability, operating margins would be impacted dramatically.” —— Zach Kirkhorn
- “Since we want to build a scalable self-driving system, we don't want to really rely on HD maps or anything.” —— Ashok Elluswamy
- “We expect to increase that capability by an order of magnitude by the end of this year. Um and probably another order of magnitude uh by the end of next year.” —— Elon Musk
---
### [03:37:31](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=13051s) 4680 干法电极制造进展
Piper 的 Alex Potter 提问，他在工厂参观中亲眼看到干法电极（Dry Battery Electrode, DBE）产线正在运转且绝非“科学实验”，希望 Drew Baglino 披露干法电极目前的良率（yield）、产线节拍速度以及相对于最初预期的推进状况。
Drew Baglino 证实，这已经是一座以全自动化模式大规模生产干法电极的真实生产工厂。团队是极致的完美主义者，并且拥有明确的终极目标。目前产线每周都在取得切实的量化进展，不论是机台运转速度，还是干法工艺本身及下游装配流程的成品良率，推进速度都未曾停滞，正极与负极两端均在同步稳步突破。
- “Every week that goes by making progress towards those end goals, whether it's speed of the tool, yield of the of that process or the downstream process.” —— Drew Baglino
- “We haven't stalled out yet on the rate of progress either. And that's both on the anode and the cathode side.” —— Drew Baglino
### [03:39:14](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=13154s) 干法电极工程化攻坚与 4680 电池规模化量产
特斯拉团队深入探讨了 4680 电池干法电极（dry electrode）工艺的量产推进情况。团队指出，当前的制造策略让特斯拉拥有边推进边试验的灵活性，而不是被一年半前锁定的技术路线所束缚。工程师引用了 Elon Musk 经常告诫团队的一句核心原则：“It's okay to scrap equipment or money. It's not okay to scrap time.”（报废设备或浪费资金是可以接受的，但浪费时间绝对不行）。团队从概率角度出发，选择最具成功希望的路径，甚至在得州超级工厂内同时运行了两条略有差异的负极生产线，在粉末进入设备的最后工序环节展开内部良率与性能的“赛马式”竞争，加速技术迭代。
Elon Musk 强调，干法电极是极具挑战性的技术难关。特斯拉当初收购 Maxwell 主要是为了获取干法电极技术，但这凸显了实验室小规模验证与工业级规模化量产之间的巨大鸿沟。特斯拉极具才华的工程团队长期以来在干法工艺的可靠性与一致性上“无论在物理层面还是隐喻层面都在拼命磨砺打拼（grinding hard literally and figuratively）”，并预计年内有望实现规模化放量。
工程师进一步披露，产线产能正以每周环比增长的节奏推进，内部目标约为每季度每周净增约 1,000 套电池包的产能。工厂现场展示的单机设备吞吐量高达每小时处理约一吨物料。这与实验室阶段完全不同，只要有 0.1% 的极细粉尘逃逸到机柜内，就会引发严重的粉尘堆积并导致电气短路。在数月处理数千吨物料的过程中，团队不断遇到全新的失效模式并逐个排查解决。
```mermaid
flowchart TD
accTitle: Dry Electrode Scaling Challenge
accDescr: 实验室验证与工业级干法电极规模化量产之间的技术鸿沟与推进逻辑
subgraph LAB["实验室小规模验证"]
A["小规模干法粉末成膜"]
B["微量粉尘无系统影响"]
end
subgraph SCALE["工业级规模化挑战"]
C["吞吐量达 1 吨/小时/设备"]
D["0.1% 粉尘逸出即可引发电路短路"]
E["数千吨级长期运行暴露全新失效模式"]
end
subgraph DEV["特斯拉工程迭代机制"]
F["并行两条负极工艺线内部竞争"]
G["宁可废弃设备与资金，不可浪费时间"]
H["以每周/季度增产目标稳步冲压推进"]
end
LAB -->|"放大倍数引发质变"| SCALE
SCALE -->|"快速试错排查"| DEV
```
- 产能增长内部目标：计划每个季度实现每周约增加 1,000 台车辆所需电池（1K a week per quarter）。
- 设备吞吐量指标：单台加工设备每小时处理约 1 吨原材料粉末。
- 规模化公差挑战：设备机罩内哪怕仅有 0.1% 的微粉（fines）泄漏，都会引发粉尘电气短路问题。
---
### [03:41:56](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=13316s) 下一代车型提问被拒与发布会计划
分析师 Chris 针对特斯拉下一代车型（Next-Gen Vehicle）提出了三个连环问题：第一，投资者何时能看到原型车；第二，能否披露关于车身尺寸、配置或性能指标的细节；第三，发布会上提及下一代车型除了墨西哥新工厂外还将在其他工厂生产，这是否意味着该车型会在墨西哥超级工厂完工前在现有工厂率先投产。
Elon Musk 礼貌地拒绝直接回答这些问题。他表示如果现在披露相关细节就属于“抢跑”，特斯拉未来会为此举办一场专门的正式产品发布会：“We will have a proper sort of product event, but uh that we would be jumping the gun if we were to answer your questions.”（我们会举办一场正式的产品发布活动，但如果现在回答你的问题，我们就抢跑了）。随后他将提问机会交给了现场的其他参会者。
---
### [03:42:54](https://www.youtube.com/watch?v=Hl1zEzVUV7w&t=13374s) 生成式 AI 的制造局限、AGI 担忧与车端神经网络架构
现场参会者提问 Elon Musk 在企业规模不断翻倍的过程中学到了哪些大型组织管理经验，以及如何看待近期爆发的生成式 AI（Generative AI）技术对造车工艺的潜在提效作用。Elon Musk 迅速打消了生成式 AI 辅助造车的设想，直言：“I don't see AI helping us make cars anytime soon.”（我认为短期内 AI 根本无法帮我们造车），并开玩笑表示如果到了 AI 能替人类造车的地步，大家其实都不用工作了，不如全去歇着。
随后，Elon Musk 表达了对尖端人工智能发展失控的深切担忧。他认为 AI 是一项极其危险的技术，社会需要建立某种监管机构来监督 AI 研发，确保其符合公众利益，并坦言自己过去在加速 AI 发展方面所做的一些事情让他心怀不安。相比于大语言模型，Elon Musk 认为特斯拉在自动驾驶领域的 AI 才是切实现实的落地。虽然他不认为自动驾驶完全等于 AGI（通用人工智能），但道路交通系统原本就是专为人类肉眼和生物神经网络设计的，因此最直接的物理对应物就是车载摄像头与数字人工神经网络。
特斯拉自动驾驶团队的技术主管随后参与讨论，指出如果打开特斯拉车端的神经网络架构图，其复杂度到了令人难以置信的程度，甚至“nets upon nets upon nets upon nets（多层网络层层嵌套叠加）”，以至于专门的可视化软件在加载整个网络图时都会直接卡死崩溃。Elon Musk 由此感叹，人类大脑在开车时内部进行的生物运算很可能也是这般狂野复杂。最后，Elon Musk 坦言 AI 话题确实让他精神倍感压力（“This one stresses me out”）。随着发布会时间来到晚上 7 点，Elon Musk 宣布 2023 年投资者日活动圆满结束并致谢离场。
- AI 制造应用判断：Elon Musk 明确认为短期内生成式 AI 无法应用于汽车制造流程。
- AI 监管主张：呼吁设立监管机构监督 AI 技术的安全性与公众利益。
- 车载神经网络技术细节：车端运行着极其复杂的深层级联神经网络，其架构庞大到内部专用可视化工具打开时会直接崩溃。
## Video description
We plan to host Tesla's 2023 Investor Day on March 1, 2023. Our investors will be able to see our most advanced production line as well as discuss long term expansion plans, generation 3 platform, capital allocation and other subjects with our leadership team.
Learn more: https://ir.tesla.com/press-release/tesla-announces-date-2023-investor-day
0:00:00 - Preshow
0:11:15 - Intro
Part 1 | Master Plan 3
0:13:24 - Master Plan 3
Part 2 | Executing Master Plan 3
0:40:21 - Vehicle Design with Franz Von Holzhausen & Lars Moravy
0:52:43 - Powertrain with Colin Campbell
1:01:17 - Electronic Architecture with Pete Bannon
1:09:53 - Software with David Lau
1:20:18 - Full Self-Driving with Ashok Elluswamy
1:30:12 - Bot Update with Elon Musk and Ashok Elluswamy
1:35:35 - Charging with Rebecca Tinucci
1:44:04 - Supply Chain with Karn Budhiraj & Roshan Thomas
2:05:05 - Manufacturing with Tom Zhu & Drew Baglino
2:18:15 - Energy with Drew Baglino & Mike Snyder
Part 3 | Impact at Tesla
2:31:15 - Impact with Laurie Shelby and Brandon Ehrheart
2:37:50 - Financials with Zach Kirkhorn
Q&A
3:03:24 - New Gigafactory Location Announcement
3:05:20 - Q&A
[Source](https://www.youtube.com/watch?v=Hl1zEzVUV7w)
Original transcript
[1:59] Mhm.
[3:54] Mhm.
[5:40] [music]
[5:49] Mhm.
[5:50] [music]
[7:44] Woo!
[9:39] Mhm.
[11:18] Good afternoon, everyone.
[11:21] Good afternoon.
[11:25] My name is Zach Kirkhorn. I lead finance
[11:27] here at Tesla, and welcome to our
[11:29] investor day.
[11:32] [applause]
[11:38] For those of you here in the audience at
[11:40] our global headquarters in Austin,
[11:41] Texas, we welcome you. Thank you for
[11:44] being here and thank you for traveling
[11:45] in.
[11:46] For those joining us virtually, thank
[11:48] you for being a part of the day today.
[11:51] So, as we reflect back on the history of
[11:53] the company,
[11:54] there's been distinct phases of product
[11:57] advancement, technology innovation, and
[12:00] rapid volume growth.
[12:02] The most recent of which has been the
[12:04] global expansion and localization of the
[12:07] Model Y program.
[12:09] And so, today we want to talk about the
[12:11] future.
[12:12] We don't want to talk about this quarter
[12:14] or next quarter. We want to go further
[12:16] out into the future.
[12:18] And we've divided today's presentation
[12:21] into three parts.
[12:22] The first of which we're going to go
[12:24] macro.
[12:26] What does it take to convert Earth to
[12:29] sustainable energy generation and use?
[12:33] The second, we want to talk about
[12:35] Tesla's contribution to that global
[12:37] need.
[12:39] And we're going to go function by
[12:40] function
[12:41] through the company, and you'll meet our
[12:44] entire leadership team, and we're going
[12:46] to get into the details of what those
[12:48] teams are doing
[12:49] as part of the broader goal.
[12:52] And then in the third part,
[12:54] we're going to bring it back up and talk
[12:56] about what this all means for the
[12:58] company as a whole.
[13:02] So, before we get started, statements
[13:04] made in this presentation are
[13:05] forward-looking statements that are
[13:07] subject to risks and uncertainties.
[13:10] More details can be found in our written
[13:12] materials.
[13:14] So, with that, let's get started.
[13:16] Part one, Elon Musk and Drew Baglino.
[13:21] [applause]
[13:26] [applause]
[13:31] All right.
[13:33] Master Plan part three.
[13:35] So,
[13:36] as uh
[13:39] Zack was mentioning,
[13:40] the the thing that I think is we wanted
[13:42] to convey probably more more importantly
[13:45] than anything else that we talked about
[13:47] here is that there is a clear path to a
[13:51] sustainable energy Earth. It's not
[13:54] um it doesn't require
[13:57] destroying
[13:59] uh natural habitats.
[14:01] Uh it doesn't uh
[14:02] require
[14:06] us to be austere and stop using
[14:08] electricity and sort of be in the cold
[14:10] or anything.
[14:11] Um
[14:12] the this the story and I think that this
[14:14] holds together quite well and we'll be
[14:16] actually publishing a detailed white
[14:18] paper with all of our assumptions and
[14:19] calculations is that there is a there is
[14:22] a clear path to a fully sustainable
[14:24] Earth uh with abundance. In fact, you
[14:26] could support a civilization much bigger
[14:29] than Earth that than than much more than
[14:31] the the 8 billion humans
[14:33] uh could actually be
[14:35] uh
[14:36] supported sustainably on Earth.
[14:38] And I'm
[14:39] I'm just often shocked and surprised by
[14:42] how few people realize this.
[14:45] Um
[14:48] most of the smart people I know actually
[14:51] don't see a
[14:52] this clear path. They think that um
[14:57] there's there's not a path to a
[14:59] sustainable energy future or at least
[15:01] there's not one that
[15:03] uh is sustainable at our current
[15:04] population
[15:06] um or that we would have to
[15:08] resort to extreme measures.
[15:10] None of this is true.
[15:12] So, we're going to walk through the the
[15:14] calculations for how to create a
[15:17] sustainable energy civilization.
[15:20] Yeah.
[15:23] And to set the stage, today our energy
[15:25] economy
[15:26] it's let's be honest, it's dirty and
[15:28] it's wasteful.
[15:29] Over 80% of global energy, primary
[15:31] energy comes from fossil fuels and only
[15:33] 1/3 of that global energy actually ends
[15:36] up delivering useful work or heat.
[15:38] This is the problem statement, but we're
[15:40] here to talk about the solution.
[15:42] Yeah, it's it's like if for
[15:44] [clears throat]
[15:44] I I'm some of this I'm going to
[15:45] elaborate because there's there's
[15:46] there's a very wide range
[15:49] of technical expertise out there from
[15:51] people who are like
[15:53] you know, whatever level nine wizards in
[15:55] this subject to people who do not do
[15:58] engineering at all.
[16:00] So,
[16:01] I like what what if you have a gasoline
[16:02] car,
[16:04] you're
[16:05] you're you're converting less than a
[16:07] third, often maybe only 25% of the
[16:11] energy in the gasoline is converted into
[16:13] motion. The rest is turned into waste
[16:14] heat.
[16:15] That does no doesn't do any good at all.
[16:17] And there's a lot of energy required
[16:19] even to get the oil out of the ground,
[16:21] to refine the oil,
[16:22] and to transport the gasoline to the gas
[16:24] station.
[16:25] So, when you when you look at all that
[16:27] for a typical
[16:29] gasoline car is is actually going to be
[16:32] using less than 20% fully considered of
[16:35] the
[16:36] energy from the oil actually goes into
[16:38] motion.
[16:40] So, this is a when when I see people or
[16:43] when we see people doing calculations
[16:45] for what does it take to create a
[16:46] sustainable energy earth, they assume
[16:48] that the same energy amount is required
[16:51] for an elec for an electric
[16:53] electrified civilization versus a
[16:55] combustion civilization. This is not
[16:57] true that because uh most of the energy
[17:02] of combustion is waste heat.
[17:05] And even to get the fuel to combust in
[17:07] the first place and get it to the end
[17:08] use, there's a lot lost along the way. I
[17:10] mean, this is the primary energy
[17:12] consumption, 165 petawatt hours a year.
[17:14] Petawatt hour is a trillion terawatt
[17:17] trillion kilowatt hours. So, it's a
[17:18] large amount of energy, but the nice
[17:20] thing about electrified economy
[17:22] it
[17:23] there's a better way. We're going to
[17:24] talk about it. Is that
[17:27] through end use efficiency and through
[17:28] efficiency along every step of the way,
[17:31] actually the total energy use
[17:33] halves.
[17:34] So, this is one of the most enabling
[17:36] aspects of electrifying everything is
[17:40] that the sustainable energy economy is
[17:41] that much easier to accomplish. It's
[17:43] actually half the problem statement of
[17:45] the fossil fuel economy.
[17:46] Yeah, and we're being conservative here.
[17:48] So, it could be
[17:49] better than half, but we're trying to
[17:52] have assumptions that are reasonable,
[17:53] not overly optimistic, in fact slightly
[17:55] pessimistic. So, it's really better than
[17:58] half, but just say for it's it's it's
[18:00] easy to make the argument that we'll
[18:02] need half as much energy with an
[18:04] electric economy versus a a combustion
[18:06] economy.
[18:07] Yep.
[18:08] Um
[18:10] So, how the master plan works? You want
[18:12] to talk about that?
[18:13] Um so,
[18:15] the
[18:17] the the the thing that is needed is at
[18:20] very large scale that is not currently
[18:21] present is a vast amount of battery
[18:24] energy storage.
[18:26] Uh
[18:26] our rough calculations are
[18:29] that this is about 240 terawatt hours or
[18:32] 240,000 gigawatt hours.
[18:34] Um
[18:35] This is a lot of batteries, but it is
[18:37] actually very achievable amount. Um
[18:41] We'll go into details on that.
[18:42] So,
[18:43] that's a combination of electric
[18:45] vehicles and stationary storage. So, if
[18:48] you've got solar or wind, you've got to
[18:49] store the energy when the wind is not
[18:51] blowing or the sun is not shining.
[18:53] Um and so, we're assuming sort of an 8:1
[18:56] ratio of
[18:58] uh
[18:58] stored energy to power. So, 30 terawatt
[19:01] hours of power that 30 30 30 terawatts
[19:04] of power.
[19:06] Um
[19:06] Our actual uh
[19:09] capital expenditure calculation for
[19:12] manufacturing investment is more like uh
[19:14] 6 trillion, but we you know, we made it
[19:18] higher to make it 10 trillion. And this
[19:21] is
[19:21] across mining, refining, you know,
[19:24] battery factories, recycling, vehicle
[19:25] factories, all the things that we're
[19:27] going to talk about needing to invest in
[19:29] to build the sustainable energy economy.
[19:31] Yeah. Now,
[19:32] if you look at the total world economy,
[19:35] it's just under 100 trillion. So, if
[19:37] this was spread out, say over 10 years,
[19:40] it would be 1% of the global economy.
[19:43] Over 20 years, uh it would be half a
[19:45] percent of the global economy. So, this
[19:47] is uh yeah.
[19:49] Not a big number relative to the global
[19:52] economy.
[19:53] Um as Drew mentioned, you need about
[19:55] half as much energy with an electric
[19:57] economy versus a combustion economy.
[19:59] And in terms of wind and solar, how much
[20:02] land would be used? It's less than 0.2%
[20:06] of the land area of Earth.
[20:08] Um
[20:10] Like, generally people don't realize
[20:11] quite how much energy is reaching us
[20:13] from the sun.
[20:14] Um it's roughly a gigawatt per square
[20:17] kilometer.
[20:18] Um
[20:19] and you know, the sun doesn't shine all
[20:21] the time, but it's uh if you multiply
[20:23] that by
[20:24] say uh
[20:27] four to get the continuous power, four
[20:30] or five, uh then that that that gives
[20:33] you the land area of solar. And you can
[20:34] put wind and solar often in the same
[20:36] place. So, a lot of places that
[20:37] currently have wind, you could have
[20:39] solar there and you double your energy.
[20:42] You can also put wind offshore. It
[20:43] doesn't even need to be on land, so wind
[20:45] is even more flexible.
[20:46] You could put solar offshore, too.
[20:48] Yep.
[20:48] So, Earth is 70% water.
[20:51] Um anyway, the point is that um
[20:53] with a with a pretty really a remarkably
[20:55] small amount of of Earth's land area, we
[20:58] can go fully sustainable.
[21:00] Um
[21:02] Yeah.
[21:03] And and and from a do Do resources and
[21:05] raw materials exist to support this
[21:07] transition?
[21:08] We'll go through that in detail, but we
[21:10] do not see any insurmountable resource
[21:12] challenges at all. In fact, in the end
[21:14] we should be
[21:15] mining less ore to accomplish this
[21:17] economy than we currently do with the
[21:18] fossil fuel economy. And we're going to
[21:19] talk through that.
[21:20] Yeah, just to emphasize that again, the
[21:23] electrified economy will require less
[21:26] mining than the current economy does.
[21:28] Yes.
[21:29] Less.
[21:30] Not more.
[21:31] Okay. Um so, this is the plan and now
[21:34] we'll get into a little more of the
[21:35] details of the plan.
[21:37] Basically, five areas of work.
[21:40] Um
[21:40] First, re-power the existing grid.
[21:43] Second, switch to to electric vehicles.
[21:46] Third, switch homes, businesses, and
[21:48] industry heating to heat pumps.
[21:51] Uh fourth, high-temp heat delivery
[21:54] and storage for high-temp industrial and
[21:57] chemical processes. And a little bit of
[21:59] green hydrogen in there for chemical
[22:01] processes that need hydrogen.
[22:02] Um And finally, sustainable sustainably
[22:05] fuel planes and boats. These are the
[22:07] five areas and we're going to go into
[22:08] detail on all of them.
[22:09] Yeah, I mean my personal opinion is that
[22:11] as we improve the energy density of
[22:13] batteries, you'll see all transportation
[22:17] uh go fully electric um with the
[22:20] exception of rockets. That's awkward. Um
[22:24] but uh
[22:25] but you can make the the fuel with uh
[22:28] CO2 and water. So, you can make methane
[22:30] with CO2 and water. So, in fact,
[22:32] And you can do that with just
[22:32] electricity.
[22:33] Yes, exactly. So, uh
[22:36] so, it's
[22:37] in fact, on Mars, if we hopefully get
[22:39] there at some point, um
[22:40] the atmosphere is CO2 and there's water
[22:42] ice uh throughout Mars. You can take the
[22:45] CO2 and H2O and turn that into CH4,
[22:49] which is methane, and oxygen.
[22:51] So, ultimately, even rockets uh can be
[22:53] electrified.
[22:56] So, first, re-powering the existing grid
[22:58] with renewables. And this is going to be
[23:00] a consistent theme. You'll see our
[23:02] estimates for the number of terawatt
[23:03] hours, terawatts, and trillions of
[23:05] investment at the bottom of the page.
[23:07] You know, this is already actively
[23:10] occurring in front of us. 60% of the
[23:12] generation added to the US grid was
[23:13] solar in 2022. And actually, on a
[23:16] year-on-year basis, solar deployment is
[23:18] growing 50% year-on-year
[23:21] as of 2022. So, this is a this is a
[23:23] serious
[23:26] upswing, and if we continue this trend,
[23:28] this is going to be behind us before we
[23:30] even know it.
[23:30] Yeah.
[23:31] Um
[23:32] Second, switching to [clears throat]
[23:33] electric vehicles. Again, 21% reduction
[23:37] in fossil fuel use by doing this alone.
[23:39] Obviously, Tesla is heavily engaged in
[23:42] this activity as along with many others.
[23:44] Overall, EV production grew 59%
[23:47] year-on-year in 2022, and EVs hit an
[23:49] amazing 10% market share. I mean, it's
[23:52] an
[23:52] awesome milestone. I've I'm super
[23:55] excited to see that. I've got to
[23:57] Yeah. I mean,
[23:58] this is uh
[24:01] obviously happening very rapidly, um
[24:04] and I mean, I think
[24:06] real really all cars will go to fully
[24:08] electric and autonomous. Um
[24:12] And so,
[24:14] writing a non-autonomous gasoline car is
[24:17] going to be analogous to riding a horse
[24:19] and using a flip phone. That's basically
[24:22] the going to be the situation.
[24:23] And we actually took a somewhat
[24:24] conservative assumption here in terms of
[24:26] how many batteries are required cuz the
[24:28] more the fleet is autonomous, the the
[24:30] fewer the the smaller the fleet needs to
[24:32] be just from a utility basis. So, we're
[24:35] not accounting for all of those benefits
[24:36] or really much of those benefits at all
[24:38] in this number. Um and what does this
[24:40] fleet look like? You know, just rough
[24:43] view from our perspective. Of course, we
[24:45] could be wrong, but you know, uh you can
[24:47] see the sort of breakdown of the fleet
[24:49] by millions of vehicles. Um you know,
[24:51] our goal is to do 20 million electric
[24:54] vehicles a year.
[24:55] Yeah. You
[24:57] Fewer vehicles will be needed, at least
[24:58] passenger vehicles,
[25:00] with autonomy. So,
[25:03] um
[25:04] there's some debate as to what that
[25:06] number is, but it's it's a number less
[25:08] than the number of vehicles needed
[25:09] today. There's roughly 2 billion cars
[25:11] and trucks
[25:13] uh
[25:14] in operation in the world today.
[25:16] And
[25:16] Yeah, so what we show here is actually,
[25:17] I think, only 1.4 million or or so.
[25:20] So, we're we're we're represent 1.4
[25:22] billion, I mean, or so. So, a smaller
[25:24] fleet. And you know, the the numbers are
[25:26] here in this presentation are around 85
[25:28] million vehicles a year produced.
[25:31] Just to give you a sense of how we're
[25:32] thinking about this. Again, we're going
[25:33] to put all these assumptions up online
[25:35] and you know, encourage people's
[25:36] thoughts.
[25:37] Yeah.
[25:38] So,
[25:41] Yeah, so we're we're basically heading
[25:42] rapidly towards an electric or
[25:44] autonomous future.
[25:46] Exciting.
[25:46] Yeah.
[25:47] Um
[25:48] and one of the reasons why EVs are so
[25:49] enabling is this end-use efficiency
[25:52] point.
[25:53] Um
[25:54] Te- Tesla
[25:55] Model 3, it's four times more efficient
[25:57] from the well to wheel than uh a Toyota
[26:00] Corolla. And that's all about the
[26:01] efficiency of getting the electricity to
[26:03] the into the car in a sustainable energy
[26:06] economy, and then how efficient the car
[26:07] is uh in transferring that stored uh
[26:10] energy to motion on the road. Um when
[26:13] compared to the engine in the Toyota
[26:14] Corolla and all the, you know,
[26:16] extraction, refining, transmission,
[26:18] distribution of the gasoline to the
[26:19] Toyota Corolla. And just for uh this is
[26:22] a fun reference, Model 3 can drive over
[26:25] a mile on the energy it takes to boil a
[26:27] pot of water for pasta. And then it can
[26:29] drive another mile on the energy it took
[26:31] to cook the pasta.
[26:33] And that pasta is 1 lb, and Model 3 is
[26:36] like 4,000 lb.
[26:37] Just to give you a sense of just like
[26:39] what like it really doesn't use a lot of
[26:41] energy to move a Model 3, that 4,000-lb
[26:43] object, down the road.
[26:45] Yeah, also heat is a lot more energy
[26:47] than motion.
[26:47] Yeah.
[26:48] Yeah, but but people, you know, you just
[26:50] boil a pot of water, you don't think
[26:51] about it. You know, it's just
[26:53] interesting how efficient these cars
[26:54] are. Um
[26:56] next, switching to heat pumps in homes,
[26:59] businesses, and industry.
[27:00] Um
[27:01] you know, right now heat pumps meet 10%
[27:03] of building heating needs. Install
[27:05] rate's growing 10% year-on-year. It
[27:06] really needs to accelerate. Heat pumps
[27:08] can serve, you know, heat applications
[27:10] up to 200°C in in businesses and
[27:12] industry. And from an investment
[27:14] perspective, as you can see on this
[27:15] page, it's actually the lowest hanging
[27:17] fruit in terms of displacing fossil
[27:19] fuels.
[27:20] Um you might be saying, "Well, what
[27:21] exactly is a heat pump?" So,
[27:24] um
[27:26] heat pumps don't move heat, don't create
[27:28] heat, they they move heat.
[27:30] Uh when you think about like the natural
[27:32] gas furnace in your house, like it's it
[27:35] is generating heat itself. But what the
[27:36] heat pump is doing is actually moving
[27:38] heat from outside of your house into
[27:39] your house. They're just an air
[27:40] conditioner or a refrigerator in
[27:42] reverse. So, we're surrounded by heat
[27:44] pumps. There's like,
[27:46] you know, they're all over this factory,
[27:47] they're in your house. Um and and all
[27:49] this really is is about bringing them to
[27:51] displace all the fossil fuel heating in
[27:53] all the homes, business, and in the
[27:54] industry that we can. And from an energy
[27:57] efficiency perspective, there's a three
[27:59] three times reduction in the total
[28:01] energy required to heat these buildings.
[28:04] So, a real obvious thing to do.
[28:06] Yeah, heat pumps.
[28:07] They're in our cars, yeah, now as
[28:10] default. And at some point,
[28:12] we might make a heat pump for our home.
[28:14] So.
[28:15] Yeah, maybe. Maybe.
[28:16] Maybe.
[28:17] Um
[28:18] ne- next, a little bit more detail on
[28:20] electrifying uh high temp uh sort of
[28:23] industrial chemical processes.
[28:25] So, over 50% of industrial heat is
[28:27] greater than 400°C. You know, cement,
[28:29] steel, fertilizers, chemicals, plastics,
[28:31] metals refining all need like 1500°C.
[28:33] So, we need a solution here. Um
[28:36] ul- ultimately, it's purpose-built
[28:37] equipment that enables electrification.
[28:40] You know, carbon graphite is stable up
[28:41] to 2000 2800°C. There's other options in
[28:44] the 1500°C range like silicon carbide,
[28:47] other other materials. So, the the idea
[28:49] here is you create and store heat when
[28:52] renewable power is available. If this is
[28:54] a sustainable energy economy, you know,
[28:56] renewable energy is intermittent. Peak
[28:58] of the day, you've got more generation
[29:00] than you need. You make a bunch of heat
[29:01] then, and then you transfer that heat
[29:03] into the industrial process 24 hours a
[29:05] day using the stored heat you created
[29:07] when the sun or the wind was blowing.
