Accelerating AI compute demand risks driving global capital toward a $5 trillion credit shortfall, raising interest rates and threatening non-AI corporate debt defaults.
## Background
Dwarkesh Patel published this podcast episode on his self-titled RSS/Substack platform on August 25, 2026. The episode features an in-depth conversation with his twin brother, SemiAnalysis founder Dylan Patel, about lab economics, compute centralization between Anthropic and OpenAI, and the macroeconomic risks of massive AI capital expenditure.
## TL;DR
Dylan and Dwarkesh Patel discuss how Anthropic and OpenAI are set to outbid competitors and dominate global compute by 2028, driving a massive reallocation of hardware from customer inference back to internal R&D and AGI development. They analyze how $10+ trillion in projected AI capital expenditures could trigger a global sovereign debt crisis, elevate interest rates, and crash non-AI equity valuations. Furthermore, they explore the accelerating concentration of effective AI labor within top labs and the systemic challenges of preventing extreme market and power centralization as recursive self-improvement nears.
## Atmosphere & style
The conversation balances high-stakes macroeconomic foreboding with a warm, fast-paced, and brotherly chemistry filled with lighthearted banter and shared jokes. It feels like an unvarnished intellectual debate between two deeply informed domain experts who seamlessly trade complex financial calculations, hardware supply-chain details, and grand geopolitical predictions. The tone shifts dynamically from playful teasing to a quiet, somber awe at the sheer scale of the financial and technological forces rapidly taking shape.
## Key takeaways
- `05:00` By 2028, Anthropic and OpenAI are projected to directly control the vast majority of usable floating-point operations (FLOPs) on Earth due to their superior monetization capabilities.
- `00:00` Frontier AI lab infrastructure spending is growing exponentially, on track to expand from over $1 trillion in annual capex in 2026 to over $2 trillion annually by 2028.
- `05:00` Hardware performance-per-watt is compounding rapidly; new hardware like GB300, TPUv7, and Trainium3 yields a 3x to 5x efficiency jump over previous-generation chips.
- `07:01` Upstream fab economics demonstrate a massive multiplier effect: $6 billion in fab capex builds 1 GW/year of capacity, enabling over $1 trillion in downstream AI revenue over five years.
- `29:43` Frontier labs will progressively reallocate compute away from external customer inference toward internal R&D and training to accelerate the development of AGI.
- `33:27` U.S. export controls have severely constrained China, cutting its share of new global AI compute additions from over 30% in 2022 to less than 10% currently.
- `43:00` Total global AI capex could reach $7 trillion to $10 trillion annually by the end of the decade when accounting for long-lead power plant and data center shell construction.
- `53:27` An estimated $5 trillion credit shortfall in cumulative AI capex between 2024 and 2029 will force hyperscalers and labs to issue massive corporate debt, driving up market interest rates.
- `48:48` Surging interest rates driven by AI capital demand threaten to trigger a sovereign debt crisis in vulnerable developing nations and depress valuations across non-AI equities.
- `01:03:30` Regulatory delays and safety mandates on public releases will likely accelerate internal lab development through hidden, hyper-fast recursive self-improvement (RSI) loops.
- `01:08:12` A single lab's effective AI workforce is expanding by roughly 10x per year, positioning leading labs to control an AI population larger than the total human workforce before the end of the decade.
- `01:12:00` Massive economies of scale in pre-training, compute bidding power, and algorithmic feedback loops create intense structural economic forces driving total centralization in the AI industry.
## Detailed walkthrough
### `00:00` Two labs will soon control most of the world’s compute
The discussion opens with light banter as Dwarkesh Patel introduces his returning guest, Dylan Patel, founder of SemiAnalysis. Dwarkesh jokes that their recurring podcast appearances have become their version of a "family Thanksgiving dinner," clarifying with a laugh that they are not actually related despite sharing a surname ("Don't tell the people this. We'll destroy the myth"). Shifting into high-stakes AI economics, Dwarkesh highlights how the trajectory of the global economy is increasingly becoming a direct function of compute markets and laboratory economics.