[29:09] That's the concept here.
[29:11] And then on the hydrogen side, we also
[29:13] need green hydrogen to decarbonize
[29:14] metals and chemical refining processes.
[29:16] This is things like ammonia, uh making
[29:19] steel, you know, there's roughly 120
[29:21] million tons of hydrogen sourced from
[29:23] fossil fuel today to do these things. Um
[29:26] and hydrogen can also directly replace
[29:27] coal, which is currently used in a ton
[29:30] of steel production through a process
[29:32] called directly reducing iron. You can
[29:34] replace blast furnaces with a hydrogen
[29:36] reducing uh direct reduced iron furnace.
[29:39] Um
[29:40] and this is the way to eliminate fossil
[29:42] fuels from these aspects of the economy
[29:44] and the CO2 associated with them.
[29:46] Yeah, I so I mean
[29:48] um some of this of this like room to
[29:50] disagree, but
[29:51] uh
[29:52] some amount of hydrogen is needed for
[29:54] industrial processes. My personal
[29:55] opinion is that hydrogen will not be
[29:57] used
[29:58] uh meaningfully in transport, uh but um
[30:01] and shouldn't be. Um if you're going to
[30:03] use a chemical
[30:04] uh fuel, you should uh use CH4 not H2.
[30:08] Uh
[30:09] but uh but it's nonetheless it is needed
[30:11] for industrial processes and can be
[30:13] produced uh just by just splitting water
[30:15] essentially.
[30:16] I mean, some of that's been done for
[30:17] decades and decades. This is not, you
[30:19] know, rocket technology.
[30:20] Yeah.
[30:23] Um and lastly,
[30:25] a small part of the pie, but a necessary
[30:26] part of the pie, is sustainably fueling
[30:29] planes and boats.
[30:30] Um shipping accounts for 3% of global
[30:32] CO2. It's ripe for electrification. Even
[30:35] with a lithium iron phosphate, long-haul
[30:37] ships can be fully battery powered. So,
[30:39] that's a a great opportunity to uh
[30:42] electrify um energy density is a little
[30:44] bit harder for planes, but short haul is
[30:46] doable today with some improvements.
[30:47] We'll get long haul underway, but even
[30:50] even in the meantime, we can leverage
[30:51] sustainable aviation fuels produced and
[30:53] stored using excess renewable
[30:54] electricity. There's a lot of work going
[30:56] on in this space
[30:58] and it's it's it's not
[30:59] Yeah. I mean, it it to to really get
[31:02] long-range aircraft and
[31:05] long-range shipping to use uh
[31:08] lithium ion
[31:10] uh you need to redesign the ship and not
[31:12] just um
[31:13] Or the plane.
[31:14] And the plane. Uh
[31:16] to take advantage of the fact that it is
[31:18] a new
[31:19] uh
[31:20] source of uh
[31:22] energy. It's It's a It's a different
[31:23] architecture. So,
[31:26] it it just like with an electric car,
[31:27] you wouldn't just, you know, take a
[31:29] gasoline car and stick a battery in it.
[31:31] That's very suboptimal. Um it's much
[31:35] more efficient to have the battery be
[31:37] the structure of the car.
[31:39] Um
[31:40] and uh
[31:41] you know,
[31:43] make it as
[31:44] make it mass efficient and optimized for
[31:45] battery for batteries. The same If if
[31:47] that's done with the aircraft, uh then
[31:49] you can get long-range aircraft
[31:51] uh at around
[31:52] with with cells at around 450 watt hours
[31:54] per kilogram, which you can buy it right
[31:57] now, actually. They're expensive, but I
[31:58] think uh that that price will come down.
[32:02] So, when we stack up all of these
[32:03] efforts,
[32:05] uh we end up with the numbers we shared
[32:06] at the beginning of the presentation. 30
[32:08] terawatts, 240 terawatt hours, 10
[32:10] trillion dollars. And you're You may be
[32:13] saying like, I need some context. Is
[32:15] this feasible?
[32:18] Spoiler alert, it's entirely feasible.
[32:21] Um just looking at it from a growth rate
[32:23] growth rate perspective, how much do we
[32:25] need to grow the deployment of these
[32:26] technologies? We're talking about only a
[32:29] three X growth rate in solar and wind
[32:31] deployment. Um solar is already growing
[32:34] at a a breakneck pace as is wind. This
[32:37] gap is going to close really quickly.
[32:39] When we look at the electric vehicles,
[32:42] they have to grow 11x. Well, they they
[32:44] grew 60% year-on-year last year. That
[32:46] growth rate is also going to close
[32:48] pretty pretty darn That that gap is
[32:50] going to close pretty quickly as well.
[32:51] And lastly, storage. Um
[32:54] you know, Tesla's energy storage
[32:56] business has grown at 65% CAGR since
[32:59] 2016. The global, you know, energy
[33:01] storage business is is is accelerating
[33:03] pace as well. I mean, all these gaps are
[33:05] going to close, especially as as this
[33:07] momentum of the transition to
[33:09] sustainable energy
[33:11] accelerates.
[33:12] Um and of course, our goal on this page
[33:14] is 20 million EVs per year and 1
[33:16] terawatt hour of stationary storage per
[33:18] year. Uh basically as soon as we can.
[33:23] Um
[33:24] and then what you know, what about this
[33:26] investment? How do I have a reference
[33:28] point on this investment? You know, Elon
[33:29] mentioned it's
[33:30] 10% of, you know, 1 year's world GDP.
[33:35] Another way of thinking about it is how
[33:36] does it compare to what we're investing?
[33:38] Like what we invested last year in the
[33:40] fossil fuel infrastructure.
[33:42] Um and and it's 60% of that investment.
[33:44] So, actually building the sustainable
[33:46] energy
[33:47] uh economy is is less than extending the
[33:50] fossil fuel economy from a
[33:52] year-over-year investment basis. So,
[33:55] very doable.
[33:56] Um
[33:57] when we look towards
[33:59] does this fit on the planet? Absolutely.
[34:01] Less than 0.2% of land
[34:04] as a reference point. Um the total land
[34:07] area intensively farmed today is 12.5%
[34:11] of all land.
[34:12] So, I mean, you drive around, you see
[34:14] some farms, but you don't see them
[34:15] everywhere.
[34:16] This is this is an order of magnitude
[34:18] more than order of magnitude difference
[34:19] between farming and what we're talking
[34:20] about for sustainable energy land.
[34:23] Yeah, it it and it doesn't need to
[34:24] displace farmland or
[34:27] you know, forests or
[34:29] jungle or any any kind of
[34:32] ecological preserve it can be used in in
[34:34] a very sparsely populated desert
[34:37] regions.
[34:38] Yeah.
[34:38] Barren areas,
[34:40] areas that are just not really fit for
[34:41] development or otherwise used so.
[34:43] Yeah.
[34:44] I mean 0.2% can fit into a lot of
[34:45] places.
[34:46] Yeah, it's it's this this essentially
[34:49] no meaningful uh
[34:51] ecological impact. In fact, um
[34:53] transition to a sustainable energy
[34:54] economy economy would result in a
[34:56] substantial reduction in current
[34:58] ecological impact.
[35:00] It's a great way to put it.
[35:01] Yeah.
[35:02] Um and what about on the mineral
[35:04] extraction side? So this is a cartoon
[35:06] that sort of gives you a sense for all
[35:08] the ore
[35:09] and the like extracted uh minerals that
[35:11] are coming out of the earth every year.
[35:14] It's about 68 gigatons. Um
[35:18] so each truck is a gigaton. What does
[35:20] this look like when we're in a
[35:21] sustainable energy economy?
[35:23] Looks like that.
[35:25] The fossil fuel extraction disappears.
[35:27] We replace it with the materials
[35:29] required to fulfill fulfill the
[35:30] sustainable energy economy. Actually
[35:32] reduces.
[35:33] Now it's not to say that we don't need
[35:35] to continue to explore, you know, bring
[35:37] on mining and refining for the
[35:40] sort of specific materials for the
[35:42] sustainable energy economy. We do, but
[35:44] the investment and mass flows are all
[35:45] very achievable just looking at what is
[35:47] already happening on the planet. Like
[35:48] this is nothing out of scale of what has
[35:50] been done and is already being done.
[35:53] Um
[35:54] and then we calculate it on a sort of,
[35:56] you know, element by element basis
[35:59] the resources are there to support the
[36:01] transition. You know, this is cumulative
[36:03] demand to move in the sustainable energy
[36:05] economy direction until 2050 relative to
[36:08] USGS resources today.
[36:10] You know, we're not breaking the
[36:12] resource bank
[36:14] for any of these materials. And then
[36:15] when we look at what really happens as
[36:17] we move forward
[36:20] history teaches the more we look the
[36:21] more we find. What people think happens
[36:23] is, "Oh, there's this many resources.
[36:25] Next year there's going to be less cuz
[36:26] we're going to extract them. What
[36:27] actually happens is as we
[36:30] uh
[36:31] extract resources, we we find more. And
[36:33] you can see on the right what has
[36:34] actually occurred with the key materials
[36:37] to the sustainable energy economy. Since
[36:38] 2000, as the sustainable energy economy
[36:40] has been growing and Tesla's been
[36:41] growing and all the industries around us
[36:43] have been growing,
[36:44] the actual resource availability has
[36:46] increased, not decreased.
[36:48] Y- Yeah, there's
[36:50] [clears throat]
[36:50] This this still seems to be quite a bit
[36:51] of confusion about the about lithium.
[36:54] Lithium is extremely common. It's one of
[36:55] the most common elements on Earth. Um
[36:58] there is there's no country that has
[36:59] monopoly on lithium or even
[37:03] close to it. Um
[37:04] there's there's enough lithium uh or in
[37:07] the United States to electrify all of
[37:09] Earth
[37:10] if if the if the United States was the
[37:11] only place producing lithium. There's
[37:13] enough domestic uh material to electrify
[37:17] uh Earth. Um it's very common. Um
[37:20] the the limit- limiting factor
[37:23] uh is the refining of the lithium into
[37:26] battery-grade lithium hydroxide or
[37:28] lithium carbonate. That's the actual
[37:29] limiting factor.
[37:30] And the same is true for these other
[37:32] materials. And And this is again, these
[37:34] are not like crazy technologies. It's
[37:35] just the investments need to make be
[37:36] made, and the investments
[37:38] they're not gigantic. They just need to
[37:39] happen.
[37:40] Right. Um nickel's maybe the the the of
[37:43] of them all the trickiest one to to
[37:45] solve. But as
[37:46] we sh- we showed with the graph there,
[37:48] maybe need like 30% of the world's
[37:49] nickel known nickel reserves. So
[37:51] And the nickel reserves have actually
[37:52] grown since 2000. So
[37:54] Uh the there's more than Yeah, exactly.
[37:56] So
[37:57] um
[37:58] and you only need nickel for uh
[38:00] basically aircraft, long-range boats, um
[38:03] and very long-range cars or trucks. Uh
[38:06] but the the vast majority of the heavy
[38:09] lifting uh for electrification will be
[38:12] uh
[38:12] iron-based uh
[38:14] uh
[38:15] cells. And the uh Earth is Iron is
[38:17] actually literally the most common
[38:18] elements on Earth. So
[38:21] uh little
[38:22] trivia point, if you say like, "What is
[38:24] Earth made of?"
[38:25] By mass, it is most it it is made of
[38:28] iron more than anything else. And
[38:29] second, oxygen. And then,
[38:32] uh,
[38:33] everything else after that. So,
[38:35] basically, we're a
[38:37] muddy rust ball is what Earth is.
[38:40] Um,
[38:41] so, an an iron cathode is sort of we're
[38:43] definitely not going to run out of iron.
[38:45] The so much iron is insane.
[38:47] Um,
[38:48] so, that's
[38:51] Yeah. So.
[38:53] Yeah, and ultimately, you know, this
[38:54] resource uh, extraction we go through
[38:58] this effort, we build these batteries,
[39:00] um, and then we recycle these batteries.
[39:02] So, ultimately, we're we're we're doing
[39:04] this to build this sustainable energy
[39:06] economy, but the maintenance amount of
[39:08] of ore that we require is is really an
[39:10] order of magnitude or more less because
[39:13] of recycling.
[39:15] Um, so, in the end,
[39:18] a sustainable energy economy is within
[39:19] our reach, and we should accelerate it.
[39:22] Yeah, I mean, so,
[39:24] this is really the the main message of
[39:26] today, um,
[39:28] and I really wanted today to be not just
[39:30] about Tesla investors who own stock, but
[39:33] really anyone who is an investor in
[39:34] Earth.
[39:35] Um, and the
[39:38] what we're trying to convey is a message
[39:40] of hope and optimism.
[39:42] Um, and and
[39:44] optimism that is based on on actual
[39:47] physics and and and real calculations.
[39:50] It's not It's not wishful thinking.
[39:52] Uh, Earth can and will move to a
[39:55] sustainable energy economy, and will do
[39:57] so in your lifetime.
[40:00] Thank you.
[40:01] Yes.
[40:03] Thank you.
[40:05] [applause]
[40:07] And, uh,
[40:08] I just want to welcome up uh, Lars and
[40:11] Franz to the stage.
[40:21] Hi.
[40:22] Hi, I'm Franz.
[40:24] I um
[40:25] lead design at Tesla.
[40:27] And I'm Lars. Uh I've been
[40:29] doing vehicles with Franz for almost 13
[40:31] years now.
[40:35] So, I joined Tesla in 2008 to vertically
[40:37] integrate design into the company. It
[40:40] didn't exist before. It was pretty small
[40:42] team. It was tasked with designing most
[40:44] beautiful, innovative,
[40:46] and well-engineered vehicles on the
[40:47] planet.
[40:48] No small task in 2008. There was not a
[40:51] lot going on in the EV sector then.
[40:53] Since then, we focused on constant
[40:55] improvement on cost, efficiency,
[40:57] innovation,
[40:58] things that you'll continue to hear
[40:59] about today,
[41:00] um while continuing to design the most
[41:03] desirable cars.
[41:06] Today, we produce cars differently than
[41:07] we did 10 years ago.
[41:09] But, the end result is always an
[41:10] exciting, futuristic, and desirable set
[41:12] of vehicles.
[41:14] Back then, we only had a handful of
[41:15] designers and engineers like myself, but
[41:17] we had a great vision to radically
[41:19] change transportation.
[41:22] So, back
[41:23] in 2008, where we designing
[41:26] Model S, we didn't have a factory. In
[41:28] fact, we had a really small engineering
[41:30] team and a tiny design team.
[41:34] But, that allowed design to lead all the
[41:37] conversations. It let us innovate
[41:39] forward-thinking ideas like, "How do you
[41:42] fit seven people into a sedan?"
[41:44] Or, "How do you make door handles
[41:46] disappear into the into the doors?"
[41:48] Or, putting a huge touchscreen into the
[41:50] center of the the vehicle, something
[41:52] that had never been done before.
[41:56] And then we won Motor Trend Car of the
[41:57] Year.
[41:57] Yes, and we won Motor Trend Car of the
[41:59] Year in 2013. Our first our first, you
[42:01] know, great award, first car. Um pretty
[42:04] good pretty good start. Kind of a home
[42:05] run, I think.
[42:07] Um but, that that whole process resulted
[42:10] in a linear process that you see on the
[42:12] screen. We We designed first, then we
[42:14] engineered, and then we figured out how
[42:16] to manufacture it.
[42:17] Yeah, I mean, I think that's really
[42:18] important. When we were designing the
[42:19] car together, we didn't even know where
[42:21] we were going to build it. And so, we
[42:22] came up with
[42:23] know, Lars, when we were designing.
[42:24] That's true.
[42:26] Uh
[42:27] we didn't So, once we got Fremont, we
[42:28] were very fortunate, and we figured out
[42:30] the manufacturing solution sort of like
[42:32] we were flying a plane and putting the
[42:33] wings and building the engine at the
[42:35] same time.
[42:36] So, we knew we had to do better.
[42:38] Yeah, we knew we had to be better in
[42:39] order to scale. And as part of the
[42:41] master plan that you've read,
[42:43] Model 3 needed to be a smaller, more
[42:46] efficient, and more affordable version
[42:48] of Model S. But it had to be equally
[42:50] great. It had to have all the things
[42:52] that people loved in their Model S or
[42:54] Model X, and but be much more
[42:56] affordable.
[42:57] And so, we we approached the process a
[43:00] little bit differently than the first
[43:01] time around. Now, we had teams that we
[43:04] all worked together. So, we were able to
[43:06] combine design, engineering, and
[43:08] manufacturing process all at the same
[43:10] time.
[43:12] But somewhere along the way,
[43:15] we changed the manufacturing process to
[43:17] be fully automated. And so, we leaned
[43:19] into this whole new way of manufacturing
[43:21] a car, but we had already engineered it.
[43:24] So, things didn't quite go as well as
[43:26] planned.
[43:27] It was an amazing product, but it landed
[43:30] us in production hell.
[43:32] Many of us who lived through that
[43:34] carry those battle scars. It was a great
[43:36] idea, but it wasn't the right timing.
[43:38] Like Franz said, automating something
[43:40] that we designed to be built manually is
[43:42] super hard. And we have many, many
[43:45] failed examples of that at the Fremont
[43:46] factory that we ripped out. But some of
[43:48] them eventually still work. This is one
[43:50] I actually worked on with a small team
[43:51] of engineers. It's still running today,
[43:53] and some of the engineers that came by
[43:55] said we couldn't do it are no longer
[43:57] with the company, but it's running, and
[43:59] it's running faster than ever.
[44:02] So, we kind of self-imposed constraints
[44:04] on the design. When we were doing it to
[44:07] be built manually, and we really didn't
[44:09] think about it.
[44:10] But despite all that, Model 3 is a
[44:12] best-selling EV ever.
[44:13] Yeah, and Model Y, which is derived from
[44:16] Model 3, is about to pass that.
[44:21] But we knew we had to improve the
[44:22] process further. And with Cybertruck,
[44:26] we designed a vehicle around a vision
[44:28] that actually started with a
[44:29] manufacturing process. And in this case,
[44:32] the materials dictated the design.
[44:35] Forming full hard stainless steel isn't
[44:37] rocket science, but it sure isn't easy,
[44:40] and it limited the way we could do it.
[44:42] Yeah, absolutely. It really forced us to
[44:44] think about designing something um
[44:46] in in in a way that you couldn't
[44:48] normally stamp panels, you couldn't form
[44:50] them in a traditional way. So you ended
[44:51] up with
[44:53] very linear um bending processes that
[44:56] are just not in automotive kind of
[44:58] language of manufacturing today.
[45:01] But it it actually created a very
[45:02] efficient and process and one of the
[45:05] most dynamic designs ever, I believe.
[45:08] It's definitely something that's going
[45:09] to change the road landscape.
[45:11] Hopefully you guys saw it down there and
[45:12] you experienced it. It's definitely
[45:14] real. Those are real trucks. We're on
[45:16] our way to build them.
[45:17] But what that stainless steel
[45:19] opportunity did for us, it is let us
[45:21] rethink the factory footprint. We don't
[45:23] stamp those. That's a huge part of it.
[45:25] We don't even paint them.
[45:26] So our footprint got smaller, and we
[45:28] started to think about innovative ways
[45:30] to take those constraints
[45:32] and make great products.
[45:34] But that constraint didn't really change
[45:36] the end result of the truck. Um it's a
[45:38] super dynamic truck and it has all the
[45:40] functionality you would expect out of
[45:43] any of the other competitive trucks.
[45:45] And the best thing about it, it's coming
[45:47] this year.
[45:51] [applause]
[45:55] So ideally,
[45:57] after all that, we would design,
[45:58] engineer, manufacture, and plan for
[46:00] automation happening together. It gives
[46:02] us the opportunity to question
[46:04] requirements. This is something that is
[46:06] fundamentally only available at Tesla.
[46:09] The places I used to work in the top
[46:10] manufacturing companies in the world,
[46:12] they don't sit together.
[46:13] Yeah, I know where I know team.
[46:15] No where I know has all these teams
[46:17] together thinking about these processes
[46:19] from the very beginning.
[46:20] In fact, all of those engineering teams,
[46:22] manufacturing, design, automation,
[46:23] they're all in one work. They all report
[46:25] to one person. We can't point fingers at
[46:27] each other. So, we have to solve them
[46:29] together, which is the best way to
[46:30] innovate.
[46:32] A traditional way of making a vehicle is
[46:34] this. You stamp it, you do a build a
[46:35] body in white, you paint it, and you do
[46:37] final assembly. And what's interesting
[46:39] is these shops are dictated by the
[46:42] the the organizational structures that
[46:44] exist.
[46:45] And they're dictated by the boundaries
[46:47] that exist in the factories that are
[46:48] laid out.
[46:49] If something goes wrong in final
[46:51] assembly, you block the whole line, and
[46:52] you end up with buffering in between.
[46:55] This is at the tail end
[46:58] of its manufacturing optimization. Henry
[47:01] Ford first invented this assembly line
[47:02] in 1922. It's been 100 years, and it's
[47:05] really hard to make a change after 100
[47:08] years.
[47:09] And when you watch it happen, it's
[47:10] actually really silly to a guy like me.
[47:13] You take all these stamp panels, you put
[47:15] them together, then you put them in a
[47:17] framing station, you build a body that
[47:19] looks something like a car, you put the
[47:22] doors on,
[47:23] and then you paint them.
[47:25] Once you get the color,
[47:27] you take the doors off,
[47:29] and then you start putting the interior
[47:31] inside the car.
[47:33] It comes in through the openings that
[47:34] already exist. I wish it went in like
[47:36] this big piece like this, but there's
[47:37] actually people coming in and out of the
[47:39] car. There's awkward, you know,
[47:41] movements. Then we lift the car up, we
[47:43] put stuff from underneath it, we put it
[47:45] down, then we put the seats in the car,
[47:47] and finally, we close it all up with
[47:49] glass, and we bring those doors that
[47:50] went away for a trip, and we put them
[47:52] back in the car.
[47:55] Most of the time, we're doing this with
[47:57] a big giant car and moving it and doing
[47:59] really nothing to it at all.
[48:01] What's funny though in this kind of
[48:03] whole process is that just recently
[48:04] Toyota just called this an engineering
[48:06] work of art.
[48:09] True.
[48:10] [applause]
[48:11] The Model Y.
[48:13] That that was humbling, but at Tesla,
[48:16] it's not good enough. If we're going to
[48:18] scale the way we want to do,
[48:20] we have to rethink manufacturing again.
[48:22] It's part of the master plan. We have to
[48:24] make another step change in cost.
[48:26] We started this on Model Y when we made
[48:29] these huge giga castings and we deleted
[48:31] hundreds of parts. We simplified
[48:33] assembly with the Model Y structural
[48:35] battery
[48:36] where we decided the floor should be a
[48:38] part of the car. The battery is the
[48:41] floor. We put the seats in the interior
[48:43] on the battery and we bring it up
[48:45] through a big open hole and we assemble
[48:47] it. And this allowed us to do things in
[48:49] parallel, fully rethinking the process
[48:51] and reducing the final assembly line by
[48:53] about 10%.
[48:55] And we thought, maybe we could do this
[48:56] other places.
[48:57] Yeah, I mean, in in a way
[49:00] the the the constraints become part of
[49:02] the solution rather than a problem.
[49:05] So, when you think about what I'm trying
[49:07] to say, I really want to hammer this
[49:08] home. When you have a car that's about 5
[49:10] m long and you have people working
[49:12] around it like we did in Model 3,
[49:14] and we change that to this process where
[49:16] we take different parts of the car and
[49:18] we can do more at the same time like we
[49:20] did with the Model Y structural battery
[49:22] pack.
[49:23] What you see here is us doing that on
[49:25] the front part of the vehicle or the
[49:26] rear part of the vehicle.
[49:28] That means we can get more people
[49:30] working on the car or robots working on
[49:33] it at the same time. That means we have
[49:35] better operator density, less time doing
[49:37] nothing. I call that space-time
[49:39] efficiency. It has nothing to do with
[49:41] quantum mechanics. We can have that
[49:42] conversation later.
[49:44] But, we get 44% more operator density,
[49:47] which means more work, less time walking
[49:49] back to the station, 30% improvement in
[49:52] space time efficiency. And because we're
[49:55] not building it in and out of the car
[49:57] with those slow movements of those
[49:58] robots I showed you from production
[49:59] hell, when we go to automate it, it's
[50:02] going to be a lot easier.
[50:05] In the end, that will probably look
[50:06] something like this where we balance
[50:08] parallel and series manufacturing in a
[50:10] way where we only do things that are
[50:12] necessary with a much shorter final line
[50:14] blocking a lot less of the entire rest
[50:16] of the factory so we can optimize
[50:19] material flow using the best practices.
[50:21] And what that means it's going to look
[50:22] something like this where we build all
[50:24] the sides of the cars independently, we
[50:27] only paint what we need to, and then we
[50:29] assemble the parts of the car once
[50:33] and only once. We put them where they
[50:34] need to go. The interior is attached
[50:37] from a bottom up or a top down strategy
[50:40] so there's more access for those robots
[50:42] and people. We aren't moving heavy
[50:44] objects around and doing nothing to it.
[50:46] And it means we're doing more work on
[50:47] the car more of the time.
[50:49] And then, when we take it all of these
[50:52] tested subassemblies and we put them
[50:54] together, we finally assemble the car
[50:56] only one time putting the sides on with
[50:58] all of their parts to a front and rear
[51:00] that was already assembled carrying the
[51:02] floor in with the seats and finally
[51:04] boxing it out with the doors one time
[51:07] just like Cybertruck.
[51:09] So in the end, you get the same car.
[51:13] But it's not going to be a Model Y.
[51:15] Yeah, this is not Yeah, this is a Model
[51:17] Y for illustration, not the next gen
[51:19] vehicle.
[51:21] In the end, what does that mean?
[51:23] To increase the scale and adoption of
[51:24] electric vehicles on the orders of
[51:26] magnitude that we just showed you,
[51:28] we have to make constraints part of the
[51:29] solution.
[51:31] It leads us to greater than 40%
[51:32] reduction in footprint, which means we
[51:34] can build factories faster with less
[51:37] CapEx and more output per per unit.
[51:41] Faster, less CapEx, more output per unit
[51:44] dollar.
[51:45] Zack's going to go into more details on
[51:47] this later, but it also means through
[51:49] this innovation and some of what my
[51:51] other engineering colleagues are going
[51:52] to talk to you about in the future,
[51:54] we'll reduce costs as much as 50%.
[51:57] This is the two-for-one concept you hear
[51:58] me and Elon talking about on earnings
[52:00] calls.
[52:03] Yeah, so I think our track record
[52:04] approves that we can deliver the best
[52:06] cars and we deliver the best cars in
[52:09] spite of because of these constraints.
[52:12] And I'd love
[52:14] to really show you what I mean and
[52:16] unveil the next-gen car.
[52:18] But you're going to have to trust me on
[52:19] that until a later date.
[52:21] Um just And I promise we'll always be
[52:24] delivering exciting, compelling, and
[52:25] desirable vehicles like as we always
[52:27] have.
[52:27] Have we ever not? We always do.
[52:29] But
[52:30] what else is Tesla known for?
[52:32] Speed.
[52:34] So with that, I want to bring on Colin
[52:36] who's in charge of making our cars go
[52:37] fast.
[52:40] [applause]
[52:43] Thank you, Lars.
[52:45] Thank you, Franz. My name is Colin
[52:47] Campbell and I have the real privilege
[52:48] of leading powertrain engineering here
[52:50] at Tesla.
[52:52] We make the fastest cars that you can
[52:54] buy for the money, whether they're
[52:55] electric or gas. And the Model S Plaid
[52:58] that we're looking at here, it has more
[53:00] than a thousand horsepower and pound per
[53:02] pound, the motors in that car are as
[53:04] powerful as jet engines. Our cars are
[53:06] super fun. People absolutely love
[53:08] driving in them.
[53:11] And the other thing that you probably
[53:12] all know about our powertrains is that
[53:13] they're efficient. Our cars go 25% to
[53:16] 30% further than other EVs in our class
[53:19] for the same amount of efficiency.
[53:22] But at Tesla, efficiency means more than
[53:24] just reducing how much energy the cars
[53:26] use. It's about how we develop, how we
[53:29] manufacture, how we refine, and how we
[53:31] scale the powertrain.
[53:35] Now the Model 3 and Y powertrain is a
[53:37] great example of this broader meaning of
[53:39] efficiency.
[53:40] So since we launched it back in 2017,
[53:43] we've continuously improved that
[53:44] powertrain and the factory that builds
[53:47] it. So, the drive unit, the engine of
[53:49] the car, is 20% lighter for the same
[53:51] power.
[53:53] We use 25% less heavy rare earths than
[53:56] when we started.
[53:58] And the powertrain the powertrain
[53:59] factory, which is behind me today, is
[54:02] 75% smaller and 65% cheaper than the one
[54:06] that we originally built.
[54:08] And what I really want to emphasize is
[54:09] that we did all of this without
[54:11] compromise. Our cars are just as
[54:13] powerful.
[54:14] They go just as far.