Dylan breaks down the recent historical transition of AI lab funding models. In late 2024, US GDP growth was heavily driven by AI infrastructure, with Anthropic and OpenAI representing roughly a third of all new compute coming online. By 2026, total AI infrastructure capex is reaching just over $1 trillion annually and is on track to cross $2 trillion by 2028. Historically funded by massive venture losses, the labs have aggressively turned the corner toward profitability: Anthropic crossed into profitability in Q2, while OpenAI is expected to turn profitable in Q3 due to heavy adoption of products like Codex and GPT-5.6.
Crucially, unit economics per megawatt have flipped dramatically. While base compute infrastructure costs around $10 million to $15 million per megawatt, early models like GPT-4 on NVIDIA Hopper GPUs initially ran on negative gross margins. Today, next-generation deployments—such as GPT-5.6, Opus 5, and Mythos/Fable 5—are generating up to $50 million per megawatt for Anthropic. This high-margin inference enables the labs to continuously plow massive profits back into raw R&D and training compute.
- **2026 AI Capex:** Projected at over $1 trillion globally, growing past $2 trillion by 2028.
- **Lab Compute Dominance:** Anthropic and OpenAI accounted for roughly 30% of incremental compute added in 2026 and are projected to swallow 40% to 50% of all new compute in 2027.
- **Cluster Scale Growth:** OpenAI and Anthropic began early 2026 at around 2 GW and <2 GW respectively, with both surpassing 5 GW by the end of the year.
- **Compute Sourcing:** Third-party providers like SpaceX are building massive compute installations to lease directly to Anthropic and OpenAI, while the labs also build proprietary capacity (OpenAI developing custom chips; Anthropic procuring Google TPUs deployed via FluidStack).
`05:00` Compute Trajectories and the Dynamic Efficiency Scaling
Dwarkesh pushes on the sheer math of frontier growth, questioning whether the trend of frontier lab compute tripling year-over-year (from ~6 GW in 2026 to 18 GW in 2027 and 54 GW in 2028) must eventually hit a hard physical ceiling against total global additions. Dylan emphasizes that analyzing raw gigawatt numbers understates performance growth because incremental hardware efficiency creates a massive compounding effect.
Because newer hardware like NVIDIA GB300s, Google TPUv7s, and AWS Trainium3s deliver 3x to 5x higher performance per watt than prior generations, a gigawatt deployed in 2027 delivers vastly more compute than one deployed two years prior. By late 2028, Dylan projects that Anthropic and OpenAI will directly control the vast majority of all usable floating-point operations (FLOPs) on Earth.
- **Incremental Power Additions:** Global compute capacity is expanding by roughly 30 GW in 2026, ~50 GW in 2027, and ~70 GW in 2028.
- **Hardware Compounding:** Newer chips (GB300, TPUv7, Trainium3) yield a 3x to 5x performance-per-watt jump over previous-generation accelerators.
`07:01` $6 billion in fab capex enables $1t+ of end revenue
Dwarkesh brings up a calculation from a prior conversation regarding semiconductor manufacturing constraints, noting that producing 1 GW of hardware (like NVIDIA Vera Rubin systems) requires roughly 55,000 N3 wafers, 6,000 N5 wafers, and 170,000 DRAM wafers. After a quick humorous tangent teasing Dwarkesh's pronunciation of "wafers" and his childhood memories of moving to North Dakota with a strong accent asking for "wedgies" instead of wedges, they dive into fab economics.
Dwarkesh runs through a wafer fab equipment (WFE) model showing that $6 billion in fab capex builds 1 GW of annual capacity. Because that single gigawatt goes on to generate roughly $100 billion in annual end-user AI revenue over a multi-year lifespan, $6 billion in fab capex compounding over five years directly enables over $1 trillion in downstream AI revenue. Dwarkesh asks why market forces and massive cash piles do not simply force ASML and Carl Zeiss to build more EUV lithography tools faster to shatter supply bottlenecks.
Dylan explains that the semiconductor supply chain operates like a delayed whip. Carl Zeiss, for example, previously targeted producing fewer optics, but has only recently adjusted to aiming for 100 EUV tools annually by 2030. Even if private equity or hyper-funded AI labs were to inject tens of billions directly into tooling suppliers, physical building constraints and massive capital requirements across energy, data center infrastructure, and WFE mean the global supply chain cannot pivot overnight.
- **Hardware Production Bill:** 1 GW of Vera Rubin-class compute requires 55,000 N3 wafers, 6,000 N5 wafers, and 170,000 DRAM wafers.
- **Multiplier Effect:** $6 billion of upstream fab capex yields 1 GW/year, generating over $1 trillion in cumulative downstream AI revenue over five years.
- **Infrastructure Market Spend:** Total WFE spending runs around $200 billion, nested within a broader data center, accelerator, and energy capex structure surpassing $2 trillion overall.
`13:08` Compute prices will rise if the labs outbid everyone
As the labs expand toward a combined 100 GW footprint by 2028, Dylan highlights that they will capture 70% to 80% of all incremental compute. This creates intense competitive disruption because virtually anyone can generate positive returns on standard $10 million to $15 million per megawatt hardware. Dylan points out that setting up a GB300 rack running open weights like Kimi or OpenRouter via SGLang is simple enough that "Codex and Fable can actually help you do this."
However, because Anthropic and OpenAI can monetize compute at vastly higher rates through advanced frontier models, they will simply outbid traditional hyperscalers, non-AI enterprises, and smaller players for raw power and hardware, forcing compute market prices to skyrocket.
> "Anthropics and OpenAI, we could make a trillion dollars right now, but we're just bottlenecked on the mirrors that go into the ASML machines." — **Dwarkesh Patel**
- **Inference Commoditization:** Standard commercial models easily turn a profit on $10M–$15M/MW infrastructure using standard serving stacks (SGLang, vLLM, OpenRouter).
- **Outbidding Dynamics:** Frontier labs' willingness to pay premium prices forces hyperscalers (Google, Meta, Amazon) to prioritize internal/lab allocation over general market availability.
`15:00` Outbidding for Compute and the Mechanics of Value Capture
The baseline cost of compute currently sits between $10 million and $15 million per megawatt, but reaching the ambitious goal of 100 gigawatts by 2028 will require compute prices to inflect drastically upward—potentially reaching $25 million, $30 million, or even $50 million per megawatt. Frontrunner AI labs generate significantly more revenue per megawatt than any other sector. If internal recursive self-improvement (RSI) gives these labs an edge through unreleased internal models, their ability to outbid all competitors will compound. Third-party compute providers, such as SpaceX or secondary hyperscalers, will naturally auction off their capacity to the highest bidder if they cannot monetize it as effectively in-house. However, regulatory headwinds—such as OpenAI pausing training for two weeks or Anthropic withholding safety-assessed models like Model 2 (widely believed to be the successor to Mythos)—threaten to stall revenue growth per megawatt by leveling the playing field for open-source and foreign competitors.
A useful mental model for monetizing compute considers the value of fully automating white-collar work. If a single gigawatt could continuously host one million automated software engineers earning an equivalent $100,000 salary, that single cluster would generate $100 billion in direct economic value. When accounting for full AGI capabilities, a gigawatt could produce hundreds of billions of dollars. Despite these massive economic projections, the vast majority of consumer surplus generated by AI models is currently captured by end users rather than the labs themselves. High-value enterprise customers utilize frontier models to extract orders of magnitude more economic value than the token fees they pay to the lab.
- Financial market firms leverage exclusive tier access, such as custom ultra-fast modes of frontier models, to capture immense downstream profits that far exceed their API expenditures.
- Current compute economics show a stark disparity, with labs achieving a fourfold or greater margin between the raw cost of compute and their internal revenue yield per gigawatt.
- In an efficient, equilibrium-driven market, the selling price of raw compute should eventually track closely with the maximum revenue yield that frontier labs can extract from it.
`20:00` Supply Chain Rebalancing and Opportunistic Compute Arbitrage
The location of value capture across the AI stack has fluctuated dramatically over time. A year prior, hardware supply chain companies reaped record gross margins while model providers burned venture capital operating on negative gross margins. Before that, memory manufacturers lost money on high-bandwidth memory (HBM) despite supplying essential components. Now, frontier labs are transitioning toward strong positive gross margins and targeting yields of $100 million per megawatt. Simultaneously, opportunistic players like Elon Musk have capitalized on near-term supply constraints by renting out buildouts to major labs at massive markups—selling compute packages for $25 million to $40 million per megawatt, or roughly $40 billion per gigawatt, allowing them to recoup entire capital expenditure budgets in a single year.