[54:16] They cost the same or less, and the
[54:19] factories have the same output.
[54:21] So,
[54:22] how did we do that?
[54:26] We did it by designing the entire
[54:27] vehicle and the entire factory together
[54:30] as one company, and this sets Tesla
[54:34] apart. We have small and highly capable
[54:37] teams, and to make a critical decision,
[54:39] we can have the battery cell chemists,
[54:41] the mechanical engineers, the
[54:42] manufacturing engineers, the supply
[54:44] chain team, the automation designers,
[54:46] the software programmers, all in one
[54:49] room,
[54:50] working together in real time.
[54:53] And that allows us to make decisions
[54:55] that are best for the whole car and to
[54:56] make them really fast. And that approach
[54:58] is unlike traditional automotive
[55:01] engineering, which is really fractured.
[55:04] And if you were to go buy like a premium
[55:05] German electric car, the engineers who
[55:08] designed the drive inverter in that car,
[55:09] they did not work for that car company.
[55:12] They worked for a contractor.
[55:13] At Tesla, we design the entire car and
[55:16] the factory that builds it.
[55:18] And I want to highlight a few examples
[55:20] of what we've been able to do in-house
[55:22] thanks to that unique approach.
[55:25] So,
[55:26] inside the charger of your Tesla are
[55:28] transistor packages, and that's at the
[55:30] top of your screen here. And every
[55:32] electron that moves you down the road
[55:34] flows through one of these packages.
[55:37] We designed our own custom package,
[55:39] which is what you're seeing here.
[55:41] And we can extract twice as much heat
[55:43] out of that package as what we could buy
[55:44] off the shelf. And so, what does that
[55:46] mean?
[55:47] It means that the silicard the silicon
[55:49] carbide wafer that's inside those
[55:51] packages can be much smaller. And
[55:54] silicon carbide is an amazing
[55:56] semiconductor,
[55:57] but it's also expensive, and it's really
[56:00] hard to scale.
[56:01] So, using less of it is a big win for
[56:03] us.
[56:05] And then on top of that, orchestrating
[56:07] all of these transistors and making them
[56:09] switch in the right ways is
[56:10] computationally extremely intensive.
[56:13] It used to require four microprocessors,
[56:16] which are shown here in the bottom left.
[56:19] We have developed our own custom
[56:21] microprocessor.
[56:22] It's purpose-built for high-power
[56:24] electronics. It's half the cost,
[56:27] and it does, in just one, the job of all
[56:29] of those four.
[56:31] And these are just two examples of many
[56:33] that I could use to showcase our
[56:35] expertise in high-power electronics. And
[56:37] that expertise has allowed us
[56:39] to take the cost of the chargers
[56:41] that were in our Model S when we
[56:43] launched it in 2012, both the cost and
[56:45] the mass, and cut both of those in half.
[56:50] And even more important, the power
[56:51] electronics they're central not just to
[56:53] our cars.
[56:55] They're also central to our
[56:56] superchargers, and to our energy storage
[56:58] products. And Rebecca and Mike will be
[57:00] talking more about that.
[57:02] So, in addition to the work that we've
[57:03] done in software and hardware,
[57:06] we've also done a lot of work in-house
[57:07] on software. So, this is the drive unit
[57:09] for Model 3 and Y. And if we take a
[57:10] cross-section, we see the stator and the
[57:12] rotor. And they're responsible for the
[57:15] core function of the drive unit, which
[57:17] is converting electricity into motion.
[57:19] And our custom software lets us simulate
[57:22] the rotating magnetic field that is
[57:24] responsible for that conversion. And
[57:26] getting that simulation exactly right,
[57:29] it's central to the cost, the weight,
[57:32] the size,
[57:34] and even the sound of the drive unit.
[57:36] And now, you can buy software that will
[57:39] do all of this,
[57:40] but our tools are faster,
[57:42] and they're more accurate, and that was
[57:44] not easy to do.
[57:46] And that allows us to quickly iterate
[57:48] through millions of possible drive unit
[57:49] designs to find the best one.
[57:53] Want to highlight one more area
[57:56] where Tesla really excels because we
[57:57] integrate work that is often farmed out.
[58:00] So, when you are making a new product,
[58:03] it's not enough to think about the
[58:04] product itself. You have to think about
[58:06] how you're going to make it at scale.
[58:09] So, at Tesla, our powertrain
[58:11] and our powertrain manufacturing
[58:12] equipment is both designed under one
[58:14] roof.
[58:15] The engineers who are designing the
[58:16] motor, they are in the same room as the
[58:18] engineers who are designing the machine
[58:20] that's going to put that motor together.
[58:22] And that collaboration pushes us from
[58:24] day one
[58:25] to design products that are not only
[58:27] high performance, but are really easy to
[58:29] assemble.
[58:31] So,
[58:34] all of this expertise that we have in
[58:35] the powertrain team, in hardware, in
[58:37] software, in manufacturing,
[58:39] it's going to have a major impact on our
[58:40] next platform.
[58:44] In our next powertrain,
[58:46] the silicon carbide transistors that I
[58:47] mentioned that are
[58:49] key component, but expensive, we figured
[58:51] out a way to use 75% less
[58:54] without compromising the performance or
[58:56] the efficiency of the car.
[58:58] And of course, we know the battery cell
[59:00] supply is one of the constraints on the
[59:02] scalability of EVs right now.
[59:05] Our new powertrain is compatible with
[59:07] any battery chemistry.
[59:09] That will give us great flexibility in
[59:11] battery sourcing.
[59:14] If we want to make EVs more accessible
[59:15] to people, they have to be cheaper.
[59:18] We've reduced the drive unit cost to
[59:20] about a thousand dollars, and we don't
[59:22] think any other automaker is even close
[59:24] to that number.
[59:27] Finally, the bigger a factory is, the
[59:30] longer it takes to build.
[59:31] If we can build the same number of cars
[59:33] from a smaller factory, we are going to
[59:35] be able to scale EV production faster.
[59:37] Our next powertrain factory is 50%
[59:40] smaller than the one that's behind me
[59:42] today, even though it has the same
[59:43] capacity.
[59:47] All of these improvements, I think, are
[59:49] going to be transformative for the
[59:50] adoption of EVs and our ability to scale
[59:52] them. There's one more thing that I want
[59:53] to highlight.
[59:56] So, I talked about how we had reduced
[59:57] the amount of rare earth in our
[59:59] powertrains. And as the world
[60:01] transitions to clean energy, demand for
[60:03] rare earth is really increasing
[60:05] dramatically.
[60:07] And not only is it going to be a little
[60:08] hard to meet that demand,
[60:10] but mining that rare earth, it has
[60:12] environmental and health risks.
[60:14] So, we want to do even better than this.
[60:18] We have designed our next drive unit,
[60:20] which uses a permanent magnet motor, to
[60:22] not use any rare earth materials at all.
[60:28] So, how does all this fit to the master
[60:29] plan?
[60:31] We can make lower cost products that are
[60:32] still efficient and compelling, and we
[60:34] can make them at scale.
[60:35] We're going to use less of constrained
[60:37] commodities, silicon carbide, rare
[60:39] earths.
[60:40] We're going to build them all in compact
[60:41] and high output factories that are
[60:43] easier easy for us to build quickly.
[60:47] We're going to make that easy to scale
[60:48] powertrains all the way up to the levels
[60:50] that Drew and Elon mentioned at the
[60:51] beginning.
[60:53] And this achievement, like all of the
[60:54] achievements that I mentioned today, is
[60:56] only possible because of the incredible
[60:58] people on our powertrain teams. They are
[61:00] absolutely committed to the cause of
[61:02] sustainable energy, and that is why we
[61:04] can do what no other company can do.
[61:08] Thank you.
[61:10] [applause]
[61:16] So, next I'd like to welcome my
[61:17] colleagues Pete Bannon and David Lau.
[61:20] Thanks, Colin.
[61:23] [applause]
[61:25] Good afternoon everybody and welcome to
[61:27] uh Texas.
[61:29] Pictured here is the low voltage system
[61:31] of a Model S from 2012. This car was
[61:34] primarily designed with vendor source
[61:36] controllers integrated by Tesla. There
[61:38] are over 300 volt low voltage devices
[61:42] that encompass everything from the
[61:43] mundane like the light in the glove box
[61:46] to the complicated like the infotainment
[61:48] computer to the safety critical like the
[61:50] airbags, steer by wire, and brake by
[61:53] wire.
[61:56] The low voltage harness is built from
[61:57] individual wires cut to length, crimped,
[62:00] and inserted into connectors. A manual
[62:03] process that is
[62:05] uh tedious, error-prone, and doesn't
[62:08] scale well.
[62:10] Going forward we want to reduce the size
[62:11] and complexity of the harness and enable
[62:14] automated manufacturing.
[62:16] These wire harnesses introduce
[62:18] extraordinary complications, especially
[62:20] in the early stages of development and
[62:21] bring up of a vehicle because when we're
[62:23] trying to bring up this entire system
[62:25] and we see that it's not working
[62:26] properly, we don't know whether it's a
[62:27] problem with the software, with the
[62:29] controller, the processor, any of the
[62:31] endpoints in this entanglement of wires.
[62:34] And so, we have to go and debug
[62:35] everything all at once. Um and Pete's
[62:37] going to tell us about how we're going
[62:38] to make it better.
[62:39] Mhm. So, I'm going to talk briefly about
[62:42] a few things that we've changed over the
[62:43] years since the Model S. With the Model
[62:45] 3 Tegra Tesla started designing an
[62:48] increasing share of our controllers by
[62:50] merging controllers together
[62:53] and we were able to simplify the design
[62:55] with a significant reduction in wire
[62:57] count and weight. Um going from S to 3
[63:00] we reduced the wire harness uh by 17 kg
[63:04] uh compared to Model S. And this is a
[63:06] pretty big deal and just to put that
[63:07] into context, our VP of engineering Lars
[63:10] will deliver a a bottle of your favorite
[63:12] spirit to your desk if you manage to
[63:14] save a kilogram of weight from the car.
[63:16] So, this improvement cost him quite a
[63:18] lot.
[63:21] Did we Did we skip one? Yeah. Um the
[63:24] original Model 3 controllers were
[63:25] enhanced and then and built into Model
[63:28] Y. Those improved controllers were then
[63:30] moved back into Model 3 when we
[63:32] introduced the heat pump. At that point,
[63:34] we had uh shared controllers across both
[63:36] cars, which helped simplify our supply
[63:38] chain.
[63:39] Those Model 3 and Y controller designs
[63:41] were then updated and enhanced for the
[63:42] new versions of Model S and X that we
[63:44] introduced in 2021.
[63:47] For Cybertruck, we are now designing 85%
[63:50] of the controllers in the car.
[63:52] For the next-gen platform, we're going
[63:53] to finish the job and be designing 100%
[63:55] of the controllers to give us full
[63:56] control of the design and the supply
[63:59] chain at the component level.
[64:01] Having full control of the supply chain
[64:02] at the component level has been critical
[64:04] to Tesla over the last few years as the
[64:07] supply constraints have hampered our
[64:09] ability to build cars.
[64:11] And with that control comes control over
[64:13] all the software, which enables us to
[64:16] develop features and functionality that
[64:18] we never even dreamed of at the time we
[64:20] designed the hardware. So, that's why
[64:22] you see software in your car is getting
[64:23] better and better over time in ways that
[64:25] we didn't even think of when we designed
[64:27] the hardware in the first place.
[64:28] Mhm. Yeah, Sentry Mode is one of my
[64:30] favorite last-minute changes.
[64:33] Uh with the introduction introduction of
[64:35] Model 3 in 2017,
[64:38] we deleted relays and fuses from the car
[64:40] in favor of eFuses. eFuses replace
[64:43] moving parts with solid-state
[64:44] transistors that provide fine-grain
[64:46] control of the power system uh to
[64:48] software and allow software to do
[64:50] advanced things like load shedding in
[64:52] the car under adverse conditions. eFuses
[64:55] also allow for software-controlled
[64:57] retries for transient faults, and they
[64:59] allow for detailed monitoring of the
[65:01] power system over time.
[65:03] And what The theme you hear a lot about
[65:05] today is software-controlled hardware.
[65:07] And when I think about that in my team,
[65:09] software-controlled hardware is
[65:11] fundamentally about being able to
[65:12] sidestep what otherwise would have been
[65:14] static trade-offs between one attribute
[65:16] or another in a piece of hardware. We
[65:19] are able with software to in instill
[65:22] intelligence, context awareness, and
[65:25] context specific behavior into what
[65:27] otherwise would have been a piece of
[65:28] hardware that had to get optimized for
[65:30] one type of scenario.
[65:32] We get more of everything.
[65:34] In 2022, we completed the transition
[65:36] from lead acid batteries to lithium-ion
[65:37] batteries and
[65:39] and which uses a new toolless connector.
[65:42] We expect lithium-ion batteries to last
[65:44] for the life of the car, so we don't
[65:45] expect anyone to wake up to a dead
[65:47] battery ever again.
[65:50] This eliminates a major source of
[65:52] failures for our car and the toolless
[65:54] connector
[65:55] makes it easier to service the car and
[65:57] it also includes a software feature to
[66:00] allow
[66:01] validation that the connector has been
[66:03] properly installed at the end of a
[66:05] service event, which removes another
[66:07] source of failures in our cars.
[66:09] Obviously, the mass and volume savings
[66:11] are also significant, which is super
[66:14] helpful.
[66:17] At Tesla, we're always trying to improve
[66:19] every single component in the car and a
[66:21] nice example of that is the 15-in
[66:23] display that was originally shipped in
[66:25] the 2017 Model 3.
[66:27] Over time, the cost of the display has
[66:30] gone down 24%. We've been able to reduce
[66:33] the weight by 12% and we've reduced the
[66:36] power by 33%.
[66:37] At the same time, we've increased the
[66:39] brightness of the display by 50% and
[66:41] improved the color accuracy. So, this is
[66:43] one of our favorite things at Tesla to
[66:44] just make a component cheaper and
[66:46] better.
[66:49] At Tesla, the drive to improve
[66:50] efficiency in the car is never ending. I
[66:52] think of it is like peeling a carrot.
[66:53] You just take a swipe at it, a little
[66:55] bit falls off, doesn't really make much
[66:57] of a difference, but if you just keep at
[66:59] it, over time it accumulates into a
[67:01] pretty significant improvement.
[67:03] One of the changes that we're looking to
[67:04] make right now is to change something
[67:07] that's been steady for the last 60
[67:08] years, 12 volts.
[67:10] Um 12 volts
[67:12] has been uh as I said for 60 years. The
[67:15] demand for power in the car has been
[67:16] steadily increasing till the point where
[67:18] we now have to have pretty large wires
[67:20] to drive over 200 amps of current around
[67:22] the car, which increases the mass and
[67:24] cost. With Cybertruck and all future
[67:26] Tesla platforms, we will be moving to 48
[67:28] volts. This reduces the current needed
[67:30] by a factor of four. And since power
[67:33] loss in the harness is resistance times
[67:35] the square of the current, a 4x
[67:38] reduction in current leads to a 16x
[67:40] reduction in lost power while
[67:42] distributing energy in the car.
[67:44] Um that allows for smaller wires,
[67:46] smaller e-fuses, and smaller
[67:48] controllers. It also allows us to make
[67:50] those heat sinks smaller or in many
[67:52] cases remove it completely, all
[67:54] benefiting the car in terms of mass and
[67:57] weight. 48 volts is the future for low
[68:00] voltage design at Tesla and likely the
[68:02] rest of the industry in due course.
[68:04] We welcome and encourage other OEMs and
[68:07] the entire supplier network to join us
[68:09] on this evolution.
[68:10] Mhm.
[68:11] You bet.
[68:13] Uh the number of wires in the car is
[68:15] driven by the number of endpoints that
[68:17] need to be powered and controlled. In
[68:19] the past, centralized control has led
[68:21] for wires spanning the entire car.
[68:24] For Cybertruck design, we have moved to
[68:26] a local controller where the wire is
[68:29] connected to the nearest controller and
[68:32] those controllers are connected over
[68:33] Ethernet.
[68:35] Wires are routed to the nearest
[68:36] controller where the data is converted
[68:37] into a network packet for transi-
[68:39] transmission to the correct location in
[68:42] the car. To be effective, the network
[68:43] must be reliable, have low latency, and
[68:45] low jitter, and these are all attributes
[68:47] that we've been able to achieve with the
[68:49] current design.
[68:50] The design has eliminated most of the
[68:52] cross-car wires in Cybertruck, and with
[68:55] the next-gen platform, we're going to
[68:56] finish the job and eliminate all of
[68:58] them.
[68:59] This um consolidated vehicle network
[69:01] allows us also to make a whole lot more
[69:02] dynamic changes on the fly Uh to how
[69:05] components in the vehicle talk to each
[69:07] other rather than the traditional
[69:08] approach of separate CAN buses that are
[69:10] spread throughout the vehicle.
[69:11] Yeah, and it's fixed in hardware.
[69:13] And one of the nice things is that you
[69:14] can see the entire vehicle through a
[69:16] single connection which wasn't possible
[69:18] in the past for debug.
[69:21] For the next gen platform, we're looking
[69:23] to optimize the controller design across
[69:24] the entire car and across all of the
[69:27] organizations, not just for a given
[69:28] subsystem. Simplifying the wire harness
[69:31] will enable automation. 48 volts will
[69:34] allow us to reduce the size,
[69:36] mass, and cost of the low voltage
[69:38] system. And it's one of the key things
[69:40] to enabling us to scale
[69:43] production of the low voltage systems to
[69:45] a further master plan three.
[69:48] David?
[69:50] Cool.
[69:52] So, I joined Tesla in 2012. And even in
[69:54] those very early days of the Model S, we
[69:57] already had the ability to update
[69:59] software over the air in all of the
[70:01] controllers throughout the entire
[70:03] vehicle. At a time when other cars were
[70:05] just starting to update software in only
[70:08] their infotainment units.
[70:10] And furthermore, we've had
[70:11] sophisticated, anonymized data logging
[70:13] and telemetry capabilities that allow us
[70:16] to understand how the fleet is doing,
[70:18] what customers are doing with it, and
[70:21] how it's responding in the field to that
[70:22] usage.
[70:24] These two capabilities combined, OTA
[70:27] updates and data insights, give us the
[70:30] ability to iterate quickly on our
[70:31] software and to maximize the amount that
[70:34] we learn and and proceed and and and and
[70:37] the progress that we make in every one
[70:38] of those iterations.
[70:40] We've used these capabilities to inform
[70:42] countless design decisions in both
[70:44] software and hardware.
[70:47] Yeah, one example is that we were able
[70:48] to
[70:49] monitor and track the use of the sunroof
[70:51] in our cars and found that our customers
[70:53] never use their sunroof. So, we we made
[70:55] this easy decision to remove it.
[70:57] Very rarely, yeah. We haven't had some
[70:59] rust for a while.
[71:01] [clears throat]
[71:02] Um
[71:03] as another example,
[71:04] we are able to use our collision data to
[71:07] design our crash safety systems for what
[71:09] happens in the real world, far beyond
[71:12] what regulatory and consumer ratings
[71:14] tests prescribe.
[71:16] And then furthermore, we're able to use
[71:18] that data to recreate in simulation all
[71:21] of those crashes every single time we
[71:24] make a change to the vehicle's design or
[71:27] to the software that controls the
[71:28] airbags and other restraint systems.
[71:31] Here on this graph, the gray dots that
[71:33] show up first are specific crash tests
[71:36] that are prescribed by regulatory and
[71:38] consumer ratings agencies throughout the
[71:40] whole world.
[71:41] And the rest of the color that fills in
[71:43] around and on top of it, those represent
[71:46] the crashes that we've seen happen in
[71:47] the real world.
[71:49] Those are what we design for and what we
[71:52] test to.
[71:53] Right. And last year with this data was
[71:55] used to change the algorithm for seat
[71:57] belt tensioning to reduce injuries in
[71:59] the field, which was OTA to all of our
[72:01] cars.
[72:04] Here are a few statistics that
[72:06] demonstrate how staggeringly quickly we
[72:08] gain new insights from the anonymized
[72:10] data that we've received from the fleet
[72:12] every day. Every day, 123 million miles
[72:17] driven, 1.9 million charge sessions, and
[72:20] those rates are only increasing as we
[72:22] deliver more and more cars into the
[72:24] world.
[72:27] As another example, we've analyzed
[72:29] patterns in the way the fleet drives and
[72:31] charges to optimally size battery packs
[72:34] for our next generations of vehicles.
[72:36] And you'll hear lots of other examples
[72:38] throughout the rest of the day about how
[72:39] we're using data to inform decisions in
[72:41] the product and in manufacturing.
[72:45] So, you've heard a lot from Pete, Colin,
[72:46] and others about how important vertical
[72:48] integration is for us in hardware, but
[72:51] it's especially important for us in
[72:53] software. And this is extremely
[72:55] difficult from the way the rest of the
[72:57] traditional automotive supply chain is
[72:59] set up.
[73:00] In most situations, all the controllers
[73:02] in the car are delivered by different
[73:04] tier one suppliers whose software is
[73:07] written by tier two suppliers who farm
[73:09] some of that software out to different
[73:11] tier three suppliers and so on and so
[73:13] on.
[73:15] Making a change that spans multiple
[73:17] components takes months of coordination
[73:19] before any work can even start.
[73:22] And furthermore, integration of software
[73:25] in the server-side with what's happening
[73:27] in our vehicles has always been a core
[73:29] part of our DNA and enables us to do
[73:31] things that nobody else can.
[73:33] Most engineers think about the vehicle
[73:36] as a complete consolidated fully
[73:38] contained system.
[73:39] But in the software team, we think about
[73:41] the system as including the vehicle as
[73:44] well as all of our back-end server-side
[73:46] applications and infrastructure and the
[73:48] resultant feedback loop from the entire
[73:50] fleet that informs all the decisions we
[73:52] make as an engineering team.
[73:56] For example, we recently released a
[73:58] feature in Model S and X that
[74:00] automatically, predictively, adjusts and
[74:03] raises the suspension for ride comfort
[74:05] before the car hits a section of rough
[74:07] road.
[74:08] We do that by leveraging the fleet to
[74:10] generate a map of road roughness
[74:12] everywhere our cars drive.
[74:15] Sounds pretty simple, but requires
[74:17] coordination of software across a number
[74:18] of different components inside and
[74:20] outside of the car.
[74:22] The restraint control module, whose
[74:24] inertial measurement unit senses the
[74:25] road roughness,
[74:27] the autopilot computer, whose GPS module
[74:29] and localizer figure out where the car
[74:32] is in the world,
[74:33] our navigation server, which aggregates
[74:36] the anonymized telemetry from the fleet
[74:38] and annotates our map with things like
[74:40] updated speed limits, lane topology, and
[74:42] now road roughness.
[74:45] The onboard navigation engine, which
[74:48] looks ahead of the route that the car is
[74:49] traveling on and determines whether
[74:51] things are about to get rough.
[74:53] And finally, the air suspension
[74:55] controller, which takes all of those
[74:57] factors into account and decides con-
[74:59] continuously whether it is appropriate
[75:02] to adjust the sus- suspension for ride
[75:04] comfort.
[75:05] And before we release this feature to
[75:07] the world, we sent prototype versions of
[75:10] it to the entire fleet that ran
[75:12] passively in the background sending us
[75:15] anonymous data about every time it would
[75:17] have engaged, which taught us exactly
[75:20] how it was going to behave in the wild
[75:22] in every environment and every
[75:24] circumstance.
[75:25] Yeah, being able to test new algorithms
[75:27] in the background without an impact in
[75:30] the car is is really critical ability
[75:32] for us, especially for safety critical
[75:33] things like automatic emergency braking.
[75:36] And we've used it uh um iterate on early
[75:39] versions of our stability control
[75:41] algorithms, which were introduced to the
[75:43] entire Model 3 fleet in 2017 with
[75:46] extreme extremely high confidence, even
[75:48] on a very foundational and safety
[75:50] critical function.
[75:53] Okay.
[75:55] So far you've heard me talk a lot about
[75:57] how we leverage our software systems to
[75:59] iterate quickly on the customer-facing
[76:00] product. But what I haven't talked about
[76:03] and what we haven't said much about
[76:04] publicly in general is how we leverage
[76:06] these capabilities in our own internal
[76:08] operations.
[76:09] So for example,
[76:11] in the early days of manufacturing Model
[76:13] S, we very quickly realized that there
[76:15] are a lot of different ways you could
[76:17] misassemble a car.
[76:19] You could forget to plug in a wire, you
[76:21] could not fully set a connector, you
[76:23] could pull a part from the wrong bin.
[76:26] So, when we designed our manufacturing
[76:28] processes for Model 3 and all of our
[76:30] other vehicles going forward, we took a
[76:32] page from the software playbook, which
[76:34] says
[76:35] test early and test often.
[76:39] And we applied that mantra to every
[76:41] single car that we build at every step
[76:44] of the assembly line.
[76:46] So now, when a production associate
[76:48] plugs something into the car, the
[76:50] central car computer sees that
[76:51] connection,
[76:52] confirms that it's the correct part for
[76:54] the type of car that's supposed to be
[76:55] built,
[76:56] installs a software update if necessary,
[76:58] checks configurations and calibrations,
[77:00] and runs a barrage of tests on that
[77:02] thing and all the other things that are
[77:03] connected to it. And if it finds an
[77:05] anomaly that can't be fixed purely with
[77:07] software,
[77:08] it throws an alert immediately that is
[77:10] displayed prominently on the vehicle's
[77:12] display as well as sent to back-end
[77:14] command and control systems, so that a
[77:15] human can come over and fix the problem
[77:18] before the rest of the car gets built on
[77:20] top of it.
[77:22] And we're taking a similar approach in
[77:24] service. Through a combination of
[77:26] onboard vehicle diagnostics, natural
[77:28] language processing applied to customer
[77:30] narratives when a customer schedules a
[77:32] service appointment, and a suite of
[77:34] internally developed tools, we are
[77:36] successfully diagnosing, scheduling, and
[77:39] ordering parts for over 33% of customer
[77:42] concerns in service.
[77:44] Yep, and we also use uh testers built
[77:47] around our own controllers to test
[77:48] subassemblies before they get to the
[77:50] car, which will be even more important
[77:52] as we go to Lars's unboxed assembly
[77:54] strategy.
[77:55] 100%. We'll know with extremely high
[77:57] confidence that before those boxes get
[77:59] put together,
[78:00] that they are that they are assembled
[78:01] they themselves are assembled correctly.
[78:06] Okay, you've heard at lots of other
[78:08] events about all the awesome things that
[78:10] our autopilot and AI teams are doing to
[78:12] make the car drive itself, bringing us
[78:14] into the future of autonomy.
[78:17] But we've also been thinking for years
[78:19] about all of the other pieces that we
[78:20] are going to need to manage a network of
[78:22] autonomous vehicles.
[78:25] A lot of this has been happening behind
[78:26] the scenes in the form of platform level
[78:28] functionality that we'll leverage later.
[78:31] But you've seen some of it surface
[78:32] already in terms of features that our
[78:34] customers and our internal operations
[78:36] teams can benefit from.
[78:38] Like in 2021, we built on top of our
[78:40] mobile app phone key and gave our
[78:42] customers the ability to share their car
[78:44] with anyone by sending them an
[78:45] invitation in a text message or an
[78:47] email.
[78:48] Last year, we introduced profile
[78:50] synchronization, which synchronizes your
[78:52] seat, steering, and mirror positions, as
[78:54] well as your settings, media favorites,
[78:56] and stored navigation locations across
[78:58] all vehicles in your account.
[79:01] Internally, we have an app that allows
[79:03] our personnel in engineering,
[79:05] manufacturing, delivery, and logistics
[79:08] to view, locate, and drive all
[79:11] Tesla-owned vehicles at their site.
[79:15] And when customers bring their car into
[79:16] service, when they receive a Tesla-owned
[79:19] service loaner vehicle, we are starting
[79:21] at some locations to automatically add
[79:23] that vehicle to the customer's mobile
[79:25] app account, so that combined with
[79:26] cloud-synchronized profiles and phone
[79:28] key, it's a completely seamless
[79:30] experience. As soon as the customer
[79:32] walks up to that loaner, it behaves and
[79:35] feels exactly like their own car.
[79:38] And all this is built on top of
[79:40] end-to-end encryption
[79:42] and cryptographically signed commands,
[79:44] so that customer data remains private
[79:46] and obscure,
[79:48] and the fleet only trust commands from
[79:50] authorized parties.
[79:52] So, that's a bit of a preview of all the
[79:54] pieces that we're building looking ahead
[79:56] to our future of autonomous fleets.
[79:59] Synchronization, permissions management,
[80:01] security, and privacy.
[80:05] And that has been a whirlwind tour
[80:07] through some of the things that we're
[80:08] doing behind the scenes to enable the
[80:11] efficiency, cost reduction, and speed
[80:13] that we are going to need for their next
[80:15] phase of growth and for the next phase
[80:17] of the master plan.
[80:18] And so now, to tell us about the latest
[80:21] in full self-driving,
[80:23] I'd like to introduce Ashok.
[80:27] [applause]
[80:33] Thanks, David and Pete. Hey everyone, my
[80:35] name is Ashok Elluswamy. I work on
[80:37] autopilot and self-driving at Tesla.