Most global compute capacity cannot be monetized at these extreme spot prices because it must be pre-financed via strict credit agreements tied to long-term enterprise commitments. Only entities with fortress balance sheets—such as Meta, SpaceX, or major cloud providers—possess the capital to construct gigawatt-scale infrastructure uncommitted. These balance-sheet powerhouses hold the unique optionality to either deploy massive clusters internally or rent them out to desperate labs at exorbitant spot rates.
- Standard uncommitted compute clusters continue to transact below $20 billion per gigawatt when bound to conventional credit market financing structures.
- Entities that can build compute upfront without pre-arranged customers exert immense pricing power over the rest of the ecosystem.
- Downstream suppliers across the entire stack—from substrate manufacturers to memory vendors—continually adjust their pricing upward as spot compute rates escalate.
`25:00` Regulatory Drag and the Revenue Cap per Megawatt
By the end of 2027, blended company-wide revenue per megawatt for leading labs could reach $70 million to $80 million, provided regulatory barriers do not completely suppress deployment. A primary drag on monetization is that top-tier models are frequently held back from public rollout due to safety evaluations and policy constraints. When state-of-the-art models remain locked in internal environments, labs cannot leverage them to optimize inference performance, refine internal developer tools, or drive external API monetization.
Beyond internal safety holds, external regulatory constraints on physical infrastructure are growing significantly. Regional moratoriums, localized data center bans, and aggressive municipal tax hikes threaten to restrict the total physical supply of power and space. This artificially constrained supply drives up baseline operational costs, creating a stagflationary effect where deployment slows down publicly while underlying frontier capabilities continue to advance behind closed doors.
- Leading labs may be forced to keep their most capable models six months ahead of publicly released versions due to safety testing mandates and competitive posturing.
- The widening gap between internal frontier capabilities and publicly deployed models serves as the primary bottleneck capping revenue per megawatt.
- Regional legislation, including local data center freezes and local tax disputes, creates structural headwinds for rapid gigawatt-scale buildouts.
Here is a detailed, faithful walkthrough of the transcript section.
`30:00` Reallocating Compute from Inference to R&D and AGI
Dylan Patel argues a highly non-consensus view: AI frontier labs like Anthropic and OpenAI are going to allocate a progressively smaller fraction of their compute to customer-facing inference and a much higher percentage to internal R&D and training over time. Currently, the standard industry belief is that most compute will naturally gravitate toward inference and forward passes. However, the economics point in the opposite direction.
Today, labs generate roughly $30 million to $40 million per megawatt, allocating around 40% of their compute to inference. As efficiency and model capabilities improve, driving that revenue generation up to $60 million or $70 million per megawatt, labs will face a core choice: do they maintain that 40% inference split, print massive profits, and issue dividends or share buybacks? Or do they direct that compute back toward building Artificial General Intelligence (AGI)? Dylan asserts that executives and boards at both OpenAI and Anthropic will unequivocally choose building AGI, as the long-term discounted cash flows and future earnings potential of achieving AGI vastly outweigh short-term token sales.
Even internal deployment of fast or frontier models yields far more marginal value to a lab than selling those same tokens to external API customers. The primary purpose of maintaining a large inference footprint today is simply to finance and scale the underlying training fleet. Dwarkesh Patel notes that this presents a fascinating economic dynamic. Dylan points out that this shift is already occurring: over recent months, despite revenue growth plateauing after massive initial spikes, labs have continuously ramped up total compute capacity, meaning the marginal megawatts acquired are swinging heavily toward internal R&D rather than customer inference.
- In early 2026, Anthropic added less compute in January than in December, yet its revenue additions skyrocketed before plateauing, showing that marginal capacity is now funneling into R&D.
- Standard historical lab compute allocation sits at roughly 60% training and 40% inference.
- Within the training budget, roughly 50% of compute goes to basic research and data generation, 10% goes to actual model development/pre-training, and the remaining 40% is allocated to inference.
---
`33:27` Global Compute Growth and China’s Severe Bottlenecks
Looking at global capacity projections, total world AI compute is on track to cross 200 gigawatts by the end of 2028. Annual capacity additions are trajectory-mapped at 30 gigawatts in 2026, 50 gigawatts in 2027, 70 gigawatts in 2028, and reaching 90 to 100 gigawatts annually by 2029. Predicting beyond four years is extremely difficult, as recursive self-improvement (RSI) regimes or 10%+ annual global GDP growth could radically shift the slope upward.