[80:39] I joined Tesla back in 2014, so I've
[80:41] been working on this problem for almost
[80:43] 9 years now.
[80:46] Yeah, some of you might be wondering,
[80:47] "Hey, what's this self-driving got to do
[80:49] with the plan to a sustainable future?"
[80:52] But it's actually a critical part of
[80:53] this plan, and here's why.
[80:57] Currently, when the car is not being
[80:59] used, it is sitting idly in parking lots
[81:02] not doing anything.
[81:03] But when autonomy is truly unlocked,
[81:06] this car, instead of being idle, can go
[81:08] serve other customers.
[81:12] This fundamentally uh reduces the
[81:16] need to scale manufacturing to extreme
[81:18] levels because each car is being used
[81:20] way more.
[81:25] But this is no trivial problem. Building
[81:27] a
[81:27] scalable self-driving system is, I
[81:29] think, one of the hardest real-world AI
[81:31] problems out there right now.
[81:33] Nonetheless,
[81:35] we at Tesla have made significant
[81:37] strides in making
[81:39] one of the most general systems uh at
[81:41] sol- solving this problem.
[81:47] There are three main parts to get right
[81:49] to build a scalable problem.
[81:51] First is the architecture,
[81:54] the architecture of the AI system.
[81:55] Second is the data, and third
[81:58] is the compute. We'll start with the
[82:00] architecture.
[82:04] At Tesla, we are betting on AI, machine
[82:07] learning neural networks to help us
[82:09] build a general vision and planning
[82:10] system.
[82:12] In the early days, we used to have
[82:14] single camera, single frame neural
[82:16] networks that produce some outputs.
[82:19] These were stitched together uh in some
[82:21] post-processing steps for the planner.
[82:23] But this was very brittle and was not
[82:25] leading to great success. So, what we
[82:27] did in the last few years is transition
[82:29] most of our stack into this multi-camera
[82:33] video neural networks.
[82:35] These neural networks take in the live
[82:38] feed from the car
[82:40] in real time of the eight cameras in the
[82:41] car
[82:43] and produce a single unified 3D output
[82:46] space.
[82:49] There are many tasks that we produce
[82:51] such as the presence of obstacles, their
[82:52] motion, lanes, roads, uh traffic lights,
[82:56] what have you.
[82:58] This is one example output that you're
[83:00] seeing here. Uh this is from our
[83:01] occupancy network that predicts the
[83:03] positions of obstacles and their motion.
[83:06] You can see that it precisely captures
[83:08] the swirly violent motion of this truck
[83:10] next to us and this helps the planning
[83:12] system to avoid a collision with this
[83:14] object.
[83:19] Some of the tasks such as lane
[83:20] connectivity are more complicated to
[83:22] model using naive methods in computer
[83:24] vision.
[83:26] So, we don't stop at the computer vision
[83:27] techniques, but we reach out to
[83:29] techniques in other areas such as
[83:30] language modeling, reinforcement
[83:32] learning, uh to model this task.
[83:35] This is similar uh these these networks
[83:37] use similar techniques such as
[83:38] transformers, attention modules,
[83:40] auto-regressive modeling of uh
[83:43] tokens similar to what large language
[83:45] models like ChatGPT do out there.
[83:49] With such an end-to-end system of
[83:51] solving perception, we have really
[83:54] removed the brittle post-processing
[83:55] steps and produce high-quality outputs
[83:57] for the planning system.
[84:01] Even the planning system is not stuck in
[84:03] the old ways. Uh it is now starting to
[84:05] use more and more AI systems to solve
[84:07] this problem.
[84:09] The neural neural network-based planners
[84:11] are needed especially in complicated
[84:13] urban planning when there's a lot of
[84:15] other objects interacting with us.
[84:17] This is, for example, an intersection
[84:19] where we have to turn left while
[84:20] yielding to crossing objects and to the
[84:22] pedestrians crossing the road. We have
[84:23] to do this both
[84:25] safely and smoothly while respecting
[84:28] everyone's right of way
[84:29] and preferences.
[84:32] If this was done naively, each
[84:34] configuration would take 10 milliseconds
[84:36] of compute and there are easily easily
[84:38] thousands of configurations to reason
[84:40] about.
[84:41] This would not be feasible using
[84:43] traditional compute, but using AI we
[84:45] have packaged all of this into a 50
[84:47] millisecond compute budget so it can run
[84:49] real time.
[84:54] The second big piece of this puzzle is
[84:55] the data.
[84:57] This is where Tesla has a unique
[84:58] advantage because it can tap into the
[85:00] fleet
[85:01] tap into the fleet to access the exact
[85:04] data that can fix the problems.
[85:09] But raw data is not sufficient.
[85:12] We In order to train these networks,
[85:14] you need label data. You need labels to
[85:17] supervise the networks.
[85:19] And if you only depended on the human
[85:22] labelers,
[85:23] this data would be too tiny to train
[85:25] these large
[85:27] multi-camera video modules. We need lot
[85:29] lots and lots of data to train these
[85:31] networks.
[85:32] Hence,
[85:33] we have built a sophisticated auto
[85:35] labeling pipeline
[85:37] that collects data from the fleet,
[85:40] runs computational algorithms in our
[85:42] data center, and then produces the
[85:44] labels to train these networks.
[85:47] Here, you are seeing a 3D reconstruction
[85:50] that is happening by collecting various
[85:51] clips from different cars in the fleet
[85:54] and assembling them all of them together
[85:55] into a single unified representation of
[85:58] the world around the car. You can see
[85:59] all the lanes, the road boundaries,
[86:01] curb, crosswalks, even the text on the
[86:03] road being accurately reconstructed by
[86:06] these algorithms.
[86:11] We don't have to stop at this kind of
[86:12] reconstruction.
[86:14] Once we have this base reconstruction,
[86:16] we can build various simulations on top
[86:18] of it to produce an infinite variety of
[86:21] data to train for all the corner cases.
[86:24] We have a capable simulator that can
[86:27] synthesize adversarial weather, lighting
[86:30] conditions, and even the motion of other
[86:32] objects to test all the corner cases
[86:34] that might even be rare in the real
[86:36] world.
[86:40] Here's an example of why this data is
[86:41] critical and how we can solve problems
[86:44] using data.
[86:45] So, back in the days,
[86:47] um
[86:48] for example, in this case, we had some
[86:49] false braking where we thought this car
[86:52] that's actually parked there was going
[86:54] to move into our path and hence we were
[86:56] precautiously braking.
[86:59] But, it is unnecessary because the car
[87:00] is truly parked.
[87:02] So, how can we solve this? What we did
[87:04] was we mined the fleet
[87:06] for similar cases where the car had
[87:08] false braking uh due to some parked car.
[87:11] We added 14,000 videos to our training
[87:13] set. And then, once we uh trained the
[87:16] networks again with this new data, it
[87:18] now correctly understands that, okay,
[87:19] there is no driver in this car, hence it
[87:21] must be parked, so then there is no need
[87:23] to brake.
[87:24] And this is a systematic way to solve
[87:26] problems. On the chart on the right
[87:28] side, you can see that every time we add
[87:30] data, the performance improves. And
[87:33] then, we can do this uh for every kind
[87:35] of task that we have in our system.
[87:38] And this is what we call as data engine.
[87:41] We identify challenging cases such as
[87:43] the one you saw earlier. And there could
[87:45] be other different types of challenging
[87:46] cases, too.
[87:47] We mined the fleet for such data,
[87:51] put it through our auto labeling system,
[87:52] and produce the labels,
[87:54] add it to the training set.
[87:56] And once you have the newly trained
[87:57] models, we deploy them to the fleet
[87:59] fleet.
[88:02] If you rinse and repeat this process,
[88:04] everything gets better and better.
[88:10] The final critical piece is the compute.
[88:13] In order to train these large models in
[88:15] a reasonable amount of time, you need
[88:16] lots of compute. In addition, compute is
[88:19] also required to produce the labels
[88:21] automatically. This is just compute in
[88:23] our data centers. In addition, we also
[88:25] have high
[88:28] compute
[88:30] computers in the car which can run up to
[88:32] like 150 terops of compute.
[88:35] On the back end, we have a 14,000 GPU
[88:38] cluster and roughly 30% is used for auto
[88:41] labeling and the remaining 70% is used
[88:42] for training.
[88:45] We also have a 30 petabyte of video
[88:48] cache and this is growing to 200
[88:50] petabytes.
[88:51] All of this is going to significantly
[88:53] increase one we once we bring Dojo,
[88:56] which is our training computer, on board
[88:57] into this.
[88:59] Just to give a reference, just the
[89:01] occupants networks that you saw earlier
[89:03] use 1.4 billion frames to train these
[89:05] networks.
[89:10] We have already shipped our FSD beta
[89:12] software to pretty much everyone that
[89:14] has bought it in United States and
[89:16] Canada. This amounts to roughly 400,000
[89:19] customers who can turn it on anywhere
[89:21] and then the car would attempt to drive
[89:23] to the destination. It's still
[89:24] supervised,
[89:25] uh but it it can already handle turns,
[89:28] stop at traffic lights, yield with other
[89:30] objects, and generally get to the
[89:32] destination.
[89:34] We have observed that people who use FSD
[89:36] are already five to six x safer than the
[89:39] user US national average.
[89:42] Like I mentioned earlier, the solution
[89:44] to scalable FSD is getting the AI
[89:47] architecture, the data, and the compute
[89:50] just right. And we have assembled a
[89:52] world-class team to execute on this.
[89:54] We're pushing the frontier on these on
[89:56] these three items and as we improve the
[89:59] safety, the reliability, and the comfort
[90:01] of our system, we can then unlock
[90:03] driverless operations, which then makes
[90:06] the car be used way more than what it's
[90:08] used right now.
[90:11] With that, I'd like to introduce uh Elon
[90:13] back on stage uh
[90:14] to give some more updates.
[90:26] Yeah, this is next slide is video.
[90:35] It's kind of weird seeing the arms and
[90:37] legs just separate.
[90:45] We have a whole lab full of arms and
[90:46] legs.
[90:56] [clears throat]
[90:58] With bearing in mind that uh when we did
[91:00] AI Day,
[91:02] uh this version of Optimus didn't walk
[91:05] walk at all.
[91:07] So, the rate of improvement here I think
[91:08] is is quite uh significant.
[91:11] Um it's obviously not doing parkour,
[91:13] uh but I
[91:16] it is walking around, and we have
[91:18] multiple multiple uh
[91:20] copies I suppose of Optimus.
[91:23] Um
[91:27] The thing that's
[91:29] I think Tesla brings the table
[91:32] that others don't have is that
[91:35] we have um
[91:41] we have the uh
[91:42] real-world AI. I We're We're the most
[91:44] advanced in real-world AI. So, the same
[91:48] AI that
[91:49] drives the car
[91:52] uh it
[91:53] which you can think of the car really is
[91:55] a robot on wheels, and this is a robot
[91:57] on legs.
[91:58] Um so,
[92:00] as we solve real-world AI, and think
[92:02] there's any I don't think there's anyone
[92:04] even close to Tesla on solving
[92:06] real-world AI.
[92:08] That same computer and software
[92:11] goes into Optimus.
[92:14] So, it's it's not that helpful to have a
[92:16] humanoid robot if you have to program
[92:18] every individual action. It needs to be
[92:21] able to
[92:22] walk around autonomously and solve
[92:24] tasks. You should be able to instruct it
[92:27] in simple things by showing
[92:30] visually what you
[92:31] what what the robot needs to do or just
[92:33] telling it what to do.
[92:35] So,
[92:36] So, I think that's a key advantage that
[92:38] we have. And then, we also
[92:40] are good at designing things for
[92:42] manufacturing and then manufacturing
[92:44] itself.
[92:45] So, the the actuators in Optimus are all
[92:49] custom-designed Tesla actuators.
[92:51] So, we designed the the the electric
[92:54] motor, the gearbox, the power
[92:55] electronics, obviously the battery pack,
[92:57] and everything else that goes into
[92:59] Optimus.
[93:01] We're actually quite we were quite
[93:03] surprised to find how little
[93:06] was available off the shelf. Because
[93:08] there's a lot of
[93:09] vast number of electric motors
[93:12] um
[93:12] gearboxes and whatnot that are available
[93:14] in the world, and we found none of them
[93:17] were useful in a in a humanoid robot,
[93:19] literally none.
[93:21] So,
[93:23] you have to custom-design the
[93:27] actuators for a humanoid robot. Um
[93:32] And so, the same team that designed the
[93:34] groundbreaking uh
[93:36] electric motors that are in the say the
[93:38] Model S Plaid designed the actuators in
[93:42] the robot.
[93:43] Um
[93:45] So,
[93:46] I mean, for practical purposes, what
[93:48] this means is that we should be able to
[93:49] bring
[93:50] an actual product to market at scale
[93:52] that is useful um
[93:56] far faster than any anyone else.
[93:58] Um
[94:00] And you know, assuming that things I'm
[94:01] saying are true, uh
[94:03] or at least you know, put it I think
[94:05] they are true. You can just It's just a
[94:07] question of the timing.
[94:08] Um
[94:10] You start getting into interesting
[94:11] questions of like, what's the ratio of
[94:13] humans to humanoid robots?
[94:16] I think it might be greater than one to
[94:17] one.
[94:20] You know, cuz you could see you could
[94:21] sort of see a use a home use for robots,
[94:25] certainly industrial uses for robots, uh
[94:27] humanoid robots.
[94:29] Um
[94:30] I think I think we might exceed a
[94:32] one-to-one ratio of
[94:34] humanoid robots to humans. Um
[94:39] It's not even clear what an economy
[94:41] means at that point.
[94:42] You know,
[94:43] since an economy is output per person
[94:45] times
[94:46] persons, but if output is much higher
[94:49] and there's no limit on persons, then
[94:53] what's the actual limit on the economy?
[94:56] You know.
[94:57] We're we're still pretty far from
[94:59] Kardashev scales here, but uh
[95:03] we're getting there.
[95:04] So, anyway, uh it's a
[95:07] It's a
[95:08] probably the least understood or
[95:10] appreciated part of what we're doing at
[95:11] Tesla, but we'll probably be worth
[95:13] significantly more than the uh
[95:15] the car side of things long term.
[95:20] So, charge
[95:20] I think Rebecca's next for talking about
[95:22] charging.
[95:25] Charge.
[95:36] Thanks.
[95:40] Hi there. My name is Rebecca Tinucci and
[95:43] I lead our charging infrastructure teams
[95:45] here at Tesla.
[95:47] And at Tesla charging, we have
[95:48] understood since day one that a great
[95:51] charging experience is the linchpin to
[95:53] electric vehicle adoption.
[95:55] And that understanding has meant that
[95:56] we've always taken a holistic approach
[95:58] to charging. That's a word that you've
[96:00] heard a lot here today, but what it
[96:02] means for charging is that we've
[96:03] considered every use case. We think a
[96:05] lot about what it means to charge at
[96:07] home, even if that home is an apartment
[96:08] or condo.
[96:10] And we spend a lot of time thinking
[96:11] about what does it mean to charge away
[96:13] from home, including if that's for daily
[96:15] commuting or if you're going on a road
[96:16] trip.
[96:17] And this holistic approach has led to
[96:19] some pretty incredible results. In 2022,
[96:22] we provided 9 terawatt hours of charging
[96:25] across our various charging methods.
[96:27] Over 50% of that was supplied via
[96:30] convenient AC home charging.
[96:33] And when our customers are away from
[96:34] home, they can visit one of our 80,000
[96:36] charging points. That includes 40,000 of
[96:39] our beloved superchargers.
[96:42] Now, getting here has meant that we've
[96:44] spent 10 years building charging
[96:46] infrastructure when basically no one
[96:47] else in the industry would do it. And
[96:49] while we certainly have a lot of areas
[96:51] that we want to improve, those 10 years
[96:53] have afforded us the opportunity to get
[96:55] pretty good at charging.
[96:58] First, we have the industry's lowest
[96:59] deployment costs. Our costs are often
[97:01] 20%, if not 70% lower than alternatives,
[97:06] and that goes for both our supercharging
[97:08] hardware and deployments, and our AC
[97:10] charging product lines.
[97:12] A lot of reasons that we're able to
[97:13] achieve this, you've heard a lot of them
[97:15] here today. We're vertically integrated,
[97:16] we manufacture and engineer all of our
[97:18] own charging equipment, we share
[97:20] components across our different product
[97:22] lines, and on the supercharger side, we
[97:25] also install and operate all of our own
[97:28] sites.
[97:29] And that's led us to be pretty obsessed
[97:31] with finding new ways to innovate around
[97:33] installation.
[97:35] As an example, we have recently extended
[97:38] our excellence in manufacturing to how
[97:40] we build supercharger sites. We are
[97:42] pre-building four post supercharger
[97:45] units at our factory in Giga at our
[97:47] factory in New York. We load them on a
[97:49] truck. We truck them to site and then we
[97:51] crane them into position. This method
[97:54] saves us 15% on our deployment costs and
[97:56] we can install a site in a matter of
[97:58] days.
[98:02] Once we install a site, we also operate
[98:04] it really efficiently. Over the last few
[98:06] years, we've been able to cut our per
[98:08] kilowatt-hour cost by 40%. Again, a lot
[98:12] of reasons for this, but one of the top
[98:14] of the list is we focused on increasing
[98:15] our site utilization.
[98:17] Site utilization is just how many
[98:18] sessions or kilowatt-hours can we push
[98:20] through a site or a post? And basically
[98:24] doing that allows us to spread our cost
[98:26] over more sessions, thus lower cost per
[98:28] kilowatt-hour.
[98:29] But of course that's easier said than
[98:31] done because when we push up site
[98:33] efficiency, of course the risk is that
[98:35] we we have a poor customer experience
[98:37] and we have wait time at our sites.
[98:40] This is where Trip Planner comes in.
[98:42] Trip Planner is our in-vehicle routing
[98:44] or navigation system.
[98:46] Other electric vehicle manufacturers,
[98:48] well some of them, have vehicle-side
[98:50] data. Other infrastructure providers
[98:53] have site data, but at Tesla we have
[98:55] both. And what that allows us to do is
[98:57] to use Trip Planner to route vehicles
[98:59] towards available sites and away from
[99:02] congested sites so we can balance our
[99:04] utilization without risking wait time.
[99:06] And the results speak for themselves.
[99:08] Over the last few years, we've been able
[99:09] to drive up site utilization by 30%.
[99:12] That means lower per kilowatt-hour cost
[99:15] while also cutting our wait time in
[99:17] half.
[99:18] And we think this can get even better.
[99:19] Today, we're feeding Trip Planner with
[99:21] real-time site information and
[99:23] information about vehicles currently at
[99:25] sites.
[99:26] Going forward, we'll be moving that to
[99:28] projecting site occupancy based on the
[99:30] understanding of what vehicles are
[99:31] currently routing to those sites. And
[99:34] ultimately, our vision for Trip Planner
[99:36] is that it's the air traffic controller
[99:38] for electric vehicle charging across all
[99:41] infrastructure on a global basis.
[99:44] And the last thing we focused on to get
[99:46] here is quicker charge times. We're
[99:48] really proud of what we've been able to
[99:49] accomplish here. We've shaved off 30% of
[99:52] our charge time, average supercharging
[99:55] sites time, visit time over the last few
[99:57] years. This has taken improvements on
[99:59] the hardware side, with our software, on
[100:02] our vehicles, and on our infrastructure,
[100:04] and we're really excited to continue
[100:06] pushing this trend down.
[100:10] Looking forward, the job does not get
[100:13] any easier. As you heard earlier, we
[100:15] aspire to a fully electrified global
[100:18] fleet. That fleet, uh from an industry
[100:20] standpoint, needs 9 petawatt hours of
[100:23] charging on an annual basis. And while
[100:26] Tesla charging certainly doesn't have to
[100:28] supply all of that, it does require
[100:30] that, for our part, we have a few new
[100:32] focuses in order to scale.
[100:36] First, if you want a fully electrified
[100:38] global fleet,
[100:39] all of these vehicles have to have a
[100:41] great and reliable charging experience.
[100:43] So, as many of you know, recently we
[100:46] have started opening opening up our
[100:48] networks on a global basis. Over 50% of
[100:52] our superchargers in Europe are
[100:53] currently open to other electric
[100:55] vehicles. We've also opened up in Asia
[100:57] Pacific with our first sites in
[100:59] Australia. And just yesterday, we opened
[101:02] our first 10 supercharger sites here in
[101:05] North America in the US to other
[101:06] electric vehicles.
[101:08] We've also invested a lot to be able to
[101:10] make this a really easy experience for
[101:13] our new customers. All you have to do is
[101:14] sign into our Tesla app to unlock a post
[101:17] and start charging. We've also, uh on
[101:20] physical site changes, we've also added
[101:22] hardware to our sites. They're called
[101:23] magic docks. And basically, that uh is
[101:26] installed in areas where we have
[101:27] different charging standards, and they
[101:28] allow for other electric vehicles to
[101:30] come to our site without having to bring
[101:32] their own adapters or hardware.
[101:34] We've also We've started installing our
[101:36] fourth-generation Supercharger posts.
[101:38] Those are being installed in Europe
[101:40] first, and while it's not a a big
[101:42] mention, not a big thing to mention,
[101:44] they do have longer charging cables so
[101:45] that we can more easily reach the charge
[101:47] ports of different vehicles.
[101:50] The second thing we need to make sure we
[101:52] do when we talk about scaling all of
[101:54] this charging infrastructure is we need
[101:56] to make sure that it's powered from
[101:58] renewable sources. We're very proud that
[102:01] over the last 2 years we've procured
[102:03] enough renewable energy to offset the
[102:05] amount of charging we've provided to
[102:07] customers, but as we look forward and we
[102:09] talk about this fully electrified fleet
[102:11] on a global basis, we really want to
[102:13] make sure that the demand for charging
[102:16] more closely follows when renewable
[102:18] sources are available. Now, this chart
[102:20] is on an aggregate basis for the US, and
[102:22] it varies based on what your renewable
[102:24] generation sources are locally, but
[102:26] basically, uh what this shows is we want
[102:29] to kind of move that charging curve and
[102:31] for solar and wind in this example, it
[102:34] means more daytime charging. And we
[102:36] think the best way to go towards daytime
[102:38] charging is to can install into install
[102:41] AC charging that is convenient and
[102:44] low-cost everywhere vehicles are
[102:46] typically parked throughout the day.
[102:48] So, that's really the plan. While
[102:49] Supercharging and home charging
[102:51] definitely stay a big part of the
[102:53] puzzle, our teams are currently scaling
[102:55] to install AC charging everywhere
[102:57] vehicles charge during the day so we can
[102:59] power them from a new renewable sources.
[103:04] So, summing it up, while we've spent 10
[103:06] years installing infrastructure that is
[103:08] low-cost, it's efficient, and it and it
[103:10] provides a great customer experience,
[103:12] we're really just getting started when
[103:14] we talk about supporting Master Plan
[103:16] Part 3. We've got to scale capacity on
[103:18] an industry-wide basis to 9 petawatt
[103:21] hours on an annual basis.
[103:24] Uh we've got to open up to non-Teslas
[103:25] and make the charging experience for
[103:27] everybody really great, and we've got to
[103:29] power all all via renewable sources.
[103:34] And one more thing, yes, we've got to
[103:36] scale
[103:38] uh we've got to scale our infrastructure
[103:39] and yes, we want to power it via
[103:41] renewable sources, but we're Tesla. So,
[103:43] we also want to make sure that we're
[103:44] continuing to focus on providing really
[103:47] incredible charging experiences.
[103:50] With that, want to welcome up our
[103:52] leaders of supply chain, Roshan and
[103:54] Karn, to talk a little bit about that
[103:55] topic.
[103:59] [applause and cheering]
[104:05] Thanks, Rebecca.
[104:10] Uh
[104:11] Hi, my name is Karn. And my team leads
[104:14] uh supply chain for uh electronics,
[104:17] power train, and battery at Tesla.
[104:19] And we've also got responsibility for
[104:21] indirect purchasing, construction
[104:22] procurement, and warehousing and
[104:24] distribution.
[104:25] Hi, my name is Roshan. I've been with
[104:28] Tesla for over 14 years.
[104:30] My team manages
[104:32] vehicle commodities, solar, logistics,
[104:35] planning, and capital equipment.
[104:38] 10 years, forgot that.
[104:40] Um so, today we wanted to provide you a
[104:42] quick overview of Tesla supply chain and
[104:44] what really separates us from the
[104:46] typical uh automotive supply chain.
[104:48] As you've seen in presentations prior
[104:50] from Pete and Drew and some others,
[104:53] Tesla designs a lot of the subcomponents
[104:55] that go into our vehicles. So, we're not
[104:57] buying things that are off the shelf
[104:59] from suppliers. So, supply chain is not
[105:01] purely a commercial relationship as it
[105:03] is in most other companies.
[105:05] We actually have an arm within supply
[105:07] chain called supply industrialization
[105:08] engineering and the burden of taking a a
[105:12] drawing, a design from concept, and
[105:14] turning it into thousands of products
[105:16] produced at the right cost and at the
[105:18] right yield falls with this group. Of
[105:20] course, they work very closely with
[105:21] design engineering, but really the level
[105:24] of detail that our engineers get
[105:26] involved in terms of building the
[105:27] capability of the suppliers is is is
[105:29] pretty detailed. So, you know, once we
[105:33] get a new part, naturally, with electric
[105:35] vehicles, a lot of the supply chain
[105:37] doesn't exist.
[105:38] The our our engineers will basically
[105:39] take the drawing, turn it into a
[105:41] manufacturing concept, do the equipment
[105:43] selection, and physically go to the
[105:45] supplier and stay there for weeks or
[105:47] months, however long it takes, to
[105:49] basically do the line bring up. Once the
[105:51] line is brought up and we're hitting the
[105:53] the right rate, the right quality, at
[105:55] the right yield, then we work on things
[105:57] like automation and and yield
[105:59] improvement and and those types of
[106:00] activities.
[106:02] So, there's there's a big group of
[106:04] mechanical, electrical, industrial, you
[106:06] name it, engineers within supply chain,
[106:08] and it's their responsibility to take
[106:10] care of this. Now, this is a huge
[106:13] strategic advantage because we manage
[106:16] every detail of our supply chain,
[106:18] because we're so integrated with our
[106:19] partners, that we know about issues
[106:22] before they happen. It's almost live.
[106:23] We've got dashboards, we've got
[106:25] connections, and across the 10, you
[106:27] know, 10,000 or so factories across the
[106:28] world that are manage making our
[106:30] components, we have a pretty good view
[106:32] of the health of those suppliers. And
[106:34] this has been a key asset in in how
[106:36] we've been able to kind of manage things
[106:38] through a lot of different issues over
[106:40] the past couple of years.
[106:41] So,
[106:43] um
[106:44] you know, supply chain is a game of
[106:46] perfection. Perfection is a passing
[106:48] grade. It's like a you know, if one part
[106:51] doesn't show up, that line goes down,
[106:53] and our CEO finds out about it in less
[106:55] than 20 minutes. Tesla's very different
[106:57] than Apple. They outsource
[106:58] manufacturing, we do it in-house. So,
[107:00] supply chain's kind of right in the
[107:01] middle. Uh and and perfection is is
[107:04] increasingly elusive and very expensive
[107:07] if you kind of look at
[107:09] all the the the issues we've had to
[107:12] encounter over the past year. And and
[107:14] today we're just going to talk about how
[107:15] we got through the past, and then what
[107:18] are we going to do going forward, and
[107:20] how are we going to get there.
[107:21] So, let me draw your attention to this
[107:23] slide. Um On the left, you will see
[107:27] a portfolio, our existing portfolio of
[107:29] products, right? At the top, you've got
[107:31] the S and X platform. In the middle,
[107:33] you've got 3 and Y. I think everyone's
[107:35] more than familiar with with uh those.
[107:37] And then the Tesla Energy platform,
[107:39] which is a very high
[107:41] rate of growth for us.
[107:43] And you see
[107:44] Don't forget the solar and
[107:45] superchargers.
[107:46] And solar and superchargers. Thank you,
[107:47] Roshan. And
[107:49] you'll see a reduction uh in So, and and
[107:52] then we basically got the tier one and
[107:54] tier two parts. This might be boring,
[107:56] but a tier one supplier is any supplier
[107:59] that builds and supplies parts directly
[108:01] for consumption in our five factories.
[108:04] And our tier two supplier basically does
[108:06] the same thing for a tier one supplier.
[108:07] So, they are the supplier to our
[108:08] supplier. And depending on the supply
[108:11] chain of uh any one of these products,
[108:13] this can actually go down to the tier
[108:15] six levels. For example, if you're
[108:16] talking about the battery cell, we
[108:18] really get involved with the mines of
[108:20] like where we're procuring the lithium,
[108:22] the cobalt, and all that sort of stuff
[108:24] because we're procuring such high
[108:25] volumes of it. So,
[108:27] you'll see a reduction in the number of
[108:29] parts on the S and X platform uh to the
[108:32] 3 and Y platform. And and this is
[108:34] because of the great work our design
[108:36] engineers have done of making the car
[108:38] simpler and easier to manufacture. In
[108:40] parallel, we actually had a similar
[108:43] strategy in supply chain. We had cast a
[108:45] wide net to a lot of suppliers for the S
[108:47] and X platform. There's a lot of
[108:48] suppliers that we invited to partner
[108:50] with us. And we used that
[108:54] platform as sort of a filter to see
[108:56] which suppliers have the technical,
[108:59] financial, and cultural capabilities to
[109:01] match us, right? Who who Like, who do we
[109:03] invite to the party when we really scale
[109:05] things up? And we you know, we we really
[109:08] dramatically reduced the number of
[109:10] suppliers that we were dealing with for
[109:11] the 3 and Y platform. And this is an
[109:13] approach we'll continue to do for future
[109:15] platforms as they come along that are
[109:17] even more high volume. So, this this has
[109:19] really been an approach. It's made
[109:21] things
[109:22] a lot easier and it it you know, we we
[109:25] could fit we could figure out the
[109:26] suppliers that were capable of of moving
[109:28] at our pace and and then the ones that
[109:30] didn't have the desire to or the
[109:31] capability to. You need both desire and
[109:33] capability.
[109:35] Um, secondly,
[109:37] the complexity on the tier two level is
[109:40] is very high, right? These are the
[109:42] suppliers to our suppliers and there
[109:44] there a lot of components. So, think
[109:46] silicon, think about resistors,
[109:47] capacitors, diodes,
[109:49] all the little assemblies.
[109:51] Raw materials, all the that's even
[109:52] further down.
[109:53] Yeah.
[109:54] But all these little subassemblies that
[109:55] have to come into Tesla,
[109:57] the the complexity is immense. So, even
[110:00] though at the tier one level we're
[110:01] talking about 8,000 parts total, it
[110:03] seems like a lot but it's not a lot. The
[110:06] management of the tier two is really
[110:08] where we excel and I'd like to
[110:10] illustrate that with an example.
[110:13] Meet the car computer.
[110:14] Very innocent sounding name and it's
[110:16] anything but. This is an absolute
[110:19] monster to manage. So, I think you've
[110:22] all seen the picture on the left of the
[110:24] autopilot board. That's the top of the
[110:26] autopilot board. The autopilot board
[110:27] also has a bottom and the bottom is
[110:29] populated heavily with components.
[110:32] On the other side of the heat sink,
[110:34] you've got the MCU, the multimedia
[110:36] cluster board. This board is equally
[110:38] complex. It's also double-sided,
[110:40] eight-layered and these boards these
[110:42] computers run so hot at peak operation
[110:44] that they have to be liquid liquid
[110:46] cooled through a heat sink. So, this
[110:48] assembly requires taking those two
[110:50] boards and then bonding them to a heat
[110:53] sink that's hermetically sealed as an
[110:55] assembly. Of course, flashing the
[110:57] software and all that sort of stuff and
[110:59] then that's one part number that comes
[111:01] to Tesla. So, this is an example of one
[111:04] tier one part number. That's a very
[111:06] complex assembly to manage at the tier
[111:07] two level.
[111:08] And
[111:10] there's more than 7,000 components here.
[111:12] There's a you know, as we stand here a
[111:15] component is being assembled onto a car
[111:16] computer every 1.4 milliseconds.
[111:18] The line that builds this computer is
[111:21] the length of a football field. So, it's
[111:23] it's quite complicated. And initially
[111:25] when we first started building this
[111:26] board due to the complexity of it, we
[111:28] had to rely heavily on labor. But once
[111:30] we dialed in the quality, the rates, and
[111:33] the yield, we started focusing on making
[111:35] this more efficient. The only way to
[111:36] control cost is by removing labor. The
[111:38] first step is removing labor. The second
[111:40] step is fully automating. The third step
[111:42] is turning off the lights and letting
[111:43] the factory run, ideally. And that's
[111:45] going to be the goal for us here. So,
[111:46] 95% but our work is not done. We're
[111:49] going to be going a lot further than
[111:50] this.
[111:51] So, this illustrates the point that
[111:54] a part is not a part. A part has a lot
[111:56] of complexity underneath it. Supply
[111:58] chain is is a game of multiple tiers.
[112:00] And and what's made us successful is our
[112:03] involvement in all the details with our
[112:05] supply engineers uh that are Tesla
[112:07] employees. They're Tesla supply
[112:08] engineers. So, they're like on our
[112:09] payroll that go and ramp up these
[112:11] capabilities at our suppliers. And then
[112:13] just managing each and every attribute
[112:14] of it. You know, 7,000 a day through the
[112:17] chip shortage, through the pandemic,
[112:19] through all the other stuff that we had
[112:20] to deal with was was very difficult to
[112:23] manage. But because we had all the
[112:25] details, we were able to pull this off.
[112:27] Um of course, supply chain is not just
[112:29] about parts. It's also about logistics
[112:31] and a lot of other things. So, I'll hand
[112:32] it off to Roshan
[112:34] for sharing some stats.
[112:38] Yeah. So, talking about inbound
[112:39] complexity.
[112:41] Just over the course course of just last
[112:43] year, we moved about 16 million pallets
[112:46] uh from our suppliers to our factories.
[112:49] Just to visualize that, if you put those
[112:51] pallets side by side,
[112:53] uh it could cover the
[112:55] uh half the circumference of the Earth.
[112:59] Uh and talking about again the tier two
[113:01] complexity,
[113:03] over 1 billion electronic components get
[113:06] moved every week to support the
[113:08] production that Karn and team manages.
[113:10] This is right now, 1 billion components
[113:12] a week, 52 billion a year. This is going
[113:14] to go up, by the way.
[113:15] And and on top of that, we also have the
[113:17] responsibility to make sure the right
[113:20] component is supplied to the right
[113:22] service location to support the growing
[113:25] fleet of our vehicles. And we have about
[113:27] 685
[113:29] uh service centers that these components
[113:31] need to show up just in time.
[113:33] Um
[113:35] And then, we've also been uh
[113:37] as as a as uh as a result of pandemic
[113:40] and even before that, we're working
[113:42] working on dual sourcing and triple
[113:45] sourcing some of our critical
[113:46] components. As a result of it, we have
[113:48] suppliers around 45 countries in the
[113:51] world.
[113:51] Yeah, approach really has been to bring
[113:54] the manufacturing of subcomponents
[113:55] closer to the point of consumption. It
[113:57] makes the supply chain more green cuz
[113:58] you're not burning diesel, moving stuff
[114:00] around. And it also gives us the
[114:02] security of supply of having multiple
[114:03] sources in case one factory is taken
[114:05] offline.
[114:05] And that's how we got through pandemic,
[114:07] also.
[114:07] Yep.
[114:10] [clears throat]
[114:10] So, last 3 years were incredibly hard.
[114:14] Karn and I used to look lot younger
[114:17] pre-COVID.
[114:19] But on a serious note,
[114:22] pandemic really changed the game of
[114:24] supply chain management.
[114:26] We were just not being a regular supply
[114:28] chain person. We're rolling up our
[114:30] sleeves and negotiating with
[114:33] state governments, city managers, mayors
[114:37] to help kick-start our suppliers to
[114:39] restart their operations.
[114:42] You know, there are some really
[114:43] inspiring stories where you our supply
[114:46] chain team really fought hard to make
[114:49] sure that the factories don't don't go
[114:51] down. And and it's also a testament to
[114:54] the resilience of our supply partners.
[114:57] Having said that,
[114:59] you know, it was not all about
[115:01] resilience and
[115:03] never give up attitude. We had also laid
[115:06] the foundation of having dual sources
[115:10] and also
[115:11] um
[115:12] you know, in many instances we had uh
[115:15] many uh
[115:17] you know,
[115:18] uh
[115:19] things that that the that uh we had
[115:22] simplified the the parts that we had
[115:24] simplified, the supply chain design that
[115:26] we had simplified due to which we were
[115:28] able to get through the pandemic in much
[115:31] better shape than rest of the automotive
[115:33] uh
[115:33] It forced us to adapt. You know, we we
[115:35] encountered issues one at a time. Uh it
[115:37] was really a team effort. We heavily
[115:39] relied on the government relations team.
[115:41] You literally had to the co- COVID
[115:44] pandemic as as you're aware, you know,
[115:45] happened in different countries at
[115:46] different times. But unfortunately, you
[115:48] need all the parts to make the car. So,
[115:50] at any given point there was one factory
[115:52] that was lying down. There was one
[115:53] jurisdiction that was being, you know,
[115:55] that didn't want to they wanted to shut
[115:56] down a factory because they had an you
[115:57] had an outbreak. So, learning the rules,
[115:59] understanding, coming up with argument
[116:01] points, leveraging the team, and getting
[116:03] those It was a It was a difficult but
[116:05] good experience.
[116:06] Exactly.
[116:06] We learned a lot.
[116:07] Yeah, and as we were just figuring out
[116:09] to work with COVID, then the port
[116:11] congestion hit and on the West Coast, uh
[116:14] logistics lead time went up by, you
[116:16] know, twice the twice the amount and we
[116:18] were operating with less than 35%
[116:21] accuracy in the system. That means at
[116:23] any given point the ETAs of the vessels
[116:26] were 75% off.
[116:29] So,
[116:30] we figured out how to work with that and
[116:32] then the chip shortage hits.
[116:34] Yeah, the chip shortage. That was a
[116:36] nightmare. Uh
[116:38] but we got through it. I mean, you know,
[116:39] I think uh
[116:41] it's a testament to the tenacity of the
[116:43] team and the resilience of our supplier
[116:44] partners. Uh
[116:46] it's mostly behind us and I think the
[116:48] results speak for themselves. Uh Tesla
[116:50] was able to grow production volumes uh
[116:52] at a time where the rest of the industry
[116:54] had a hard time staying flat. I think
[116:56] they shrunk, actually, uh cumulatively.
[116:59] Uh and you know, we're often asked the
[117:00] question, how do we manage through it?
[117:01] And and the answer is very much kind of
[117:03] baked through the presentations today.
[117:05] First, we design our own electronics.
[117:07] So, in order to control cost, we really
[117:09] had to go and build relationships with
[117:12] all the tier two suppliers uh to
[117:13] negotiate them directly. So, these
[117:15] relationships were formed, you know,
[117:17] over 10 years ago, and and we've really
[117:19] gotten to know them well. So, when a
[117:21] shortage hit, we knew, you know, who to
[117:23] escalate to, who to call, who to ask for
[117:25] help, and and uh really kind of swarming
[117:27] the issue at at the point of constraint.
[117:30] So, the best team, you know, strong
[117:32] relationships with vendors, and then uh
[117:34] we also have a pretty sophisticated
[117:36] capacity planning function.
[117:38] Tier two capacity needs to come on
[117:40] earlier than tier one capacity. Tier
[117:42] three capacity has to be even earlier.
[117:44] And our team does a very good job of
[117:46] having that strategic alignment with all
[117:47] our partners to make sure that the
[117:49] capacity is ready before we need it.
[117:51] We're a mission-based company. Our goal
[117:53] is to put as many EVs out there. We did
[117:55] not let the pandemic phase us. We were
[117:57] full throttle. And it was pedal to the
[117:58] metal in terms of our demand signal.
[118:00] That also helped. Uh the the silicon
[118:03] industry has very long lead times. They
[118:04] invest billions of dollars in uh
[118:06] sophisticated fabs that take years to
[118:08] build. Uh so, that stability and demand
[118:11] is something they appreciated, and it's
[118:12] something that is very uh very much a
[118:14] Tesla thing.
[118:16] Um so,
[118:17] this was looking back, right? This is
[118:20] how we got here. This is how we got from
[118:22] zero to 40,000 cars a week.
[118:24] Let's talk about about how we get to 20
[118:25] million.
[118:27] So, the question gets asked, you know,
[118:28] is Tesla going to basically break the
[118:30] chip industry if we grow to 20 million
[118:32] vehicles? The answer is no. But, let me
[118:34] walk you through our logic on that. So,
[118:36] as it stands right now, a Tesla enabled
[118:40] with a full self-driving computer has
[118:42] four times the silicon content of a
[118:43] regular vehicle.
[118:45] Uh
[118:45] so, at about the 2 million run rate,
[118:47] we're consuming about 700,000
[118:49] 12-in wafer equivalents. That's the
[118:51] industry term that's thrown around for
[118:53] capacity in silicon.
[118:54] That's how much we're using, so a little
[118:56] bit below a million. And the global
[118:58] wafer capacity based on the estimates
[118:59] we've seen is about 135 million. So,
[119:02] we're about half a percent of the
[119:03] industry. Not a lot.
[119:06] Now, if you get to 20 million, when we
[119:07] get to 20 million, and if we don't make
[119:10] any simplifications in terms of part
[119:12] count, which is also a separate work
[119:14] stream, I think Colin talked about it,
[119:15] Pete talked about it. We're not going to
[119:17] have all these different components.
[119:18] We're really going to simplify the
[119:19] architecture because simpler is better.
[119:21] But even if you don't do that, and we
[119:23] take the most pessimistic case, we're
[119:25] going to need 8 million wafers, okay?
[119:27] And the industry is going to scale to 20
[119:29] to 200 million by that point. So, still
[119:31] we're less than 5% of the market share.
[119:33] And if we run the same offense we ran
[119:35] all along and continue to work with our
[119:37] partners and and you know, invest in
[119:39] fabs and invest in capacity,
[119:41] uh we'll have the capacity ready by the
[119:43] time we need it. So, I really don't see
[119:44] this as an impediment uh from a supply
[119:47] chain perspective to get to 20 million.
[119:54] Now, I would like to talk about a page
[119:57] that we took out of the electronic
[119:58] playbook and applied it to the
[119:59] mechanical side.
[120:01] Our vehicles use currently the heat
[120:04] pump-based technology for the thermal
[120:05] system.
[120:06] But before heat pump, our vehicles used
[120:09] to have a dispersed thermal system
[120:11] connected by hoses.
[120:13] Engineering was inspired from circuit
[120:15] circuit boards and then created a system
[120:19] which was extremely integrated where 60%
[120:23] of the system was the size of a
[120:25] basketball si-
[120:27] basketball where it contained about 100
[120:29] components, 50 uh sealing interfaces,
[120:33] and several several different
[120:35] manufacturing processes all were
[120:37] packaged into this tight tight size.
[120:40] Now, why are we talking about this?
[120:43] Well, we launched the system right
[120:46] during the outbreak of the pandemic in
[120:48] Marawi.
[120:50] The team was forced to manage 100 plus
[120:52] components from different suppliers
[120:55] around the world remotely.
[120:59] So,
[121:01] there are four videos playing at the
[121:02] same time. One is the video of the
[121:05] manual line that was created
[121:07] uh when we started the production. And
[121:09] this manual line
[121:11] produced a lot of quality issues, let
[121:13] alone throughput.
[121:15] So, we quickly
[121:16] uh you know, pivoted to a semi-automated
[121:19] line.
[121:20] But in spite of that, although the three
[121:22] throughput was uh great,
[121:25] uh the quality issues sustained.
[121:28] So, our team decided that fully
[121:31] automated is the only way to go.
[121:34] So,
[121:35] before we automate the line,
[121:38] we created a 3D simulation. Now, not
[121:41] many companies or teams have this
[121:44] capability. Where at this point, now
[121:47] when a part gets specked, we have the
[121:50] ability to create a 3D simulation of all
[121:53] the complex sub-assemblies that would be
[121:56] involved in making the part, and tell
[121:58] the suppliers exactly what equipment to
[122:01] purchase, what to test, and and and and
[122:04] create an end-to-end line in simulation.
[122:06] And that's what we did here. So, with
[122:09] the help of the simulation, uh we had a
[122:11] very detailed uh sub-assembly cell,
[122:14] which was juxtaposed with the with the
[122:17] suppliers factory layout.
[122:20] And then, we uh after a back and forth
[122:24] of uh iteration and optimizing the
[122:27] processes, we implemented this fully
[122:30] automated line in less than 8 months.
[122:34] And the results were remarkable.
[122:37] 99% reduction in labor force.
[122:42] So, and a line that would need thousand
[122:44] associates to build the super manifold
[122:47] in full volume, now just takes 10 10
[122:51] associates.
[122:53] And the quality also went up by an order
[122:55] of magnitude.
[122:57] Uh we have a defect rate of less than
[122:59] 0.005%
[123:01] and a super manifold rolls out every 7
[123:05] seconds because we have created many
[123:07] automated lines and these factories are
[123:10] the size of two football fields, by the
[123:12] way.
[123:14] Rohan had to one-up me on that one.
[123:15] [laughter]
[123:16] 99% two football fields.
[123:18] Uh
[123:19] so, why are we talking about this?
[123:22] In a world where labor costs are
[123:25] increasing,
[123:26] labor participation is decreasing, there
[123:28] is a huge turnover,
[123:31] and the cost of implementing automation
[123:34] is going down by 3x in the next 10
[123:36] years.
[123:37] This is how we are thinking about
[123:40] complex assemblies and we are going to
[123:42] push this type of thinking into our
[123:44] supply base more and more who will be
[123:46] building our complex assemblies.
[123:48] This is the only way we think we can
[123:50] confidently scale to the 20 million
[123:52] vehicles target.
[123:57] To summarize,
[123:59] we have laid the foundation in getting
[124:01] ready to execute Master Plan 3.
[124:04] Our strat- strategy is
[124:06] supply chain design simplification.
[124:09] We are going to make sure that we have
[124:12] more control into the tiers of the
[124:14] supply chain.
[124:15] We are going to grow responsibly and
[124:18] sustainably with our long-term partners
[124:21] and we'll automate.
[124:23] Yeah, it's like, you know, going from
[124:24] zero to 40,000 cars a week was was
[124:27] tough. I mean, it took it took a lot of
[124:29] trial and error. It took a lot of
[124:30] learning. There was a lot of pain.
[124:32] There's a lot of mental scar tissue, as
[124:34] Elon calls it, uh uh through that
[124:35] process. But now now that we're at that
[124:38] level and the foundation is set, going
[124:40] from 40 to 400 is not going to It
[124:41] doesn't really phase us.
[124:43] Uh we have a capable team. We've got
[124:45] capable external partners that have, you
[124:46] know, gone through hell with us. Our
[124:48] leadership team is galvanized behind the
[124:50] mission, um and we're going to get it
[124:51] done.
[124:52] So, I'd like to thank you for your time
[124:54] and hand it off to Drew and Tom for the
[124:56] next presentation. Thank you.
[124:59] [applause]
[125:06] Hi, everyone. Um my name is Tom Zhu. Uh
[125:09] I'm now responsible for global
[125:11] production and sales and delivery
[125:12] service. I joined the company back in
[125:14] 2014. Um and um I've been running the
[125:17] company's business in Greater China and
[125:19] APAC. Um today, I'm representing all the
[125:21] Gigafactories and talk about how we can
[125:24] make more cars and faster.
[125:29] So, right now, um we have four uh
[125:31] vehicle factories on three continents to
[125:33] serve our markets. Um the total number
[125:36] of a manufacturing employee is over
[125:37] 65,000. This is also including our
[125:40] amazing team in Reno uh they are making
[125:42] power chains, um battery packs, and
[125:44] drive unit for our vehicle factories.
[125:46] Um the total installed capacity in all
[125:48] these vehicle factories about 2 million
[125:50] cars a year. That said, we always
[125:52] looking for opportunity to grow more
[125:54] capacity from the existing footprint.
[125:56] So, you expect this number will grow
[125:58] over time.
[126:02] Um to build more factories um is to
[126:05] start to build the more cars. And we
[126:07] certainly are expert of it. Um
[126:09] this is a before and an after image um
[126:11] showing you the progress we made back in
[126:13] 2019 when we built the Gigafactory in
[126:16] Shanghai. Um so, in 9 and 1/2 months
[126:19] time, um we uh built um a greenfield
[126:22] project uh to establish a vehicle
[126:24] factory. And the 3 months later, we
[126:26] start deliver our first customer
[126:28] vehicle.
[126:29] Um people always ask me, you know, how
[126:31] Tesla can build a factory that fast?
[126:34] Really,
[126:35] we learned from our Fremont factory a
[126:38] lot. We talked to the survivor from the
[126:40] production hell and try to avoid all the
[126:43] mistakes we made.
[126:45] And we decided to design a straight line
[126:47] with minimum number of up and downs and
[126:49] turns for easier manufacturability and
[126:52] easier construction. We also challenged
[126:54] all assumptions people ever know to
[126:56] build a vehicle factory
[126:57] and we delete and simplified all the
[127:00] redundancies and the buffers that's
[127:03] helped us to save a lot of time.
[127:05] Also, we have a a very strong in-house
[127:09] construction team in Tesla.
[127:11] If there
[127:13] ever a job cannot be done by others
[127:15] better and we bring this in-house. So,
[127:17] we have this in-house construction team
[127:19] have a full control over all the
[127:21] activities on sites from design
[127:24] engineering to construction, site
[127:26] management.
[127:27] So, this team not didn't just build
[127:30] Gigafactory they also built Giga Berlin,
[127:32] Giga Texas, and Giga Nevada.
[127:35] And they're really the hero behind the
[127:36] scenes.
[127:38] Like
[127:39] Lars shared earlier, going forward with
[127:42] the new platform, the more modular
[127:45] design and
[127:47] densified Gigafactory will be able to
[127:50] make more cars from even a reduced the
[127:52] footprint. That also means we can build
[127:54] more factories at the same time. With
[127:57] the joint effort of
[127:59] manufacturing teams around the globe,
[128:02] I'm happy to share early this morning we
[128:05] hit the 4 million mark
[128:07] for total Tesla ever built.
[128:09] [applause]
[128:13] And the uh
[128:14] the 4 millionth vehicle actually built
[128:17] in this factory. When you took a tour on
[128:19] our shop floor, you probably walked past
[128:21] it. All [laughter] right. So, took us 12
[128:24] years to build the first million
[128:27] Um, and um, about um, 18 months um, to
[128:30] the second uh, build the second million.
[128:32] The third million took us 11 months and
[128:35] just a shortly uh, less than a 7 months
[128:37] we built the fourth million cars.
[128:39] So, we're getting better and faster.
[128:41] Exponential growth.
[128:43] [applause]
[128:44] Really kudos to the team.
[128:47] So,
[128:48] um, what it takes to ramp a Gigafactory?
[128:50] Well, if you have a 600 robots, 10,000
[128:53] trained employees or 5,000 human and
[128:56] 5,000 um, Optimus and hundreds of
[128:59] processes, you can do it.
[129:01] Sounds simple, but it's extremely hard.
[129:03] So, there's two um, key metrics that we
[129:06] predominantly focused on. It is the um,
[129:09] overall equipment effectiveness and the
[129:11] cycle time. Um, in Tesla um, we're
[129:13] setting the passing grade for our
[129:15] vehicle factories um, with 90% OEE and
[129:18] 45 second uh, cycle time. What that
[129:21] means, um, the OEE really evaluate um,
[129:24] the equipment uptime, the um, machine
[129:27] performance and the quality. Simply put,
[129:29] um, this is the um, actual um,
[129:32] production time on a good quality
[129:35] product versus the planned productive
[129:37] time. And the higher the better. Um, the
[129:40] 45 second uh, cycle time
[129:43] that means, you know, you expect every
[129:45] 45 seconds there's a car running off the
[129:47] final assembly line uh, in the factory.
[129:50] Um, and the faster we ramp, um, uh, the
[129:52] faster we can get the economy of scale.
[129:55] Um, if you look at um, the chart on the
[129:57] right on the left, um, Shanghai um, be
[129:59] able to significantly drop our labor
[130:01] hours per car um, during the ramp. Um,
[130:05] the little dip there is a happened in
[130:06] the last um, Q2 2022 um,
[130:09] because of the the COVID shutdown. Um,
[130:12] and on the right is the Fremont um,
[130:14] Model Y shop. Even this is a 6-year-old
[130:17] facility, the team there still be able
[130:19] to optimize material flow, eliminate all
[130:21] the um, single point of failure, and the
[130:23] drive higher output,
[130:25] hence reduce the labor hours.
[130:28] And actually this factory keep setting
[130:30] new record. Yesterday they just had a
[130:33] new factory daily record.
[130:36] Congrats Fremont team.
[130:38] Absolutely.
[130:39] [applause]
[130:43] So,
[130:44] how can we reduce the cycle time,
[130:47] keep improving efficiency?
[130:49] We follow the philosophy
[130:51] Elon shared about building rocket. Which
[130:54] is questioning.
[130:55] Basically find the right problem to
[130:56] solve. And we start with delete,
[131:00] simplify, then we try to accelerate,
[131:02] pressurize the line, and find the if the
[131:05] the solution actually work. And at last
[131:08] we think about the automation.
[131:10] There's one example in Gigafactory
[131:13] Shanghai. In the paint shop we find that
[131:15] there is a
[131:17] overlap baking range between
[131:19] the
[131:20] PVC sealer and the top coat. It was done
[131:24] by two different ovens, and we decided
[131:26] to combine the two processes.
[131:28] Eventually that helped us didn't just
[131:30] help us to reduce the cycle time, but
[131:32] also save the 9% of energy consumption
[131:35] and the 9% of the
[131:36] CO2 emission. Also we commonized the SNX
[131:41] design in Fremont.
[131:43] Right now we have a common
[131:45] body and common headlamp.
[131:47] We eliminate eliminated about 40 parts
[131:51] and reduce about a 10% of the cycle
[131:52] time. So all these improvements really
[131:56] help us to
[131:58] get to
[131:59] a faster ramp trajectory. And we didn't
[132:03] just learn from these high volume
[132:04] existing factories. We also learn from
[132:06] the new factories.
[132:08] Most recently our Berlin factory team
[132:12] implemented a 5G private network on the
[132:14] shop floor, helped us to
[132:17] overcome about a 90% of the over cycle
[132:20] issues for a particular process
[132:23] in GA shop. We're going to soon
[132:25] implement this globally.
[132:27] Um, so
[132:30] um
[132:31] right now we have an integrated global
[132:34] organization from from production all
[132:36] the way to sales and delivery service.
[132:38] This will help us to strengthen feedback
[132:41] loop between manufacturing and the
[132:43] service. We obviously want to bring a
[132:45] delightful customer
[132:46] ownership experience to our
[132:50] car owners.
[132:51] And
[132:52] with this direct feedback loop we be
[132:54] able to turn customer escalations and
[132:57] the feedback into a quick actions and
[132:58] improvement on the shop floor
[133:00] immediately. Um, with that effort in the
[133:02] past 6 months we've been able to reduce
[133:04] the time in service
[133:06] and early ownership service also service
[133:09] appointment wait time significantly.
[133:13] So,
[133:14] um
[133:15] four factories not enough if we want to
[133:18] hit the 20 million car
[133:20] a year target we're going to keep
[133:21] building new factories and new lines
[133:24] and that will come up come up with new
[133:27] product as well.
[133:29] And with all gigafactories now
[133:32] in one organization we'll be able to
[133:34] replicate
[133:35] unified ideas across factories.
[133:38] Also we will help the new factories ramp
[133:41] faster and also
[133:43] produce a better quality at the lower
[133:45] cost. Um, now I'm going to hand over to
[133:47] Drew to talk about cell manufacturing.
[133:51] Thanks Tom.
[133:52] Yeah, we're not just talking about new
[133:54] vehicle factories but also new cell
[133:55] factories.
[133:57] And and we're going to talk about how we
[133:59] follow the same model when we do do so.
[134:01] But first I thought I'd just provide a
[134:02] little bit of an update
[134:04] on cell production. So, remember battery
[134:07] day we showed this
[134:08] video with the spoon.
[134:10] Uh, and how we went from dry powder to
[134:12] film. Let's just say there's no spoon
[134:15] now.
[134:16] So, many of you in this room saw uh,
[134:19] this on your tour today, but you know,
[134:21] here is our dry electrode machine here
[134:23] in Texas, one of the lines we have
[134:24] installed here. Fully automated, no
[134:26] spoon from powder and foil in to coated
[134:29] electrode out. Um,
[134:31] uh, from from a peak productivity
[134:33] perspective, per tool perspective, this
[134:35] is over 20 times the productivity of the
[134:37] tools that we showed folks on the tour
[134:39] uh, in Kato back in battery day. So,
[134:41] we've made a lot of progress on the the
[134:43] key one of the key parts of the cell
[134:44] manufacturing process.
[134:46] Um, and we've also continued to focus on
[134:49] uh, refining the way we make cells and
[134:52] the factories that the cells go inside.