A stark disparity exists between Western labs and China. Back in 2022, the U.S. added about 45% to 50% of global compute, China added 30% to 35%, and the rest of the world absorbed the remainder. Following strict U.S. export controls and massive American capital deployment, China now accounts for less than 10% of all incremental data center AI compute additions globally. By 2028, China will likely hold 30 gigawatts or less of total AI compute. While Chinese domestic fabs (such as SMIC and CXMT) and memory suppliers (like CXMT) will begin scaling up production around 2027–2028, their chips remain vastly inferior in performance and efficiency compared to Western hardware from Nvidia or Google. Consequently, a single leading American lab in 2028 may possess more quality-adjusted compute than the entire country of China will control in 2029 or 2030.
Dwarkesh reflects on this gap, noting that he had previously steel-manned anti-export control arguments from a libertarian perspective, wondering if cooperation would be better given China's dominance in supply chains and robotics. However, seeing how severely export controls and Western financial markets have restricted China's compute stock changes the picture, revealing the sanctions as a notable policy success. Dylan cautions that while Western capital markets "YOLO" into startups faster, the Chinese state provides immense subsidies—far exceeding the rest of the world's semiconductor subsidies combined—meaning China could catch up significantly if technological progress takes longer than expected.
Furthermore, Chinese labs currently remain surprisingly competitive on benchmark models despite their tiny compute budgets. Most leading Chinese labs operate on only 100 to 200 megawatts of total compute (with ByteDance's Seed team being the primary exception), whereas Anthropic is approaching well over 5 gigawatts by the end of the year. Historically, actual pre-training runs have taken up a relatively small footprint—for instance, pre-training Anthropic's Mythos model required under 200 megawatts over two months, with reinforcement learning (RL) using even less single-site compute. The vast majority of lab capacity went to human researchers testing ideas, architectures, and hyperparameters across fragmented clusters. However, as automated AI researchers, automated coding, and continual learning take over, the boundaries between research and training will blur, allowing labs to throw multi-gigawatt clusters directly into model training and widening the gap against compute-starved Chinese competitors.
- Global annual compute additions: ~30 GW (2026), ~50 GW (2027), ~70 GW (2028), and ~90–100 GW (2029).
- China's share of new global compute additions dropped from ~30–35% in 2022 to under 10% currently.
- Projected 2028 total AI compute: Global ~200 GW; China $\le$ 30 GW. China may add ~50 GW incrementally in 2029, but on lower-quality domestic hardware.
- Anthropic's Mythos model pre-training run required under 200 MW for approximately two months.
---
`43:00` The Trillion-Dollar CapEx Wall and Economic Friction
Deploying compute at a scale of 100 gigawatts per year imposes staggering capital costs that extend far beyond buying server hardware. At current IT hardware pricing, 100 gigawatts represents approximately $5 trillion in annual capital expenditures. However, because data centers (15-to-20-year assets) and power infrastructure (30-year assets) must be funded and constructed years in advance, true annual expenditures will skyrocket much higher.
When industry commentators cite $40 billion to $50 billion in AI CapEx, they are typically referring only to critical IT infrastructure—servers, networking gear, optical transceivers, and fiber. They fail to account for physical data center shells, power grid interconnects, and dedicated power plants. To support 150 gigawatts next year, the physical buildings must be paid for today; to support 200 gigawatts the following year, power plant turbines must be ordered immediately. As a result, total incremental AI CapEx could easily approach $7 trillion to $10 trillion annually by the end of the decade.
This sheer scale would represent nearly a tenth of the entire global economy and upwards of a third of the U.S. economy devoted strictly to data center construction and power generation. Both speakers express awe at these figures, questioning whether political systems, credit markets, and broader capital markets will even permit such an extreme reallocation of physical and financial resources toward a single industry without triggering severe economic friction.
- Deploying 100 GW of server compute per year costs ~$5 trillion in direct IT CapEx alone.
- Total annual AI CapEx (factoring in advance power plant and data center shell construction) could reach $7 trillion to $10 trillion by the end of the decade.
- At $10 trillion annually, AI infrastructure spending would consume roughly 25% to 33% of the total U.S. economy.