[134:54] Um,
[134:55] we, as you can see on this chart, from
[134:57] typical 2170 cylindrical cells to 4680,
[135:00] we made a huge leap, which is basically
[135:02] a 5x reduction in the factory footprint,
[135:05] um, and volume and
[135:07] volume and footprint. Um, and then from
[135:10] going from what we did in Fremont in our
[135:12] pilot line to Texas, we improved further
[135:14] and we're improving again when we go
[135:16] into Nevada. And what this actually
[135:18] represents is a a series of actions
[135:21] taken by a very integrated holistic uh,
[135:24] design team across, you know, the
[135:26] product design, the manufacturing
[135:27] design, the process design, equipment
[135:29] design, and the facility design. They
[135:31] all need to work together to make this
[135:32] happen. Um, and you can see on the
[135:34] bottom, just as an example of simple
[135:36] simplicity, uh, up, investment down, and
[135:38] scaling up, you know, parts, we've
[135:40] reduced the number of parts in the cell,
[135:42] uh, the number of processes, significant
[135:44] reduction in the number of processes,
[135:45] um, and and collaborating with those
[135:48] five design groups, we've been able to
[135:50] uh,
[135:51] result, you know, build a end up in a
[135:52] with a factory that's 10 times smaller
[135:55] volumetrically, which means it's uh,
[135:57] faster to get uh, built, it's much lower
[135:59] CapEx per gigawatt hour output, um, and
[136:02] we can go and scale, you know, to our
[136:04] objectives of, you know, 240 total
[136:07] terawatt hours, 1 terawatt a year of
[136:09] stationary storage and 20 million
[136:10] vehicles a year.
[136:13] With the scalability that is required to
[136:15] to achieve our goals.
[136:17] Um
[136:19] and we're not just looking at the cell
[136:20] factory itself, but also upstream
[136:22] materials where necessary. What you see
[136:24] here is a the rendering of the 50
[136:27] gigawatt hour a year Corpus Christi
[136:29] lithium refinery that we've already
[136:30] broken ground on
[136:32] here in Texas. The the facility will
[136:34] start commissioning by the end of 2023
[136:37] this year. This is a good example of of
[136:40] something where we're we're we're
[136:41] basically talking about breaking ground
[136:43] and and and starting commissioning
[136:45] within 10 months and with with with
[136:48] actual production within 12 months.
[136:49] That's the target. Similar to what we
[136:51] did in Shanghai.
[136:53] Again, the result of collaboration and
[136:55] and you know, internal execution of
[136:58] construction
[136:59] in partnership with local communities.
[137:01] 30 This This site is 30 minutes from the
[137:03] Corpus Christi ports located directly on
[137:05] rail.
[137:06] The process route we're taking is a
[137:09] direct soda ash leach of the input
[137:11] material, which means there's
[137:13] no acid roasting. We don't have any of
[137:15] the sort of waste products associated
[137:17] with
[137:19] acid roasting step.
[137:20] We're designed to consume
[137:23] lithium spodumene, which is a very
[137:25] commonly traded lithium rock, but it is
[137:27] flexible to other feeds from primary and
[137:29] secondary sources.
[137:30] And similarly, we're working on our
[137:32] cathode facility here in Texas. Maybe
[137:33] you've seen it as you're driving around.
[137:35] Which is a 60 gigawatt hour year cathode
[137:38] facility behind the main building here.
[137:40] 10 months build time there as well.
[137:43] The equipment's being installed for the
[137:44] first line as we speak and will be
[137:46] commissioning starting next quarter. So,
[137:48] we we are doing this where we need to.
[137:51] Our plan is not to do it always. There
[137:52] are lots of competent companies out
[137:54] there,
[137:55] but we are also trying to sort of
[137:56] accelerate the pace of the industry by
[137:58] trying some new things that are a a bit
[138:00] more scalable. And and de-risking
[138:03] certain innovations that improve
[138:04] productivity per uh in terms of CapEx
[138:06] per gigawatt hour and things like that.
[138:08] Um,
[138:09] and with that, I think that sums up our
[138:11] efforts on the factory side.
[138:13] Um,
[138:14] um, Mike, you want to come up and we'll
[138:15] talk about energy?
[138:16] All right.
[138:18] [applause]
[138:25] All right.
[138:26] Take it away.
[138:27] All right.
[138:28] Here, you can
[138:29] Hello everyone. My name is Mike Snyder.
[138:32] I lead our Megapack organization. I've
[138:34] been with Tesla for almost 9 years.
[138:37] I'm excited to talk to you today about
[138:38] the Megapack product and the business
[138:40] and some of the exciting things we have
[138:41] ahead of us.
[138:44] From the beginning with this business,
[138:46] we have always focused on building
[138:48] successful projects for our customers
[138:51] and not just the batteries in the box.
[138:55] We've built a hardware and a software
[138:56] platform that is able to adapt to
[138:58] environments all over the world
[139:01] and that is able to scale for from small
[139:03] island projects to large gigawatt hour
[139:05] scale batteries.
[139:08] We've invested our time and our talent
[139:10] to best understand every aspect,
[139:13] every step in the process and and every
[139:15] risk of a project
[139:18] to ultimately provide a solution that is
[139:19] that is thoughtful, intuitive,
[139:22] provides great value to our customers,
[139:24] and is as plug-and-play as possible.
[139:28] And yeah, I just looking at these
[139:29] projects, it's so incredible to see them
[139:31] all come together in one in in one
[139:33] montage.
[139:33] Seeing the future.
[139:35] Yeah, yeah.
[139:35] Compressed.
[139:36] It's 10 10 years of building these
[139:37] projects and and it's it's incredible to
[139:39] see the impact and to to see what we
[139:41] have in front of us.
[139:42] Absolutely.
[139:45] We're on our sixth generation of our
[139:46] industrial product, which is the
[139:48] Megapack XL that we're building out of
[139:50] the out of Lathrop, California.
[139:52] We've deployed over 16 gigawatt hours of
[139:54] industrial and residential products
[139:55] across 50 countries.
[139:57] And really, Megapack is a market leader.
[139:59] It's best in class in efficiency,
[140:02] reliability, energy density,
[140:05] and easiest to install, lowest cost to
[140:06] install.
[140:08] And we call it XL because it is actually
[140:11] the like largest, heaviest object that
[140:13] you can
[140:14] transfer around the roads of the world
[140:16] without, you know, having to shut them
[140:18] down and get crazy permits and things
[140:20] like this. So, it is extra large and
[140:22] there is going to not going to be an
[140:23] XXL, let's just say it that way.
[140:26] [laughter]
[140:27] There's been incredible demand for the
[140:29] project or for the product. 2023 is
[140:31] going to be a great year. We have
[140:32] gigawatt hour scale projects being
[140:33] built. Um the Lathrop factory continues
[140:36] to to scale and ramp up and there's new
[140:38] products on the near-term road map. So,
[140:39] a lot to look forward to.
[140:42] How did we get here?
[140:44] Number one, a maniacal focus on all
[140:46] aspects of delivering stationary
[140:47] storage.
[140:49] What do we mean by that? What are some
[140:50] examples?
[140:52] As you see here, the Megapack enclosure,
[140:53] for one example, we designed the
[140:55] wireways directly into the base of the
[140:57] enclosure. So, typically you need to
[140:59] install a bunch of conduits and cables
[141:00] under the ground. Now, you can lift them
[141:02] up and install them directly into the
[141:04] base, lower uh increasing the speed of
[141:06] deployment and lowering the cost.
[141:09] Next, we see the batteries and the power
[141:11] electronics. And the power electronics
[141:12] is really
[141:14] the the beauty and the elegance of the
[141:16] Megapack. So, this is what converts the
[141:17] DC power to AC power on the grid.
[141:20] The brain is built directly into the
[141:21] product.
[141:23] It is able to connect to any grid in the
[141:25] world right out of the box.
[141:28] It allows it to all the Megapacks on a
[141:30] given site to work as one unit.
[141:34] And really, it's it's it's what makes it
[141:36] what it is. So, I think it's the most
[141:37] incredible feature of of of the
[141:38] Megapack. As you build out the full
[141:40] site, what we end up with is the most
[141:42] energy dense solution on the market,
[141:44] upwards of 300 megawatt hours per acre.
[141:47] And just as a frame of reference, this
[141:49] solution is two times more power dense
[141:52] than a typical gas peaker plant.
[141:54] So, th- this is the future. This this
[141:56] this is where we're going.
[141:58] This is the product that retires the
[142:00] fossil fuels.
[142:01] Yep.
[142:01] One power plant at a time.
[142:03] We talk about power electronics a little
[142:05] bit more. Um
[142:06] Tesla is a leader in power electronics.
[142:08] We've deployed over 1.4 terawatts across
[142:11] energy storage and vehicle.
[142:13] And we deliver more power electronics
[142:15] than the solar and the wind industry
[142:17] combined on a per annum basis.
[142:19] And power electronics it's it's hard to
[142:22] under to overstate how impactful they
[142:25] are. They are really the glue in the
[142:26] sustainable energy economy between
[142:29] generation, storage, and the end use.
[142:31] You know, the the the those power
[142:33] electronics devices are switching, you
[142:35] know, thousands of times per second,
[142:36] hundreds of thousands of times per
[142:37] second to efficiently react to whatever
[142:40] is needed uh either in the car or in the
[142:42] grid. Um and because they are so like
[142:45] sort of software driven at their core,
[142:48] they can provide functionality that uh
[142:50] hasn't been available to the grid in the
[142:51] past, and it's one of the reasons why
[142:53] renewables and storage together are are
[142:55] such a great solution.
[142:56] Yep.
[142:58] And one reason that we focus on power
[143:00] electronics and and control so much is
[143:02] because of the impact that it has on the
[143:04] projects directly.
[143:06] So, just as an example here on the left,
[143:08] we have a we had a have a firmware
[143:10] feature that we we call virtual machine
[143:12] mode.
[143:13] And what virtual machine mode mode is is
[143:16] it contributes to grid stability like a
[143:18] car shock absorber that dampens
[143:20] oscillations or vibrations and keeps the
[143:23] ride smooth.
[143:24] So, you can imagine if if you can't if
[143:26] you don't dampen those those vibrations,
[143:27] the vehicle could lose control or or or
[143:30] um someone could get hurt. On the grid,
[143:32] you could have a blackout.
[143:34] We have one grid operator that's
[143:35] utilizing virtual machine mode and said
[143:37] they will not operate their grid at 100%
[143:39] renewables unless they have this feature
[143:41] unless they have virtual machine mode
[143:42] working.
[143:43] Yeah, and
[143:44] They tried and it didn't work very well.
[143:45] Why is it virtual machine mode? It's
[143:47] like synthetic inertia where you turn
[143:49] the battery power plant into
[143:51] like through software behaving like it's
[143:53] a giant spinning machine literally. And
[143:56] and
[143:57] and it and and that like inertia
[143:59] stabilizes the grid.
[144:01] And you don't but you don't need a giant
[144:02] spinning machine. You don't need like a
[144:03] huge fossil fuel power plant or a giant
[144:05] hydro turbine. You can just have the
[144:06] battery do it. And you can program it to
[144:08] whatever you need at that part of the
[144:09] grid or even have it be dynamically
[144:11] changing as the as the grid conditions
[144:12] change. Right.
[144:14] And that's why we focus so much on it.
[144:15] We believe it's the future of how
[144:16] batteries are going to add value. It's
[144:17] going to be more about adding power
[144:18] stability just as it is about energy.
[144:21] Yep.
[144:21] Yep.
[144:22] Autobidder is another software feature I
[144:24] wanted to mention. Autobidder is an
[144:25] autonomous energy trading platform that
[144:28] in its most basic sense is
[144:30] buying energy low and selling it high
[144:32] and the the owner can net the difference
[144:33] of that. But operating battery storage
[144:36] is actually very complex.
[144:38] Um upwards
[144:40] needing upwards of hundreds or even
[144:41] thousands of decisions every 5 minutes.
[144:43] It's much more complex than
[144:45] a PV a solar plant or a wind plant or a
[144:47] thermal thermal battery or thermal
[144:50] generation plant.
[144:52] And in its mostly because of the
[144:53] versatility and the ability of the
[144:55] battery to provide so much value in
[144:56] different ways that these decisions need
[144:58] to be made in real time in order to
[144:59] optimize and and get the most value out
[145:01] of the batteries it can.
[145:03] So there wasn't a solution that existed
[145:05] in the market so we built one ourselves.
[145:07] In the market that we've been markets
[145:08] we've deployed them in in Australia and
[145:10] Texas and the UK it has proven to be a
[145:12] market leader. We're continuing to
[145:14] invest in Autobidder as we expand to new
[145:16] markets.
[145:19] Second point of how we got here.
[145:21] A relentless focus on speed of execution
[145:23] and and there's two points to be made
[145:25] here. First one is in building
[145:27] factories. What you're seeing here is a
[145:29] time lapse of the megafactory that was
[145:31] built in Lathrop. We took it from a J.C.
[145:34] Penney distribution center to a
[145:35] world-class manufacturing facility in
[145:38] less than 12 months which is which is
[145:40] incredible. And really the the way that
[145:42] we did this is similar to what my
[145:43] colleagues have described. It's It's
[145:45] leveraging
[145:46] the vertical integration of Tesla. It's
[145:48] It's getting the
[145:49] vehicle manufacturing team in the same
[145:51] room as construction and engineering and
[145:53] making decisions quickly with all the
[145:54] decision with all the stakeholders and
[145:55] decision makers.
[145:57] We're using the lessons learned from
[145:59] Lathrop as we speak as we plan for our
[146:00] next factories.
[146:04] Second, it's about installing projects
[146:06] faster.
[146:07] I've mentioned this already, but
[146:08] plug-and-play is kind of at the core of
[146:10] what we're going for.
[146:12] Over the past 4 years, we've we've
[146:14] increased the installation speed by 4x.
[146:18] And we've reduced the total labor
[146:20] involved across both construction and
[146:22] manufacturing by 3x.
[146:25] We think that this is key to unlocking
[146:27] our ability to scale and our customers'
[146:29] ability to scale with us. We need to
[146:30] make be laser-focused on reducing that
[146:32] time from when the Megapack leaves the
[146:34] factory to when it is operational on the
[146:36] grid.
[146:40] And it's not just about uh
[146:43] centralized storage. It's also about
[146:44] distributed storage. The future uh
[146:46] roadmap of of uh well, at least here at
[146:49] Tesla. Um and and maybe some of you know
[146:51] about this sort of retail plan we have
[146:53] here in in Texas and I just wanted to
[146:55] talk a little bit about how that all
[146:57] fits together.
[146:58] So, Tesla Electric really unlocks the
[147:01] full value of distributed energy and
[147:03] storage products.
[147:04] So, um it enables our customers to
[147:06] become their own utility. The data on
[147:08] the page is from our South Australia uh
[147:12] virtual power plant um another Tesla
[147:14] Electric uh uh setup that we have. Um
[147:17] over the over 2022, it's 5,000 customers
[147:20] that we we have the data for.
[147:22] You can get a sense for what's
[147:23] happening. So, just if they were being
[147:25] provided default utility service and you
[147:26] looked at the cost of serving them that
[147:28] electricity, on average, $140 a month.
[147:31] If they all had solar and Powerwall, but
[147:33] the solar and Powerwall wasn't
[147:34] interacting with the grid, wasn't
[147:35] participating in the energy market,
[147:37] the cost of serving that customer would
[147:38] go from a would would have go from $140
[147:40] a month to $70 a month.
[147:42] If you If Tesla Electric is operating um
[147:45] those assets in an intelligent way to
[147:47] benefit the grid,
[147:49] uh Tesla Electric being basically a
[147:50] software that we developed out of
[147:52] Autobidder for the purposes of of of
[147:53] these distributed energy resources, we
[147:56] can actually
[147:57] pay the customers to bring their their
[148:01] their energy services to the grid.
[148:03] That's what happened in Australia last
[148:04] year.
[148:05] And Australia is a little bit of a
[148:06] special case. They're They represent the
[148:08] future, South Australia in particular.
[148:10] Solar and wind supplied 70% of South
[148:12] Australia's energy in 2022.
[148:15] That compares to 30% in Texas and 35% in
[148:17] California. But it But this is a This is
[148:20] an indication of where this is all
[148:21] headed.
[148:22] Of
[148:23] both centralized and distributed um
[148:26] storage resources
[148:27] uh providing the the the key to
[148:30] unlocking fully renewable grids.
[148:32] And Tesla Electric is the retail plan
[148:34] that we're using to bring that to bear
[148:37] for our our our customers that have our
[148:38] products.
[148:40] Um
[148:42] And you know, this is how we're rolling
[148:43] it out.
[148:44] Uh
[148:45] first uh as it exists, uh it's available
[148:48] for uh in Texas for people who have a
[148:51] Powerwall in their homes.
[148:52] But we need to extend beyond that. Over
[148:54] a billion people live in markets with
[148:56] competitive retail electricity.
[148:58] And there's over $2 trillion in annual
[149:00] energy spend in these markets. This is a
[149:02] huge opportunity.
[149:04] And the our intention is to bring this
[149:07] to, you know, market by market in the
[149:08] same way that we we've approached Tesla
[149:10] Insurance, where we can bring value to
[149:12] our customers to reduce their total cost
[149:14] of ownership of our products.
[149:17] So,
[149:18] actually what we're going to do next,
[149:19] and this is pretty exciting here in
[149:20] Texas, uh by by this summer, we're going
[149:23] to offer a a retail electricity plan to
[149:25] people who have our cars where they can
[149:27] have unlimited overnight home charging
[149:29] for $30 a month.
[149:31] This is part of reducing the total cost
[149:33] of ownership of our vehicles, and the
[149:35] reason why we can do this is because
[149:36] Texas has a ton of wind. And in Texas,
[149:39] the wind blows at night. So, actually
[149:41] serving these customers with electricity
[149:43] at night for their cars
[149:45] is the best thing to do for everyone.
[149:47] And so, this is a way to incentivize
[149:48] people to charge at home at night
[149:50] directly from renewable power. It's part
[149:52] of the grand master plan we talked about
[149:54] at the beginning. Um so, we're we're
[149:56] we're very excited about this. Um and we
[149:58] do see this similar to Tesla insurance,
[150:00] as I said, as further reducing the total
[150:02] cost of ownership of electric vehicles.
[150:05] So, putting this all together,
[150:07] we're really at the beginning of this uh
[150:10] massive ramp in energy storage
[150:11] deployment.
[150:13] And
[150:13] Yeah.
[150:14] Yeah, reflecting back on the on the
[150:15] master plan, we talked about tens of
[150:17] terawatt hour needing tens of terawatt
[150:18] hours of stationary storage. So,
[150:20] we have our sights set on annual
[150:22] production rates of 1 terawatt hour,
[150:23] which is 25x our capacity at Lathrop.
[150:26] Um in the in in in the near term, we see
[150:28] we see really strong demand for the
[150:30] Megapack product over 100 gigawatt hours
[150:32] per year in 2023 and growing by over 100
[150:35] gigawatt hours per year over the next
[150:37] few years. So, the demand will is there.
[150:40] Um and naturally, as we continue to
[150:41] focus on cost and speed and value, the
[150:43] things that we've mentioned, it's clear
[150:44] we need to build more capacity, and we
[150:46] need to ramp it quickly.
[150:48] So, while the challenge is big, it's
[150:49] also a huge opportunity for Tesla, it's
[150:51] a huge opportunity for for the Megapack
[150:53] business. Um and I'm excited to see the
[150:55] impact that it's going to make on our
[150:56] grid transition.
[150:57] As am I.
[150:58] Yeah.
[150:59] All right.
[150:59] All right.
[151:00] Thank you.
[151:00] Thank you all. Yes.
[151:02] [applause]
[151:05] All right, and uh
[151:07] Laurie?
[151:09] Brendan? Come on up.
[151:17] Hi, I'm Brendan with the legal team.
[151:19] Thank you.
[151:20] Not just for coming, but being
[151:22] investors.
[151:24] Now, we have we have an active and
[151:26] engaged board and management team. We've
[151:28] met with you. And today is the
[151:30] culmination of that. We've heard you.
[151:33] Which is why we're excited to share this
[151:35] investor day with you. Now, you will
[151:37] hear more from our board at the
[151:38] appropriate time. But for now, I'm
[151:40] excited to introduce Laurie Shelby who's
[151:42] going to discuss sustainability and our
[151:44] employees. Laurie.
[151:47] [applause]
[151:50] Thank you, Brandon.
[151:52] Hi, everyone. My name is Laurie Shelby
[151:54] and I lead the environmental health,
[151:56] safety, security, and sustainability
[151:58] team.
[151:59] I've been with Tesla for 5 and 1/2
[152:02] years.
[152:03] So, I'm going to talk about Tesla's
[152:05] impact.
[152:08] And of course, that starts with our
[152:12] awesome team members.
[152:14] So, you've heard a lot today about our
[152:16] amazing products, our process, our
[152:19] software. You've seen lots of numbers
[152:21] and lots of graphs. You've also heard a
[152:23] common theme. And that common theme is
[152:26] that Tesla's team is what drives our
[152:28] success and makes a difference.
[152:31] Whether you're in our factories, in our
[152:34] sales, service, and delivery centers,
[152:36] our warehouses, or in the field, you can
[152:40] see and you can feel the collaboration
[152:43] that's happening. And you also can see
[152:47] the pride that everyone has in working
[152:50] for Tesla.
[152:52] So, before I go on, I want to give a big
[152:54] shout out and a thank you for our Tesla
[152:57] team globally. So, let's
[153:02] [applause]
[153:06] That was great. That was great. Thank
[153:08] you. So, we are creating a lot of
[153:11] meaningful work through our jobs.
[153:15] We're a huge employer and we're growing
[153:17] every day. We're at 129k.
[153:21] And more than half of our team works in
[153:23] vehicle manufacturing.
[153:26] And almost 60%
[153:29] of our team is based in the US.
[153:35] People want to join our mission.
[153:38] And we need to grow and we need to hire
[153:42] the best people. Engineers want to work
[153:45] for Tesla. Look at number one and number
[153:48] two, SpaceX and Tesla.
[153:51] In 2021,
[153:53] we got 3 million
[153:56] applicants. 3 million. That's
[153:58] incredible.
[154:02] And our culture, our culture at Tesla is
[154:05] anyone at any time
[154:08] can raise an idea or make an improvement
[154:11] suggestion across
[154:14] safety, people, cost, production,
[154:18] quality.
[154:21] Cuz we have a very simple process called
[154:23] take charge. And take charge is really
[154:26] employee engagement. And as take charges
[154:28] go up, as you can see in this graph,
[154:31] injuries go down.
[154:33] And you know why?
[154:35] People who do the work, the people who
[154:37] do the work know how to improve the
[154:39] work.
[154:42] Sustainability, sustainability is
[154:44] everything. It's our mission. You've
[154:45] heard about it all day, part of our
[154:47] master plan.
[154:49] But at Tesla, we don't just tick boxes
[154:53] to meet a requirement. We really focus
[154:56] on the impact that our products and our
[154:59] software
[155:00] will deliver.
[155:02] In 2021,
[155:04] our customers avoided 8.4
[155:08] million metric tons of greenhouse gas.
[155:12] That's equivalent to taking 1.7
[155:16] million ice vehicles off the road for a
[155:20] full year. Yes.
[155:22] Yes.
[155:24] [applause]
[155:27] We're also focusing on doing really good
[155:30] work and driving sustainability into our
[155:34] operations. We have year-over-year
[155:37] reduction in our water use, in our
[155:40] waste, and we're growing and building a
[155:43] renewable energy program within our
[155:46] operations, just like supercharging, as
[155:49] you heard.
[155:50] So,
[155:52] it's not just about our products, it's
[155:54] the way we make our products, both
[155:55] phases.
[155:57] Tesla, our company, is solving both
[156:00] sides of the clean energy equation.
[156:03] We've got the clean generation, the
[156:05] clean storage. We also have zero
[156:08] emission transport. As this slide shows,
[156:11] from 2012 to 2021,
[156:15] our products generated more energy
[156:18] that was consumed by all Tesla vehicles
[156:21] and our factories.
[156:23] It's amazing.
[156:25] Yeah, I like the claps. Yeah.
[156:28] I'll take I'll take all I'll take claps.
[156:32] So,
[156:33] our vehicles use less emissions than
[156:35] gas-powered vehicles.
[156:38] Even in the worst-case scenario, 100%
[156:42] coal grid,
[156:44] the Model 3 has less emissions than the
[156:47] average ice car.
[156:49] And we're only going to get better. You
[156:52] heard about that today. Every product we
[156:55] make is more efficient. Every factory we
[156:58] build is more sustainable. So, it's only
[157:01] going to get better and better.
[157:03] So, as we build our zero emission
[157:07] future,
[157:09] we've been talking with our
[157:11] stakeholders, with our shareholders,
[157:14] about how do we improve our reporting
[157:18] framework. And we are aligning with the
[157:21] TCDF.
[157:23] So,
[157:24] TCFD.
[157:26] [laughter]
[157:26] So, we're aligning with the TCFD, which
[157:29] I know many of you have been asking us
[157:31] about.
[157:32] I also want you to stay tuned because
[157:34] there's going to be much more
[157:36] coming in our impact report. So, thank
[157:39] you all, and up next is Zach Kirkhorn.
[157:44] [applause]
[157:54] Thank you, Laurie.
[157:57] All right. So, this is our last section
[158:00] of our prepared remarks today.
[158:03] Um we'll kind of take what we've
[158:05] discussed today and summarize it at the
[158:07] corporate level for our financials.
[158:10] Maybe before I jump into the details
[158:12] here, we've never done an investor day
[158:14] as a company. We've never brought our
[158:16] leadership team out and asked them to
[158:18] talk about the things that they're
[158:19] working on.
[158:20] So, you know, I feel very fortunate to
[158:22] be able to work alongside this group of
[158:24] people and support their teams, and I
[158:26] also just want to thank them. They're in
[158:27] a room in the back listening to this,
[158:30] for all of the work that they put into
[158:31] today and getting the company to where
[158:34] it is today. So, thank you to the Tesla
[158:35] team.
[158:37] [applause]
[158:44] When we were preparing for today, folks
[158:46] were asking, "Well, Zach, what should we
[158:48] talk about?" And really, the only
[158:50] guidance we gave folks is, "Well, talk
[158:53] about the things that you're working on,
[158:54] and talk about it in the context of the
[158:56] master plan."
[158:58] And what did almost everybody talk about
[159:00] today in great detail? All of the work
[159:03] that they're doing to take cost out.
[159:05] Because in this industry, in this
[159:07] business,
[159:09] you survive or you die based upon the
[159:11] ability to manage your costs.
[159:13] And so, I'd like to talk about costs as
[159:15] well.
[159:18] If we look at our longest-running scale
[159:21] production product, this is the Model 3,
[159:24] we reached 5,000 cars per week, which
[159:26] was our design capacity in mid-2018.
[159:30] Uh and since then, we've taken 30% of
[159:32] cost out of this product.
[159:35] And there's two points I want to make
[159:36] about this. The first is that cost
[159:38] reduction, as you have heard throughout
[159:40] the course of the day, is deeply
[159:42] ingrained in our culture, and I think
[159:44] one of the most important reasons why we
[159:46] are here today as a company.
[159:48] The second point that I want to make is
[159:50] that
[159:53] when we're working on cost reduction,
[159:55] it's easy just to take costs out and
[159:56] make our products worse.
[159:58] But we have to take costs out and
[160:00] improve our products at the same time.
[160:02] This is the hard thing to do, but it's
[160:04] the necessary thing to do to continue to
[160:06] move forward.
[160:07] And if you look at a version of a Model
[160:09] 3 that we built today, and you compare
[160:11] it back to a Model 3 that we built in
[160:13] 2018, there's a long list of
[160:15] improvements that we put into the car,
[160:17] while also taking cost out of the
[160:19] product.
[160:22] Cost reduction doesn't just come from
[160:24] one place. There's no silver bullet
[160:26] here.
[160:28] And we pushed the boundaries on volume
[160:30] with the Model 3 program, increasing
[160:31] volume 3x over this period of time.
[160:34] We've improved productivity. So, in our
[160:35] Fremont factory, we are twice as
[160:37] productive now as we were in mid-2018.
[160:40] We've made a lot of progress on overhead
[160:42] efficiencies and product improvements,
[160:44] as we've discussed.
[160:45] And a long list of other things,
[160:47] including localization, our factory in
[160:49] Shanghai.
[160:51] And as we ramp up volume and as we find
[160:53] efficiencies, we work with our suppliers
[160:55] to do the same. And that leads to
[160:57] material cost reductions that improve
[160:58] the affordability of our products.
[161:02] And as we look forward to our next gen
[161:04] vehicle,
[161:05] our target is to reduce 50% of cost, and
[161:08] we've talked about that a bit earlier
[161:09] today.
[161:11] From going from the the Model S and X
[161:13] platform to the 3 and Y platform, we
[161:15] took out 50% of cost. So, the task here
[161:17] is to do it again.
[161:19] And this is very important because as we
[161:21] improve affordability,
[161:24] the number of customers who have access
[161:26] to our products dramatically increases.
[161:28] And as we link this back to our master
[161:30] plan,
[161:31] it enables an exponential growth in our
[161:33] volume with linear reductions in the
[161:35] cost of our products.
[161:37] The second point I'll make here is that
[161:39] and again, cost reductions don't come
[161:42] from any single one place. And so, you
[161:44] can see the buckets here on the vehicle
[161:46] side, battery and powertrain,
[161:48] manufacturing cost reductions, and
[161:49] others. These buckets are relatively
[161:51] equal in size. And so, in order to take
[161:53] 50% of cost out of the product, we have
[161:56] to go through everything.
[162:01] But more importantly than just the cost
[162:04] of the car upfront,
[162:08] when we're transitioning to a
[162:09] sustainable economy, particularly with
[162:11] vehicle ownership, it's really important
[162:13] to think about the lifetime cost of a of
[162:16] the products.