`45:00` The Scaled-Up Funding Gap: Who Pays for $3–4 Trillion in CapEx?
Dylan and Dwarkesh break down the financial reality of scaling global AI compute to 100 gigawatts by 2028 or 2030, which requires roughly $3 to $4 trillion in capital expenditure across IT ($2.5T), data centers, energy infrastructure, and semiconductors ($1–2T). Historically, hyperscalers like Google, Microsoft, Amazon, and Meta funded the vast majority of compute out of internal cash flow—for example, by redirecting funds previously earmarked for stock buybacks into purchasing compute infrastructure. However, today's hyperscalers no longer generate spare cash flow; they spend every dollar on CapEx and have begun taking on massive amounts of corporate debt.
As the hyperscalers and their supply chains prepare to raise hundreds of billions of dollars in incremental debt over the next few years, the key question becomes who will provide this capital. Dylan outlines several emerging sources of funding:
* **Semiconductor vendors:** Nvidia, Broadcom, and memory makers reinvesting profits directly back into funding customer CapEx.
* **Infrastructure investors:** Traditional private infrastructure funds redirecting capital from projects like bridges and roads into data center assets.
* **Broad capital markets & debt investors:** Retail and institutional investors shifting away from home mortgages, equities, or government treasuries into high-yielding corporate debt issued by hyperscalers, data centers, or AI labs.
To illustrate this shift, Dylan notes that an AI lab like Anthropic might be fully willing to pay a 20% interest rate on a $1 billion debt raise to build out data center capacity. Because their projected returns from AI models are astronomically high, paying a 20% yield on debt remains far cheaper than renting compute from third parties at exorbitant prices (e.g., $50 billion per gigawatt).
- Projected total CapEx needed by 2028: $3 trillion to $4 trillion ($2.5T IT CapEx; $1–2T energy, data centers, and supply chain).
- Shifts in capital allocation: Broad consumer and institutional pivot from real estate debt and sovereign bonds toward high-yield AI infrastructure bonds.
---
`48:48` The Threat of an AI-Driven Sovereign Debt Crisis and Interest Rate Spikes
The discussion pivots into an off-air debate between Dwarkesh and Dylan about whether the massive demand for AI capital could trigger a global sovereign debt crisis through a severe "crowding out" effect. Because AI data center investments offer extraordinarily high returns—turning $1 of depreciated CapEx into $2 to $10 of annual revenue when rented out to frontier labs—lenders can demand much higher interest rates. This forces the entire global economy to borrow at significantly higher costs.
Dwarkesh monologues on the fiscal vulnerabilities this creates for national governments, noting that while the U.S. might weather the storm by taxing domestic AI data centers, its existing debt dynamic is fragile:
* **Corporate vs. Payroll Taxes:** Corporate income tax makes up less than 10% of U.S. federal revenue, while over 80% comes from payroll and income taxes. As AI automates human labor, the traditional tax base could contract.
* **Debt Servicing Costs:** Around 20% of federal spending currently goes to interest payments on national debt. Because much of the U.S. debt rolls over every five years, a 500-basis-point (5%) increase in market interest rates—combined with $2 trillion in annual deficit borrowing—could push debt servicing costs to well over 60% of total tax revenue.
The twin brothers share a lighthearted, self-deprecating moment when Dylan laughs at Dwarkesh's deep dive into macroeconomics, teasing him that he only learned about government debt rollover in the past month despite his financial economics degree. Dwarkesh laughs back, acknowledging how funny it is that a podcast host with a newly acquired grasp on macro dynamics is broadcasting to a million listeners.
Moving beyond the U.S., Dwarkesh points out that heavily indebted developing countries with low tax revenues and frequent debt rollovers (such as Pakistan or Nigeria) face severe default risks in this high-interest regime. Dylan adds that debt-reliant corporate sectors—consumer packaged goods, telecom, and banks—will experience massive squeeze. The widening credit spread between official central bank rates and what market participants demand (due to massive bond issuances from companies like Amazon or Meta) imposes a structural brake on AI expansion, slowing down the buildout below pure economic demand.
- U.S. Federal Revenue Structure: <10% corporate income tax, >80% payroll/income tax.
- Projected debt servicing cost: Rising from ~20% of federal tax revenue to over 60% if market interest rates increase by 5 percentage points amid $2T annual deficits.