[162:17] And
[162:20] uh this chart here is showing what the
[162:22] total cost of ownership per mile is over
[162:24] the course of 5 years.
[162:26] And we have to think about financing
[162:28] costs, insurance costs, the cost of
[162:30] power. Drew talked a little bit about
[162:32] that with the plants that we're doing in
[162:34] Texas, wear and tear and maintenance on
[162:36] the cars, etc.
[162:38] And we're already at a place today in
[162:39] the US where a base Model 3 on a cost
[162:42] per mile basis is less than a Toyota
[162:44] Corolla, which is the highest selling
[162:46] car in the world. And as we move towards
[162:48] our next gen platform, we will continue
[162:50] to reduce this. And as we work on
[162:53] robotaxi variants of this platform, this
[162:55] cost will come down even further.
[162:58] And so, this is a product that has that
[162:59] we expect to have substantially lower
[163:02] cost per mile than the highest volume
[163:03] products in the world.
[163:07] The next thing on cost reduction I want
[163:09] to talk about is operating expenses. We
[163:11] spend a lot of time on our earnings
[163:12] calls talking about gross margins and
[163:14] product cost. We don't spend a lot of
[163:16] time talking about OpEx, but I think
[163:18] this is one of the really important
[163:19] parts of our story.
[163:21] So, since 2018, we've reduced our
[163:24] non-GAAP OpEx 60% as a percent of
[163:27] revenue and 75% on a per delivered car
[163:29] basis. Uh this is quite staggering, and
[163:33] I want to go into the details about how
[163:34] we've done this.
[163:36] The first component of OpEx, which is
[163:38] R&D,
[163:39] the the single most important thing here
[163:42] is that we constrain the number of
[163:44] programs that we work on. So, minimizing
[163:47] the number of programs that we're
[163:48] running in parallel, and the key here is
[163:50] to maximize the revenue and cash
[163:52] generated from every program that we
[163:53] work on. And also, maximize the
[163:56] technology sharing between the programs.
[163:59] So, we talked a little bit about how
[164:00] power electronics in our energy business
[164:02] is shared with power electronics in our
[164:04] vehicle business.
[164:08] If we look at the SG&A side of this, um
[164:10] I think the story here is pretty
[164:11] dramatic. So, we're showing here a
[164:13] comparison of SG&A per car delivered
[164:16] compared to the traditional auto
[164:17] industry, which is both an OEM and the
[164:19] dealership network. And our estimates
[164:22] are that we're 60 to 70% lower on a per
[164:24] car basis.
[164:26] And if you turn this into dollars, this
[164:27] is many thousands of dollars per car.
[164:30] This is part of the reason why on the
[164:31] last earnings call, I made the comment
[164:34] that we think about operating margin
[164:37] more so than gross margin as a company
[164:39] because the
[164:41] uh the integration that we have here on
[164:43] the SG&A side, particularly with
[164:44] vertical integration into our dealership
[164:46] network,
[164:47] this provides efficiencies that um um,
[164:50] give us a competitive advantage from a
[164:51] cost structure standpoint relative to
[164:53] other companies.
[164:55] The other thing that I'll note here is
[164:57] that this wasn't always the case.
[164:59] And it wasn't that long ago where these
[165:01] bars would be inverted.
[165:03] And if we go back a little bit in time,
[165:05] and we talk about when we launched Model
[165:07] 3 in the Fremont factory,
[165:09] we went from 2,000 cars per week, which
[165:11] was S and X, adding 5,000 a week to that
[165:14] for Model 3 got us to 7,000 cars a week.
[165:17] You've heard some folks here today talk
[165:19] about production hell, but I don't think
[165:21] we've ever talked about as a company is
[165:23] back office operations hell, which also
[165:25] happened at the same time.
[165:27] And at 7,000 cars a week, 350,000 cars a
[165:30] year, we want to go to 20 million cars a
[165:33] year, and we're struggling at 7,000.
[165:35] And we're saying to ourselves,
[165:37] we have to completely rethink the way
[165:40] that we're managing our back office
[165:41] operations if we have any hope of
[165:43] efficiently scaling this company.
[165:46] And so this began a process that we're
[165:48] still continuing to work on today, but
[165:50] it's something that we refer to as the
[165:51] Tesla operating system.
[165:53] So in the same way that we're vertically
[165:54] integrated into the software that
[165:56] manages our cars, we've also vertically
[165:58] integrated into the software that runs
[166:00] our company.
[166:01] And all of our major, uh, departments in
[166:03] the company are using,
[166:06] uh, not all of them yet, but almost all
[166:08] of them are using our in-house software.
[166:10] Most recently we, uh, removed
[166:12] third-party recruiting software, and
[166:13] we're now running our own recruiting
[166:15] software as a company.
[166:16] And this matters a lot because rather
[166:19] than having complex web of third-party
[166:21] systems that are generic solution that
[166:23] take a lot of effort to customize for
[166:25] our particular needs, we've been able to
[166:27] put in place with an in-house
[166:29] applications engineering team dedicated
[166:31] lightweight software that does only what
[166:33] we need it to do and nothing more.
[166:36] And that team works very closely with
[166:38] process improvement teams that sit
[166:39] within the business,
[166:41] who are looking through all of the
[166:42] processes that we're managing as a
[166:44] company, and following the exact same
[166:46] process that Tom mentioned in the
[166:47] manufacturing space to look at our
[166:50] processes, delete process steps that we
[166:51] don't need, simplify process steps that
[166:54] are existing, and automate what's
[166:55] remaining.
[166:57] And the results of this have been quite
[167:00] staggering, to be frank. I think it's
[167:02] exceeded our expectations as well.
[167:05] Uh on the left side of this chart are
[167:06] some examples of the efficiency
[167:08] improvements that we've seen within our
[167:10] SG&A areas as a result of the strategy.
[167:13] The North American sales team is four
[167:15] times more efficient now than they were
[167:17] in 2018.
[167:18] Uh order operations team, financial
[167:20] services. I mean, these levels of
[167:22] improvement are large, and I don't
[167:24] There's not enough pixels on the slide
[167:25] here to list out everything, so it's
[167:27] just a couple of examples.
[167:29] And then in the same way that when we
[167:31] look to take cost out of our products,
[167:33] we want to feature up the product or
[167:35] improve the product while taking cost
[167:37] out, the same has been true with the use
[167:40] of Tesla operating system and that
[167:41] integration with our process improvement
[167:43] teams within the company. And so we've
[167:45] in- increased performance and added
[167:47] capabilities while also dramatically
[167:50] reducing cost on a per car basis.
[167:53] There's a list of self-service
[167:54] functionality that we've put in place,
[167:56] which customers like. It also simplifies
[167:58] our operations behind the scenes. We've
[168:00] used the Tesla operating system to
[168:02] expand into our captive insurance space,
[168:04] which we are fully vertically integrated
[168:05] into. We run our own in-house software.
[168:08] We have our own agents, our own claims
[168:10] software, and and in claims folks who
[168:12] run our insurance business.
[168:14] This is also very important because as a
[168:17] technology company, data is very
[168:18] important. And uh we use these systems
[168:22] and the access to the data in those
[168:23] systems to pull out extremely granular
[168:26] and targeted reporting that enables us
[168:28] to see every aspect of our business in
[168:30] real time, so we can make adjustments to
[168:32] our operations as needed. And there are
[168:34] sometimes examples where we say, "Man, I
[168:36] wish we were tracking this piece of
[168:38] data." Well, you know, it's a Teams
[168:40] meeting away from the applications
[168:42] engineering team to make a request, and
[168:44] then we can start tracking that data and
[168:45] make those changes. And so the feedback
[168:47] loop associated with process improvement
[168:50] inside the company is is pretty
[168:52] astounding.
[168:54] The last thing I'll just mention here,
[168:55] which is something I'm personally quite
[168:56] proud of, we've made a lot of progress
[168:59] in our close processes within the
[169:00] finance team, getting our 10-Q and 10-K
[169:02] filed is amongst the fastest of the 20
[169:05] largest market cap companies. So, great
[169:07] work to the finance team here.
[169:12] So, all of this work on cost reduction
[169:14] is extremely important because we have a
[169:16] lot of money to spend ahead of us to
[169:19] achieve our goals
[169:21] within the master plan. So, we've
[169:22] mentioned over the course of the day
[169:24] today 20 million annual vehicle
[169:25] production as our target, 1
[169:27] terawatt-hour of annual energy storage
[169:30] production, and then expanding cell
[169:32] production, service, and charging in
[169:33] line with the growth of those other
[169:35] businesses.
[169:36] And so we've estimated what we think the
[169:38] total cost to get there will be, and
[169:40] there's certainly error bars around
[169:42] these numbers as as we continue to
[169:44] progress and innovate, but of this we've
[169:46] spent about $28 billion of that so far
[169:48] in the history of the company.
[169:50] And
[169:52] maybe this total investment looks large.
[169:54] I actually think it's quite small
[169:56] relative to our ambitions. And if you
[169:58] look at our 2022 operating cash flows
[170:00] and you you just say, well, let's assume
[170:03] some modest growth to that, maybe not
[170:06] all that much,
[170:08] if you're being conservative, but the
[170:10] ability to pay for this level of
[170:12] investment
[170:14] from the forecast that we have is is
[170:16] very achievable for us.
[170:21] I'm often asked about capital
[170:23] allocation. I think our strategy here is
[170:24] very straightforward. I touched upon it
[170:26] a little bit in our last slide, but just
[170:28] to be explicit about it,
[170:31] uh
[170:32] obviously the priority here is to ensure
[170:34] that we're using capital to run the
[170:35] day-to-day operations of the business.
[170:37] This is a working capital intensive
[170:40] business, and this has to be managed
[170:41] very carefully. And so, you know, over
[170:44] the course of a quarter, there can be
[170:46] multi-billion dollar swings in our cash
[170:48] balance up and down just depending upon
[170:49] the timing of when we build and deliver
[170:51] cars. So, we obviously have to have
[170:52] money set aside for that.
[170:54] Uh downside protection is another really
[170:56] important area where we have money set
[170:58] aside for that.
[170:59] Uh and over the last couple of years
[171:01] throughout the pandemic, I think this
[171:02] just highlights why being ready for
[171:04] downside side protection is important.
[171:07] Uh preserving daily operations is the
[171:10] engine that then generates cash that
[171:12] allows us to reinvest
[171:14] into the growth of the business.
[171:16] And I'm very proud that over the last
[171:18] couple of years, despite the ups and
[171:20] downs in the macro environment and
[171:22] interruptions to supply chains, we have
[171:24] never once pulled back on our
[171:25] investments in growth. So, that remains
[171:27] a huge priority for us.
[171:30] We do then, after investing in growth,
[171:33] we look at opportunistic ways to spend
[171:35] our cash. These are more financially
[171:36] oriented where if a return exceeds a
[171:38] hurdle,
[171:39] then we can place cash in that. This is
[171:41] part of our story of cost reduction in
[171:43] the Fremont factory.
[171:45] Um and then on the excess side, you
[171:47] know, this is where, you know, the board
[171:49] meets on a regular basis and they're
[171:51] thinking about what our cash flow
[171:53] projections are looking forward and
[171:54] whether it makes sense to do buybacks or
[171:56] dividends.
[172:01] So, to wrap up the finance portion of
[172:03] the presentation today,
[172:05] uh couple of key messages I just want to
[172:07] leave you all with here
[172:09] is that we use innovation at an intense
[172:13] level
[172:14] to drive cost down and improve the
[172:16] efficiency of the business.
[172:19] And the reason this is so important is
[172:20] it allows us to improve product
[172:22] affordability. And this is particularly
[172:24] the case when we move to our next
[172:26] generation platform.
[172:28] Improving affordability allows us to
[172:31] comfortably make investments that grow
[172:33] volume, that volume generates cash that
[172:36] then allows us to make more investments,
[172:39] and as we integrate that going forward,
[172:42] you know, it's our belief as a
[172:43] leadership team here at Tesla that we're
[172:46] going to achieve unprecedented scale in
[172:48] the manufacturing space.
[172:51] And this is what's ultimately required
[172:53] to accelerate the world's transition to
[172:55] sustainable energy.
[173:00] So, with that, this completes the
[173:02] prepared portion of today.
[173:05] We're going to take a short break, and
[173:07] then we'll resume with the live Q&A.
[173:09] Thank you all very much.
[173:11] [applause]
[176:24] Mhm.
[176:25] [music]
[176:41] [music]
[176:47] [music]
[176:55] [music]
[177:01] [music]
[177:08] [music]
[177:16] [music]
[177:27] [music]
[177:33] [music]
[177:57] [music]
[178:40] [music]
[178:46] [music]
[179:25] [music]
[179:32] [music]
[179:47] [music]
[179:55] [music]
[180:14] Mhm.
[180:15] [music]
[180:52] [music]
[181:09] [music]
[181:22] [music]
[181:41] [music]
[182:02] [music]
[182:09] Woo!
[183:07] Oh, I can talk.
[183:23] All right, so
[183:25] um
[183:27] Well, let's see. So, this is uh
[183:30] uh probably probably the most
[183:31] significant announcement of the day is
[183:32] that uh we're excited to announce that
[183:33] we're going to be building a a
[183:35] gigafactory in Mexico.
[183:41] [applause]
[183:45] So, um
[183:46] yeah, and and we'll have a uh
[183:49] you know, obviously a grand opening and
[183:51] you know, uh groundbreaking and whatnot.
[183:53] But, um we're excited to announce that
[183:55] that uh the next uh uh Tesla gigafactory
[183:58] will be in Mexico uh near Monterrey.
[184:01] So, super excited about it.
[184:04] [applause and cheering]
[184:07] Um
[184:08] Now, I do want to emphasize we will
[184:09] continue to expand production at all of
[184:12] our existing factories, so including
[184:14] California and Nevada. Um
[184:17] at here in Texas, obviously, and uh
[184:20] Berlin and Shanghai. So, we're we're
[184:21] intend to increase production uh at all
[184:23] factories. Uh so, the uh Giga Mexico
[184:26] would be uh supplemental to uh the
[184:29] output of all the other factories. So,
[184:31] this is not, to be clear, moving moving
[184:33] output from anywhere to anywhere. It is
[184:35] simply about expanding uh total global
[184:38] output. Uh
[184:39] Yeah. So, it's going to be good.
[184:41] Um
[184:44] So,
[184:46] [laughter]
[184:47] Peanut Gallery.
[184:49] Um anyway, so this so I think that's uh
[184:52] Yeah, we're you know, we're quite
[184:54] excited to announce that. Um
[184:55] in fact, I do believe the governor is
[184:57] here.
[185:00] It's hard to see. Head, welcome.
[185:03] Uh
[185:04] [applause]
[185:07] I know.
[185:09] Sec- Secretary of State, I believe. Uh
[185:11] anyway, so anyway, uh
[185:13] it's uh we're we're looking forward to
[185:14] it and and having a a big event uh to
[185:16] for the great groundbreaking and and and
[185:18] opening. So, let's see. With that, uh we
[185:20] can move to uh I guess Q&A. Uh we've
[185:23] obviously got uh significant bench
[185:26] strength here.
[185:27] Um
[185:28] Should probably have like a bench.
[185:30] And uh
[185:32] Anyway, we're
[185:33] we should uh
[185:34] We're maybe have too many people on
[185:36] stage, but
[185:37] Um we'll try to answer questions within
[185:39] reason. Um
[185:40] you know, which
[185:41] this is meant to be kind of more of a
[185:43] long-term
[185:45] sort of discussion as opposed to
[185:48] uh
[185:49] you know, what will be the production
[185:51] for the rest of the quarter type of
[185:53] thing. Um
[185:54] So,
[185:56] the let's try to orient our questions
[185:58] towards uh long-term value creation
[186:01] and with that fire away.
[186:05] Rob.
[186:11] Hi, it's Rod Lash for the Wolfe
[186:12] Research. Um
[186:14] very exciting plans about the the next
[186:17] generation vehicle and
[186:19] power trains and and batteries. I was
[186:20] hoping you can maybe talk to us a little
[186:22] bit about the timeline
[186:24] for deploying this and it sounds like
[186:27] it's more than just a vehicle. This is a
[186:28] kind of a paradigm change on how
[186:31] vehicles are assembled,
[186:32] how batteries are put together and and
[186:35] everything and does that also just get
[186:38] reconfigured into everything that you
[186:40] do? So,
[186:42] the Model Ys that are being built here
[186:43] will be built very differently in the
[186:45] future. Maybe just give us some feel for
[186:47] what happens from here and what's the
[186:49] timeline for implementation.
[186:52] Well, I'll talk a little bit about that,
[186:53] but
[186:55] um
[186:57] it broadly speaking the
[186:59] most profound architectural changes will
[187:01] be in future vehicles.
[187:03] Um retooling a factory means bringing
[187:06] the factory down for an extended period
[187:07] of time
[187:08] and uh
[187:10] that's that would prefer not to do that,
[187:13] I think.
[187:14] Um
[187:16] But do you But there are variants in in
[187:18] how Model Y is produced. So, we've got
[187:21] uh variants where there's a rear
[187:22] casting,
[187:23] uh where there's a front and rear
[187:25] casting, and we have the uh
[187:28] structural battery pack.
[187:30] Um
[187:31] and then there are a number of smaller
[187:33] improvements that occur, but um I think
[187:35] for for really really big changes, those
[187:37] would be future vehicles.
[187:40] Um yeah, I don't know if you guys want
[187:41] to add.
[187:44] Lars, maybe?
[187:46] Me?
[187:46] Yeah, so as I mean as far as I agree
[187:48] with you 100% Elon, and it's it's really
[187:51] easy to put innovations in new vehicles,
[187:53] but long term we'll obviously bring them
[187:55] back. We've always talked about that,
[187:56] but we don't want to take our factories
[187:58] down.
[187:58] As far as a timeline goes, um you know,
[188:00] we're going to go as fast as we can left
[188:02] to right. Uh as always, you know, Elon
[188:05] alluded to the fact that um Mexico will
[188:07] build our next gen vehicle, but we will
[188:09] also be doing that in our other plants.
[188:11] Um and so it's really about getting them
[188:13] all up and running. Um we expect that to
[188:15] be a huge volume product and um
[188:18] yeah, we're going to move that quickly
[188:19] over the next couple of years.
[188:28] Let's go to Adam.
[188:30] Uh thanks. Uh first uh enhorabuena uh
[188:33] Mexico vamos. That's great.
[188:35] Congratulations. Uh Elon um
[188:38] a question on
[188:40] applying first principles thinking
[188:43] and innovation to an area that up to now
[188:46] has seemingly been outside of of your
[188:49] control and that is on mining and
[188:50] extraction of some of the key materials.
[188:53] I believe a couple years ago you you had
[188:55] a patent on sodium chloride uh to
[188:57] extract lithium from some from some
[188:59] clays and spodumene clay and and things
[189:02] of that nature. Um any how how does that
[189:05] fit into the plan of maybe bringing real
[189:07] innovation into a mining sector that
[189:09] could use a little, you know, maybe
[189:10] waking up and getting those costs down
[189:12] cuz that could be a real gating factor,
[189:13] it seems.
[189:16] Well, we're we're we're going to address
[189:17] whatever we think the limiting factor is
[189:20] at any point in time.
[189:21] Uh so we we would like to do the least
[189:24] amount possible. Uh
[189:26] so we don't want to get into the mining
[189:29] or refining
[189:30] sector. We will do that if we have to.
[189:33] Um
[189:35] I I do think the the the focus really
[189:37] should be on refining capacity.
[189:40] Um
[189:42] you know, do we need to make just a a
[189:45] very uh
[189:46] giant
[189:48] amount of anode cathode lithium
[189:52] lithium hydroxide lithium lithium
[189:53] carbonate there's it's really the
[189:55] refining capacity that is
[189:58] the biggest choke point.
[190:02] Yeah, so that's that's why we're
[190:03] building a lithium refinery in Corpus
[190:05] Christi.
[190:08] Uh [clears throat] in terms of
[190:11] the mining companies that are out there
[190:14] and looking at that part of the value
[190:16] chain. So we do have large suppliers of
[190:18] lithium right now and we they are aware
[190:21] of you know how we're approaching the
[190:23] Corpus
[190:25] refinery and the technologies we're
[190:27] trying there and the reason we're
[190:29] making them aware of it is because we
[190:31] think they're fundamentally more
[190:32] scalable and as we prove them out we
[190:34] plan to
[190:35] to share that with them cuz as Elon says
[190:37] like we it's not really like we want to
[190:38] do these things. We're doing that
[190:39] because it's not happening fast enough.
[190:41] So if if we can prove that it can be
[190:43] done faster the intention is to transfer
[190:45] that knowledge to our our large our
[190:47] current suppliers and the same is
[190:49] actually true it was actually a clay
[190:51] process that that we were we're playing
[190:53] with and we continue to work on. Same is
[190:55] true on that. So we've worked with our
[190:57] suppliers as well on trials and and
[190:59] we're we're sharing knowledge there and
[191:01] the intention is just to help the whole
[191:03] world do this better
[191:04] and and ultimately the this here is
[191:07] getting the lithium out or whatever it
[191:08] is out of the ore.
[191:12] And we're we're obviously building a
[191:14] cathode processing
[191:16] facility just adjacent to this this
[191:19] building. So a little further down the
[191:21] road you'll see another large
[191:22] construction that's for cathode
[191:25] refining.
[191:27] Um
[191:29] But like I said we'd really prefer if
[191:30] others did that.
[191:32] We're doing it cuz we have to not cuz we
[191:34] want to.
[191:35] Yeah and and in that case there just
[191:36] isn't really any large scale cathode
[191:38] production in the United States and it
[191:40] needed to be done.
[191:41] And and again if we're going to do it,
[191:43] we're going to try to do it from a first
[191:44] principles perspective. So, we're we
[191:46] have tried a bunch of new things there.
[191:48] We're confident that they will work. And
[191:49] as they prove out, again, we want to
[191:51] bring them back to our suppliers so that
[191:53] they can uh, build new facilities more
[191:55] quickly with less than investment.
[192:00] All right, let's go to Ben.
[192:02] Hi. Uh, Ben Kallo from Baird. Uh, so
[192:04] similar, um,
[192:05] on the renewable side, is there more
[192:07] that you can do
[192:09] uh, in in Tesla uh, to to nudge the rest
[192:12] of the renewable industry to to speed up
[192:14] since it's such big uh, pillar of uh,
[192:17] the Master Plan 3?
[192:21] Well, uh, I mean, I don't know what what
[192:22] more we can do, but
[192:24] I
[192:25] I can say that if there's like
[192:27] entrepreneurs out there that want to
[192:29] have like a guaranteed chance of
[192:32] success, it would be refining lithium or
[192:36] anodes and cathodes
[192:38] or any any materials whatsoever for
[192:41] uh,
[192:41] lithium-ion cells.
[192:43] It is a no-brainer.
[192:44] Um,
[192:47] Yep.
[192:48] I think uh,
[192:50] I we're doing everything we can.
[192:52] So,
[192:53] Yeah, on the on the renewable energy
[192:54] thing,
[192:55] uh,
[192:56] the thing that that we can do Tesla, um,
[192:58] is the more we reduce the cost of
[193:00] storage, the more we reduce the cost of
[193:02] stationary storage, and the more we
[193:04] bring like flexible load to the grid in
[193:06] the form of like cars charging at the
[193:08] right times, the more valuable
[193:09] renewables are.
[193:11] Um, because in like Texas right now, I
[193:13] wasn't joking. Like, there's a lot of
[193:15] wind.
[193:16] Um, and at night, the wind is there's
[193:18] like almost too much wind cuz they don't
[193:19] want to turn the nukes off or whatever.
[193:21] So,
[193:22] um,
[193:23] so so so so bringing really low-cost
[193:26] storage onto the grid makes renewables
[193:29] more valuable, which ultimately will
[193:30] accelerate their deployment. So, that
[193:32] that's how we focus on it, I guess.
[193:36] Okay. Philippe?
[193:46] Thank you. It's Philippe Buchard
[193:48] Jeffries.
[193:50] I've got two questions. The first one
[193:52] when I think about
[193:53] in this industry everybody wants to be
[193:55] Tesla. Every car maker is trying to
[193:57] emulate what you're doing.
[193:59] Well done. The one thing they don't do
[194:01] is
[194:02] um
[194:04] focus on having as much growth as
[194:06] possible with as few models as possible.
[194:09] So, I'm just trying to understand as you
[194:10] aim for 20 million units in 2030,
[194:14] how many models do you think you need to
[194:17] get there? And how does it fit into your
[194:19] drive to hyper scale?
[194:21] Um how do you manage this? And also the
[194:24] fact that probably consumers at some
[194:26] point don't want to see a Tesla or the
[194:28] same Tesla at every street corner. So,
[194:30] I'm just trying to get your your your
[194:31] sense of that.
[194:33] And my other question is on
[194:36] on bidirectional. I mean, you
[194:38] you talked a lot about making you know,
[194:40] a world more renewables and better use
[194:42] it of cars. I think bidirectional
[194:44] charging is one way of better using
[194:46] cars, but you seem to have been
[194:48] reluctant about
[194:49] doing that in the past. So, I'm just
[194:51] wondering what your latest views are on
[194:53] the topic.
[194:57] [clears throat]
[194:57] Sure. On bidirectional um
[195:01] it it uh
[195:04] it wasn't like a conscious decision to
[195:06] not do it. It just wasn't a priority at
[195:08] the time. I think is maybe the way to
[195:09] think about it. Um as we looked as we
[195:12] continue to improve the power
[195:13] electronics in our vehicle,
[195:15] um
[195:15] we've found ways to bring
[195:17] bidirectionality while actually reducing
[195:19] cost of power electronics in the
[195:20] vehicle. And as at all things Tesla, the
[195:23] the goal is usually to
[195:25] get more for less. Um and so, uh we are
[195:28] in the middle of kind of like a power
[195:29] electronics retool, I would say, that
[195:32] will bring that functionality to our all
[195:34] of our vehicles over the next, you know,
[195:36] 2 years, let's say. Um,
[195:38] but
[195:39] but it's
[195:41] Yeah, I guess that's how I'll say there.
[195:43] I don't think very many people are going
[195:44] to use bidirectional charging unless you
[195:46] have a power wall because if you unplug
[195:49] your car, your house goes dark.
[195:51] And this is extremely inconvenient.
[195:57] Yeah, most of the value it comes in
[195:59] charging the car at the right time. It's
[196:01] not really about sending energy the
[196:02] other way.
[196:04] Um
[196:04] Yeah, I mean if if you have a power wall
[196:06] um
[196:07] that can take the house load, then you
[196:08] can use the the your car as supple as a
[196:11] supplementary energy source
[196:13] to the power wall and then,
[196:15] you know, you're not going to drive
[196:16] everyone crazy by unplugging your car
[196:19] um and having the house go off.
[196:21] So,
[196:22] I think there's some value there as a
[196:24] supplemental energy source down the road
[196:27] uh
[196:28] where if you have a power wall, you
[196:30] not
[196:30] uh diminish the
[196:33] convenience of of of the people in the
[196:35] house.
[196:38] And the question of number of models?
[196:42] Sorry, the what? Oh, how many models? Oh
[196:44] vehicle models
[196:45] I have not that many. Um really
[196:48] 10? I don't know. Not that many.
[196:51] Um
[196:52] The there's
[196:53] I mean what's happened with conventional
[196:55] cars is people have run out of things to
[196:56] do. So,
[196:58] they run out of things to do, they just
[196:59] end up reshuffling the deck and you have
[197:01] pretty much the same
[197:03] I mean, how many variants of car are
[197:05] there on the road? There's like
[197:07] hundreds.
[197:10] But they're are they are they good
[197:12] variants? No, I mostly not.
[197:15] They're just variants for the sake of
[197:16] variants. Um
[197:18] And but look at how how is have things
[197:20] converged with the phone?
[197:23] I mean, there used to be hundreds of
[197:24] flip phones. Now, what what do we have?
[197:27] It'll be like that.
[197:32] George.
[197:33] Hi, George Generikas from Canaccord
[197:36] Genuity. I just had a question about
[197:38] your plans to grow
[197:40] market share in China. And also whether
[197:42] or not the political tensions between
[197:45] United States and China impact your
[197:47] long-term ambitions there. Thank you.
[197:53] I don't know.
[197:55] Hey Tom, not to brag don't don't don't
[197:57] sweat too hard.
[197:59] [laughter]
[198:01] Yeah.
[198:02] Well,
[198:03] um
[198:05] Well, we're still growing out our market
[198:06] share in China um quite a strongly
[198:08] actually. Um
[198:10] especially um you know early this year
[198:12] um we had a
[198:13] price adjustment. Um after that um we
[198:16] actually generated a a huge demand um
[198:18] more than we can produce really. Um
[198:21] and um um as Elon said as long as you're
[198:24] um
[198:25] um offer a
[198:28] a product um
[198:29] with value um at affordable price, you
[198:32] know, um you don't have to worry about
[198:34] demand. Um that's basically the
[198:36] philosophy that we we follow. Um we try
[198:38] everything to cut cost um
[198:40] from supply chain um from improve the
[198:43] efficiency in the factory um and the
[198:45] pass down that value to our customers.
[198:47] Um I think you as as long as we continue
[198:49] to do that um I'm not too concerned
[198:51] about the market share in China. Um and
[198:54] the second part geo politic no one else.
[198:56] Um we do our best. Um we we actually
[198:59] create a lot of jobs um to the local
[199:01] community. Um and we um generate also um
[199:05] and
[199:06] with um
[199:07] at our suppliers um
[199:09] factories so we create a lot of jobs as
[199:10] well and um we contribute with a lot of
[199:13] to the um
[199:14] local
[199:15] economy. Um I think you know, as long as
[199:18] we're needed um in this country, I don't
[199:21] see there much of a risk of that.
[199:25] Colin?
[199:26] Uh Colin Rush from Oppenheimer. You
[199:28] know, one of the things I was struck by
[199:29] in the presentation was the the
[199:31] operational efficiency metrics that you
[199:32] guys are talking about. Can you talk a
[199:34] little bit about the velocity of
[199:36] learning cycles and how you guys track
[199:38] that and and think about that as an
[199:40] organization as you work into a variety
[199:42] of other areas from from innovation
[199:43] perspective?
[199:51] Anyone want to try that?
[199:52] I mean, I'll I'll say one very
[199:53] high-level thing, which is you can't
[199:55] improve something you don't measure, and
[199:57] we're like ruthless measurers at Tesla.
[200:00] And um
[200:02] once you start measuring things that
[200:04] contribute to operational efficiency,
[200:06] you actually have a path to doing
[200:07] something about it. Um so, I guess the
[200:10] key is a good measuring stick.
[200:15] Okay.
[200:17] But you know, something I should say
[200:19] with respect to uh
[200:21] demand, which I've said a few times over
[200:23] the years, but it sometimes uh
[200:26] needs to be said said again, which is
[200:29] um
[200:30] the
[200:32] overwhelmingly the uh the there's the
[200:35] the desire for people to own a Tesla is
[200:38] extremely high. The limiting factor is
[200:40] their ability to pay for a Tesla, not do
[200:43] they want a Tesla?
[200:45] Um it's it's easy for people in this
[200:48] room to lose sight of that. It is if if
[200:51] your income is far in excess of what a
[200:53] car costs, then you look at value for
[200:55] money, but you do not consider
[200:56] affordability.
[200:58] But for the vast majority of people, it
[201:00] is affordability driven.
[201:03] This is why we cannot simply double the
[201:04] price of the car,
[201:05] um or you could say and to think about
[201:08] things in the limit, where if if you had
[201:09] an
[201:10] you know, infinitely desirable car, but
[201:13] uh
[201:14] it costs $10 million, uh it wouldn't
[201:17] matter, because people not kind of most
[201:19] people do not have that. So,
[201:22] demand is
[201:24] very much a function of affordability,
[201:26] not desire.
[201:28] Very important.
[201:30] One of the things we weren't sure about
[201:32] was the price elasticity of demand for
[201:35] for Teslas. So, like as we lower the
[201:37] price,
[201:39] how much does demand increase?
[201:41] Um
[201:43] and we found that even
[201:45] small changes in the price have a big
[201:48] effect on demand, very big.
[201:50] So, that was a good thing to learn. Um
[201:55] Yeah.
[201:57] And then as
[201:59] the sort of autonomy question
[202:01] is very very big. Um
[202:04] cuz you could potentially have, I don't
[202:06] know, five times the utility
[202:08] of of the asset that you currently have.
[202:10] So, if passenger car is
[202:12] 10 or 12 hours a week of usage,
[202:15] um plus a lot of parking expenses,
[202:18] uh
[202:19] an autonomous car could be 50
[202:22] to 60 hours a week or something like
[202:24] that.
[202:25] Um
[202:26] and you could get rid of a lot of
[202:27] parking expenses. So,
[202:33] you know, if if this if this is true,
[202:34] then as
[202:36] autonomy is effectively turned on
[202:39] for the fleet, uh it may be the probably
[202:43] will be the biggest uh asset value
[202:45] increase in history overnight.
[202:50] Emmanuel
[202:55] Thank Thank you so much. Emmanuel Rosner
[202:57] from Deutsche Bank.
[202:58] Um
[202:59] So, as you uh start launching this
[203:02] next-generation vehicle and ramping up
[203:04] volume, what will be your nearest-term
[203:06] priority in terms of
[203:09] segment of vehicles focus? Uh the the
[203:11] slide that you showed with two vehicles
[203:13] on the ramp seemed, based on the form
[203:15] factor, one of them maybe looks like a
[203:17] van, another one looks like maybe like a
[203:19] smaller uh vehicle, like potentially a
[203:21] Model 2, I guess. What is the near-term
[203:23] focus for you in terms of uh ramping up
[203:25] the next-gen vehicle? And how do you
[203:27] make sure that by lowering the price
[203:30] point uh so much, because the cost is
[203:32] going down 50%, you're not cannibalizing
[203:35] demand, you know, for your existing
[203:36] vehicles?
[203:45] All right, demand for our vehicles in
[203:46] terms of desire to own them may as well
[203:48] be infinite.
[203:50] Uh
[203:51] it's
[203:51] it's indistinguishable from infinite at
[203:53] this point. Uh
[203:55] so, affordability is what matters. So,
[203:57] as you make the car more affordable, uh
[203:59] we'll have demand to go crazy,
[204:02] basically.
[204:03] Uh the the issue is how do we build the
[204:05] cars? The hard part is building the
[204:06] cars.
[204:08] I can't emphasize that enough. The hard
[204:09] part is building the cars and the entire
[204:11] supply chain that goes with the cars. Uh
[204:14] this is a
[204:16] logistics challenge
[204:18] of extraordinary difficulty.
[204:21] Um
[204:24] All the things that have to go into the
[204:25] car have to scale with the car
[204:27] while everything is doing an
[204:29] an exponential ramp.
[204:31] And if you miss even one of those
[204:33] things,
[204:35] doesn't matter why, earthquake, flood,
[204:38] fire, revolution,
[204:40] I thought I'd heard them all. Uh I mean,
[204:42] there's
[204:43] uh
[204:45] any part of that supply chain gets
[204:46] interrupted,
[204:47] you're now you you know, you have a
[204:48] seizure.
[204:53] The hard part is building the cars,
[204:56] by far.
[204:58] And the supply chain that goes with it.
[205:00] I mean,
[205:01] you guys want to talk about supply chain
[205:02] stuff?
[205:05] Uh well, they might.
[205:06] I think, you know, and the presentation
[205:08] I'd mentioned, um,
[205:10] perfection [clears throat] is a passing
[205:11] grade. We really need everything to
[205:13] happen perfectly, and the strategy for
[205:15] mitigating the different risks, some of
[205:17] which were anticipated and some weren't,
[205:19] is really bespoke to the situation. And
[205:21] and and really the
[205:24] you need subject matter expertise and
[205:26] and and a deep knowledge of the
[205:27] particular supply chain you're managing
[205:29] to come up with those strategies. So,
[205:31] we've we've seen everything from, as
[205:32] Elon mentioned, you know, first the the
[205:34] tariffs that flew back and forth, then
[205:36] the ocean logistics issue, a chip
[205:39] shortage, uh, COVID, floods, there was a
[205:42] fire in a fab in Japan that knocked down
[205:46] There was a massive, uh, COVID spike in
[205:48] Malaysia where a lot of the chips are
[205:50] the back end of a lot of chips is was
[205:52] done there. And more. Uh, this is one
[205:54] that Elon was involved with as well.
[205:56] And, um,
[205:58] yeah, that's
[205:59] it's nerve-wracking, but somehow it
[206:02] works.
[206:04] Yeah, and
[206:05] [clears throat]
[206:06] although it's difficult, I think we have
[206:07] been
[206:08] laying the foundation to be as intimate
[206:11] to our all the tiers of the supply
[206:13] chain, uh, building that, uh, control
[206:16] and, you know, directing the, uh, the
[206:18] different tiers of supply chain. That's
[206:20] the way we are uh, trying to mitigate
[206:22] the risks that come with it. Um, you
[206:25] know, again, dual sourcing, triple
[206:27] sourcing, having redundancy,
[206:30] uh, is the way we've been trying to
[206:32] mitigate for it.
[206:35] Yeah, on on on that point, when we think
[206:37] about vehicles, when you think about 3Y
[206:39] as an architecture, SX as an
[206:41] architecture, our next-generation
[206:42] platform is more than one segment. And
[206:44] really we're thinking about all the
[206:45] segments that are available that we
[206:46] haven't captured and where the market
[206:48] would be in designing it with our supply
[206:50] chain partners so that we can go quickly
[206:52] through those segments for where we
[206:54] need. But to Elon's point, um, if you
[206:56] make a car desirable
[206:58] and affordable,
[206:59] um, you know, oftentimes it it doesn't
[207:02] necessarily matter what segment it's in
[207:04] because it's one that you want. And um
[207:06] we've seen that with Model 3 when a lot
[207:08] of people thought the sedan was not
[207:09] going to be a great hit, but we sell
[207:11] tons of them. So, um the next generation
[207:14] platform is not one vehicle, it is it is
[207:16] multiple, and it's on a segment um
[207:18] that we will um you know, really turn
[207:20] focus on that affordability and
[207:21] desirability point moreover than where
[207:24] we started.
[207:26] Yeah, I mean there's like I think
[207:29] you know, there's an old saying like
[207:30] battles are won with tactics, wars are
[207:31] won with logistics.
[207:33] Uh the logistics challenges here are
[207:34] enormous.
[207:36] Um and when you you start like being a
[207:38] very significant percentage of an
[207:41] industry,
[207:42] uh you can't over capacitize. It's not
[207:44] realistic. Um
[207:46] and some of these
[207:47] things like you say like well you
[207:49] dual source or triple source. You can do
[207:51] that maybe for small things, but you
[207:52] can't do it for big things cuz
[207:55] y- you if you're if you're triple
[207:57] sourced
[207:58] and one of it's like having a a plane
[208:01] with three engines where if any of the
[208:03] three engines fails, you crash.
[208:06] It's like
[208:07] you know, so
[208:09] th- you you you have to either over
[208:10] capacitize, which drives your capex up,
[208:13] um and it has idle suppliers uh
[208:15] somewhere and big warehouses, uh or you
[208:19] design to some overage that's you know,
[208:22] reasonable.
[208:23] Um
[208:25] and then you have expedite costs because
[208:27] inevitably there's something goes wrong
[208:29] somewhere and you got to fly things
[208:30] around.
[208:31] Um
[208:32] so it's really just the the rate of
[208:35] progress is the rate at which we are
[208:36] able to scale a uh
[208:39] 10,000 logistics problems.
[208:42] Um most significant of those is the cell
[208:44] production.
[208:45] Um
[208:47] [clears throat]
[208:47] and so w- we we actually
[208:50] deliberately try to overdo
[208:52] uh cell production or s- cell
[208:55] supply um
[208:57] to have that exceed what is needed in
[208:59] vehicles
[209:00] uh because if if it goes below what's
[209:02] needed in vehicles, then the with the
[209:04] factories stall.
[209:06] Um
[209:08] So, but then what do you do with all
[209:09] these extra cells? It's like, well,
[209:11] okay, so the the much easier thing to
[209:13] scale up and down is
[209:15] is Powerwall and Megapack output
[209:17] stationary storage.
[209:19] So,
[209:20] we we can then over capacitize in cells
[209:23] and packs
[209:24] um and scale production of stationary
[209:28] storage, which is much easier to scale
[209:30] than vehicle production.
[209:32] So, that's
[209:33] strategically, I think a good thing. Um
[209:37] Yeah, I mean, the CapEx for Megapack is
[209:39] tiny compared to CapEx for vehicles.
[209:42] Um and uh
[209:43] w- also Megapack demand is
[209:45] quasi-infinite. It It basically uh as
[209:48] long as we are competitive with
[209:50] utilities, uh we can sell as many mega
[209:52] mega like
[209:54] It's It's Yeah.
[209:57] Quasi-infinite demand for that, really.
[209:59] Um or multi-ter- many terawatt hours.
[210:02] And we got a long way to go to get to
[210:04] many terawatt hours per year.
[210:08] Dan.
[210:10] Let's go to Dan.
[210:13] Thank you. Dan Levy, Barclays. Um I
[210:16] think we know competitive dynamics
[210:19] differ significantly by region. You
[210:22] know, cost dynamics differ by region. We
[210:23] saw that with your China Gigafactory,
[210:26] far superior cost to what you had
[210:27] Fremont. I realize it was a new factory,
[210:29] but uh there are clearly different
[210:30] dynamics. So, to what extent are the the
[210:33] cost strategies that you've laid out
[210:35] today, do they differ by region or uh or
[210:38] is there more of a global
[210:39] one-size-fits-all
[210:41] approach on reducing cost?
[210:48] Who was talking?
[210:50] Um
[210:51] Yeah, I think um um
[210:53] we're [clears throat] pretty consistent
[210:54] on the strategy here. We try to
[210:56] as much localized as possible.
[210:59] Like in China, 95% over 95% of our
[211:02] supply chain are localized. Um, all the
[211:04] way from the first tier to the second
[211:06] tier and then to the third tier. Um,
[211:08] and that give us
[211:11] generate a tremendous savings on costs.
[211:14] Um, and we also have a very localized
[211:18] um
[211:19] um labor force um and we have access to
[211:23] skilled labor force in the region and
[211:25] Yangtze Delta region in in Shanghai I'm
[211:27] particularly referring to.
[211:28] Um, and and
[211:30] among over 30,000 of
[211:33] employees in China, we have probably um
[211:37] less than 20 experts.
[211:39] So, it's
[211:40] you know, very deep localization we have
[211:42] done there
[211:43] to be able to
[211:45] have such cost structure and we try to
[211:47] replicate
[211:49] this approach elsewhere in
[211:51] in
[211:52] different gigafactories. Of course, you
[211:55] know, the supplier base are different,
[211:57] the
[211:58] you know, labor market are different
[212:00] region by region and country by country,
[212:02] but we try our best to to localize. Um,
[212:05] yeah, and and um
[212:08] I think that's that's absolutely
[212:11] give us an advantage to compete with the
[212:14] auto OEMs around the globe. Um,
[212:18] another thing is you know, we have this
[212:20] direct selling mode like Zach said, you
[212:22] know, we will save a tremendous money on
[212:25] on OPEX as well. We have full control
[212:27] over
[212:28] all the expenditures across the
[212:31] the company and you know, we don't spend
[212:34] money on
[212:36] marketing or advertising.
[212:38] Everything
[212:40] we saved
[212:42] will become
[212:43] you know, the value we can offer to the
[212:44] consumers. Um, so
[212:47] that part on that part we also follow
[212:49] pretty consistent strategy over the
[212:50] years.
[212:54] I mean, certainly the manufacturing
[212:55] stuff that Colin, Pete, and myself
[212:57] talked about, whether you make it here,
[212:59] or you make it in Europe, or you make it
[213:00] in Asia, it applies everywhere. We think
[213:02] about that from the ground up, not not
[213:04] specific to any region.
[213:07] Yeah, just one thing I would add
[213:09] to Tom's point about localization and
[213:12] Lars's point here.
[213:13] And I would actually say that we're
[213:15] moving in a direction of more
[213:16] standardization in terms of our
[213:17] factories and our processes and our cost
[213:19] reduction approaches. So, as we've been
[213:21] thinking about the next generation
[213:22] platform, we've been thinking about the
[213:25] volume that we aspire to build against
[213:27] that, how many individual factories do
[213:30] we need to build, and what is the
[213:32] fastest possible way to expand that
[213:35] footprint around the world.
[213:37] Uh, what we've done with Model Y is each
[213:39] factory is an incremental improvement of
[213:41] the previous factory, which requires
[213:43] engineering hours, engineering spend for
[213:45] each factory design, and then you end up
[213:46] with factories that are slightly
[213:48] different from each other.
[213:49] And as we move towards the next
[213:50] generation platform,
[213:52] uh, I think the the term you used, Lars,
[213:55] is copy-paste.
[213:56] So, to get it right from the first time,
[213:59] um, you know, certainly there will be
[214:01] things that we learn and we replic- and
[214:03] we make adjustments to, but try to have
[214:05] as much standardization and
[214:06] commonization as possible, which allows
[214:08] us to go as quickly as possible with
[214:10] expansion.
[214:13] Omer Rokah on the side.
[214:19] Thanks. Um, two questions, one for Zach,
[214:21] one for Ashok. Um, so, for Zach, um,
[214:24] how do you think about a long-term
[214:27] growth and operating margins for the
[214:28] business? And when we look at a 150 to
[214:31] 175 billion in CapEx, is um, the
[214:35] remaining balance of that, is that
[214:37] essentially the CapEx guidance through
[214:39] 2030?
[214:40] Um, for Ashok, um, the multi-trip
[214:43] reconstruction that you highlighted, how
[214:45] is that different from what Mobileye
[214:46] does, what the REM map that they have?
[214:49] And can you give us any color on the AP4
[214:52] hardware?
[214:57] So, with respect to operating margins,
[215:00] you know, we intend to continue to
[215:01] improve operating expenses as a
[215:03] percentage of revenue over time.
[215:06] And we're continuing to take cost out of
[215:08] our products, um, and try to keep gross
[215:10] margins at a place that's healthy as
[215:12] well.
[215:13] And so, you know, from the hardware
[215:15] perspective of the business, um, you
[215:17] know, it's our expectation that we'll
[215:18] continue to stay in a healthy place over
[215:20] time.
[215:21] Uh,
[215:22] and then there's a software portion
[215:24] that's added on top of that. And so,
[215:25] Elon has commented on this many times
[215:27] about the impact that full self-driving
[215:29] software can have on the economics of
[215:31] the business. And in that space, you
[215:34] know, profitability, operating margins
[215:36] would be impacted dramatically.
[215:38] Uh, specifically to your point about
[215:40] CAPEX guidance, you know, the intent of
[215:42] that slide was not to provide specific
[215:44] guidance, but to just be transparent
[215:45] about internally
[215:47] what our rough estimates are, and to
[215:50] provide context that we think getting to
[215:52] 20 million vehicles per year and 1
[215:54] terawatt hour of energy storage is very
[215:56] feasible relative to our expected cash
[215:58] generation for the business. You know,
[216:00] that's that's a number that will move as
[216:02] we learn. We may choose to vertically
[216:04] integrate more into things. We may find
[216:06] efficiencies elsewhere. To the whole
[216:08] conversation we were having earlier
[216:09] about do we do mining, do we not do
[216:11] mining. But, uh, the entire point is to
[216:14] say that that this is something that's
[216:16] entirely possible based upon our
[216:17] forecast.
[216:20] All right, regarding the multi-trip
[216:21] reconstruction, uh, my my point I was
[216:23] trying to make was that um, we want to
[216:26] auto label most of the data and we can
[216:28] collect data from the fleet and then
[216:29] reconstruct local areas that can act as
[216:31] supervision for those clips. Since we
[216:33] want to build a scalable self-driving
[216:35] system, we don't want to really rely on
[216:37] HD maps or anything. Even though we
[216:39] could like trivially build something
[216:40] using the same technology. Uh but this
[216:42] is mostly for auto labeling.
[216:44] Uh and this provides like precise 3D
[216:47] labels for all the video sequences
[216:49] involved in the reconstruction.
[216:54] Uh it's something that's uh
[216:56] I don't know if
[216:57] if we touched on this much, but in terms
[216:59] of training uh we we have one of the
[217:01] biggest uh neural net training systems
[217:03] in the world, and that
[217:05] uh
[217:06] we expect to increase that capability by
[217:08] an order of magnitude by the end of this
[217:09] year.
[217:10] Um and probably another order of
[217:13] magnitude uh by the end of next year.
[217:16] Yeah, with some combination of Nvidia
[217:19] and Dojo.
[217:21] So,
[217:22] like I'm not sure if the scale of that
[217:24] is quite um
[217:26] dawned on people. It's enormous.
[217:31] Maybe the final two questions, one from
[217:33] Alex and one from Chris.
[217:40] Okay. Alex Potter with Piper. Um
[217:43] so I definitely want to ask Drew or or
[217:46] anybody else up there uh for an update
[217:48] on on dry battery electrode, right? So,
[217:50] if you're try to over capacitize towards
[217:53] cells, um scale a lot of this uh clearly
[217:56] there's a lot of moving pieces and it's
[217:58] a complex sort of orchestra with the
[218:00] supply chain, but a lot of it comes down
[218:02] to dry battery electrode, at least to
[218:03] me. Um so how are you trending? Uh that
[218:07] was the most fascinating part of the
[218:08] factory tour for me, and looking through
[218:09] that window and seeing uh that this is
[218:11] clearly not a science project. Um
[218:13] but anything that you're willing to
[218:15] disclose, yields um
[218:17] progress, where you are today versus
[218:18] where you thought you were going to be.
[218:20] Um yeah, thanks.
[218:27] Um
[218:28] [clears throat]
[218:29] uh yeah, I mean, as you saw, right? Like
[218:31] it's this is this is a real factory
[218:33] making a lot of dry electrode in an
[218:35] automated fashion. Um
[218:37] we've made a lot of progress.
[218:39] Uh
[218:41] it's it's a it's a spectrum. Like we
[218:44] we're perfectionists and we we have
[218:46] clear end goals.
[218:48] Um
[218:49] and we are
[218:51] I
[218:51] Every week that goes by making progress
[218:53] towards those end goals, whether it's
[218:55] speed of the tool, yield of the of that
[218:57] process or the downstream process.
[218:59] Um
[219:02] We we haven't stalled out yet on the
[219:04] rate of progress either. And that's both
[219:06] on the anode and the cathode side. Um
[219:08] and I think
[219:10] the the great thing about where we are
[219:12] with the overcapacity that Elon
[219:14] mentioned is it's giving us the
[219:15] opportunity to
[219:17] um experiment
[219:19] as we go rather than just being like
[219:21] stuck to something that uh we happen to
[219:23] kick off a year and a half ago and it is
[219:26] here. Uh something that Elon has said to
[219:28] the team many times is it's okay to
[219:31] scrap equipment or money. It's not okay
[219:33] to scrap time. So, the way we've been
[219:36] approaching it is probabilistically
[219:39] what do we think [clears throat] is the
[219:39] most likely thing to succeed? And even
[219:42] actually in the factory here, you saw
[219:44] more than one anode line. Uh they they
[219:46] are actually operating two slightly
[219:48] different versions of the the final
[219:49] process step of the powder entering the
[219:51] tool. And
[219:53] it's a competition. You know, which is
[219:55] going to be more higher yield, which is
[219:56] going to perform better. And we have the
[219:57] luxury to be able to do that. Um and
[220:00] we're taking full advantage of it to
[220:01] advance the technology as quickly as
[220:02] possible.
[220:04] The the dry electrode problem is is
[220:06] really quite a hard problem. Uh I mean,
[220:09] we acquired Maxwell really just for the
[220:11] dry electrode uh technology, but uh
[220:15] just just to illustrate what a gigantic
[220:17] gap there is between something working
[220:19] at small scale and at large scale.
[220:21] Um
[220:22] and we've had a
[220:24] an extremely talented team of engineers
[220:27] working on sort of scaling the dry
[220:29] electrode process and having it be
[220:31] reliable
[220:32] uh
[220:33] consistent uh
[220:35] and
[220:36] we've we've been grinding hard literally
[220:38] and figuratively on this
[220:40] for quite a while. Um
[220:43] it seems likely that we will be able to
[220:45] scale it to volume this year.
[220:48] Yeah.
[220:48] Yeah, I mean we're we're basically
[220:50] increasing the output week over week uh
[220:52] like roughly 1K a week per quarter is
[220:54] our our internal target. Um and we're
[220:58] you know, we're tracking to that. I
[220:59] think the what you saw is
[221:03] effectively like a ton of material per
[221:05] hour per tool.
[221:07] It's kind of like hard to like
[221:09] rationalize like what that really means.
[221:12] Um
[221:13] but it's
[221:14] it it it's it is it's different than
[221:16] like oh, something works in a lab. It's
[221:18] like well, when it's tons of material
[221:19] per hour,
[221:20] there's just different kinds of
[221:22] problems. Like even if you have like
[221:23] 0.1%
[221:25] escape of like fines into the enclosure
[221:29] that your equipment is in, now you have
[221:31] a dust problem and you have and like
[221:33] that could short out some electronics.
[221:34] It's like just things like this
[221:36] where when you're on a lab scale, you
[221:38] don't even notice it, but when you're
[221:39] doing thousands of of tons over the
[221:41] course of months, it's like oh, a new
[221:43] failure mode we found. And that's where
[221:45] we're at. Um but we're we're we're
[221:46] knocking those out though. Um and and
[221:49] and the team is
[221:50] grinding through through it, but but
[221:52] progress every week.
[221:56] And the last question from Chris.
[221:59] Thanks for taking the question. I have a
[222:00] a few follow-ups on the next gen
[222:02] vehicle. Um first, when do you think
[222:05] we'll get a look at it, maybe a
[222:06] prototype?
[222:08] Second, are there any details that you
[222:10] think you can share in terms of the the
[222:12] size, the content, the performance? And
[222:14] then third, I think you mentioned that
[222:16] you would produce it in other plants in
[222:18] addition to Mexico.
[222:19] Should we take that to mean that you can
[222:21] launch it at an existing plant before
[222:23] you're finished constructing the new
[222:25] plant in Mexico?
[222:28] Uh
[222:28] I think we'll actually have to probably
[222:31] decline that
[222:32] answer.
[222:33] Uh
[222:35] We will have a proper sort of product
[222:38] event, but uh that we would be jumping
[222:40] the gun if we were to answer your
[222:41] questions. Maybe another question if
[222:43] this
[222:44] Yeah. Uh
[222:45] yeah.
[222:46] I don't know. Anyone?
[222:49] Will Will they not?
[222:54] I I have two questions. This has really
[222:56] been a very impressive afternoon. Thank
[222:57] you so much. Elon, I'm curious as you've
[223:00] doubled and you'll double again,
[223:02] how what you've learned about sort of
[223:05] managing a larger enterprise and what
[223:07] you might have to, you know, do to
[223:10] manage a bigger enterprise. And then
[223:12] secondly, I'm curious on your thoughts
[223:13] on
[223:14] how, you know, generative AI and, you
[223:17] know, these
[223:18] rapid breakthroughs in AI in the last
[223:20] months could help you make cars
[223:24] sort of
[223:26] you know, less hard to make.
[223:28] Thank you.
[223:32] I don't see AI helping us make cars
[223:34] anytime soon.
[223:35] Um
[223:37] at that point, I mean
[223:40] I would no point any of us working, I
[223:42] think. Big problems.
[223:44] I think we should all just chill out.
[223:46] Um
[223:48] I mean, I'm a little worried about the
[223:50] AI stuff. Um
[223:55] I think it's uh something, I don't know,
[223:57] which we should be concerned about. Uh
[224:04] I don't know. I think we should need
[224:04] need some kind of like regulatory
[224:06] authority or something that's overseeing
[224:08] AI
[224:10] development and just making sure that
[224:11] it's operating within the public
[224:13] interest and
[224:14] you You it's quite a dangerous
[224:17] quite a dangerous technology.
[224:19] Um and I I
[224:21] I fear I may have
[224:23] done some things to accelerate it, which
[224:24] is
[224:26] I don't know.
[224:28] So,
[224:31] I mean, some of the AI stuff I think is
[224:33] just obviously useful like what we're
[224:35] doing with
[224:37] self-driving,
[224:39] which is, you know, some people think is
[224:42] an AGI
[224:43] type problem. I don't think it's quite
[224:45] an AGI problem, but it's
[224:46] certainly requires
[224:49] uh
[224:50] very sophisticated neural nets uh
[224:52] because the the road system is designed
[224:54] for eyes and biological neural nets. So,
[224:57] naturally, the analog to that is
[225:01] uh
[225:02] cameras with uh
[225:05] digital neural nets.
[225:06] Um
[225:07] and uh
[225:10] Yeah, the thing we've we found is it's
[225:11] just there's a sig-
[225:13] if if you actually look at our neural
[225:15] net architecture in the car, it's
[225:17] kind of insane, frankly.
[225:19] Uh
[225:20] nets upon nets upon nets upon nets.
[225:22] Yeah, the the visualization tool crashes
[225:25] when you open
[225:25] Yeah.
[225:26] It's like so complicated that no one can
[225:27] visualize this right now.
[225:29] Yeah, it's hard to vis- Yeah, it's just
[225:30] like literally it looks insane.
[225:32] Um
[225:33] So, then I I guess something like that's
[225:34] happening in our brains while we drive
[225:36] around, which is pretty wild.
[225:38] Um
[225:40] So,
[225:41] you know,
[225:43] I don't know.
[225:44] Um
[225:49] I think Tesla's doing good things in in
[225:50] AI.
[225:56] I don't know. This one stresses me out,
[225:57] so I don't know what to think to say
[225:58] about it.
[226:01] You know, um
[226:04] Okay, given it's 7:00, I think that's
[226:06] all the time we have.
[226:08] Thank you very much for coming, and uh
[226:10] Thank you.
[226:10] We'll see you later.
[226:13] [applause]