---
`53:27` $11 Trillion in Capex: $6T Cash Flow vs. $5T Credit Demand
Dylan breaks down the proprietary financial modeling done by his research group for the 2024–2029 timeframe. The model forecasts a staggering $11 trillion in cumulative AI CapEx globally over this five-year period. Even under optimistic assumptions where AI revenues expand rapidly, organic lab and hyperscaler cash flows can only cover roughly $6 trillion of this spending, leaving a massive $5 trillion credit shortfall that must be funded through debt issuance across the ecosystem.
```
Global AI CapEx (2024–2029): $11 Trillion
┌──────────────────────────────┬──────────────────────────────┐
│ Funded by Cash Flow: $6T │ Funded by Credit: $5T │
└──────────────────────────────┴──────────────────────────────┘
```
Dwarkesh pushes back, asking whether explosive AI revenue growth—doubling or tripling year-over-year—could allow labs to fund a larger share out of cash flow. Dylan replies that structural bottlenecks, diffusion delays, regulatory hurdles, safety restrictions, and rising real-world capital costs prevent unlimited revenue growth from fully offsetting the need for credit. Therefore, the market must issue ~$5 trillion in corporate debt, exerting sustained upward pressure on interest rates.
- Total projected global AI CapEx (2024–2029): $11 trillion.
- Capex financing breakdown: $6 trillion funded by internal cash flows; $5 trillion funded via credit and debt markets.
---
`55:00` Economic Ripple Effects: Equity Crashes, Discount Rates, and a "Second Volcker Shock"
Expanding on the consequences of rising interest rates, Dylan estimates that corporate bond yields for top-tier tech companies like Meta could jump from their current 5%–6% range up to 8% or higher, reflecting a 250+ basis-point increase. Because AI compute yields immense returns, hyperscalers will happily pay 8% to secure debt, but this forces higher borrowing costs across the rest of the global financial system.
Dwarkesh and Dylan detail three major macroeconomic fallout scenarios resulting from this interest rate surge:
1. **Banking Sector Stress:** As market credit spreads blow up, bank liabilities (short-term debt) reprice upward much faster than their long-term assets, threatening bank profitability and balance sheets.
2. **Equity Market Devaluations:** Higher market interest rates increase the corporate discount rate used in discounted cash flow (DCF) models. A jump in discount rates from 3%–5% to 8%–10% drastically depresses net present values, causing traditional cash-flow stocks (such as Berkshire Hathaway-style value equities) to crater, even if the broader S&P 500 holds up due to AI tech growth.
3. **A "Second Volcker Shock" for Emerging Markets:** Referencing insight from economist Basil Halper, Dwarkesh notes that when Federal Reserve Chair Paul Volcker raised real interest rates sharply in the early 1980s to fight inflation, it triggered sovereign debt defaults in roughly 40 developing nations (primarily across Latin America). A similar wave of defaults is likely to strike emerging markets prior to any technological singularity, as capital is pulled away from sovereign debt into high-yielding AI infrastructure.
Looking further ahead to a potential post-automation economy, Dwarkesh references research by Damon Binder regarding input-output tables. He explains that while human labor currently acts as a hard physical bottleneck on GDP growth (capping traditional economic expansion at ~3% annually), a fully automated AI economy could theoretically remove this constraint and drive annual GDP growth to unprecedented, compounding levels.
- Corporate debt pricing: Meta/hyperscaler bond yields expected to rise from ~5.5% to 8%+ (+250 bps).
- Discount rate impact: Increase in discount rates to 8%–10% severely reduces net present value for non-AI cash-flow equities.
- Historical parallel: Paul Volcker's 1980s rate hikes caused ~40 sovereign defaults in developing nations; AI debt demand threatens a similar global shock.
`01:00:00` Economic Regimes, Escalating Interest Rates, and Market Multiples
Dwarkesh Patel opens with a staggering macro economic prediction: if an AI-driven economy can double the effective labor force every single year, overall economic growth could easily compound at tens or even hundreds of percent annually throughout the 2030s. Under standard economic theory, interest rates must reflect this staggering real growth rate. However, a world with interest rates in the tens or hundreds of percent creates catastrophic systemic consequences for the non-AI financial system. Every nation not actively engaged in AI production would default on its debt; equities for non-AI businesses would crater as discounted cash flows vanish into near-zero valuations; and normal consumer finance—such as getting a mortgage—would completely collapse.
In plain terms, the massive opportunity cost of capital shifts the entire global paradigm. Dylan Patel points out that when the return on capital spent building "a robot factory that builds a robot factory that builds a robot factory" is astronomically high, government borrowing to pay pensions becomes prohibitively expensive. This dynamic crushes equity markets. AI-believers often wonder why memory producers like Micron, SK Hynix, or Kioxia trade at low multiples of two to three times earnings despite roaring demand; Dylan argues that if AI causes this level of macroeconomic disruption, *everything* across the global market should logically trade at a compressed 2x to 3x earnings multiple.
```
+-------------------------------------------------------------------------------+
| THE HIGH-INTEREST RATE CASCADE |
+-------------------------------------------------------------------------------+
| Exponential AI Labor Scaling --> Global Economic Growth at Tens/Hundreds % |
| | |
| v |
| Astounding Capital Opportunity Cost --> Real Interest Rates Spike Skyward |
| | |
| v |
| Non-AI Sovereign Debt Defaults <-- Equities & Valuations Crater (2-3x PER) |
+-------------------------------------------------------------------------------+
```
- Meta's $1.5 trillion valuation looks conservative when considering their cash flows, infrastructure hoarding, and ability to resell compute for absurd dollars-per-watt yields.
- The fundamental limiter on AGI isn't how fast frontier research engineers (such as their roommate Sholto Douglas) can work; it is how much the rest of the world lets capital reallocation happen before stepping in with regulations, higher interest rates, and bans on data centers or fabs.
- Anthropic and OpenAI are responding to external bottlenecks by directly contracting power, designing their own chips, and building proprietary infrastructure to bypass traditional market dependencies.
---
`01:03:30` Regulatory Bottlenecks, Internal Deployment Restrictions, and Self-Improvement
The conversation turns to political and societal pushback, which both brothers expect will act as a major brake on a straight-line, explosive takeoff. Governments are already imposing heavy deployment restrictions, and Dylan warns that politicians and voters alike will become deeply hostile to AI as stock portfolios plummet and jobs shift. Dylan envisions a scenario in the late 2020s where regulatory authorities force frontier labs to sit on new models for six months before public release. Paradoxically, this attempt to slow down AI might trigger an unmanageable internal Recursive Self-Improvement (RSI) loop: during those six months of delayed external deployment, labs will run models internally, achieving years worth of AI capabilities while the public remains trapped behind legacy systems.
```
+-------------------------------------------------------------------------------+
| THE DANGEROUS DISCONNECT OF DEPLOYMENT DELAYS |
+-------------------------------------------------------------------------------+
| Government Mandates 6-Month External Deployment Delay |
| |
| PUBLIC REALM: Stuck on legacy models; years behind frontier capabilities. |
| INTERNAL LAB: Runs continuous recursive self-improvement (RSI) loops. |
| Gains 3 to 6 years of technical progress in 12 months. |
+-------------------------------------------------------------------------------+
```
Dwarkesh highlights the irony that preventing external release might accelerate a hidden, dangerous singularity inside lab walls. Furthermore, regulatory agencies aren't stopping at public release bans—they are beginning to restrict internal lab usage as well. Dylan references reports that Anthropic had to temporarily halt giving access to their internal model, Mythos, to foreign employees. As regulatory panic grows, the brothers predict political backlash will hit close to home, joking that even their parents will call them complaining that their work is accelerating societal collapse.
- A single year of delayed deployment alongside 2x–3x annual compute scaling represents a massive 3 to 6 years of compressed AI progress happening entirely behind closed doors.
- Systemic political hostility will be driven by voters losing non-AI equity value and facing widespread employment shifts.
- Government intervention could force a chaotic environment where societal institutions "tear themselves apart" before reaching safe AGI.
---
`01:06:40` Jane Street ML Internships
Dwarkesh pauses the discussion for a sponsor message highlighting Jane Street's internship opportunities. The quantitative trading firm is hiring for two distinct ML tracks for 2027: Machine Learning Engineering and Machine Learning Research. Dwarkesh shares insights from a conversation with Alok, who helps run the research track, detailing how interns tackle real-world financial problems rather than contrived exercises.
- The ML research program focuses on understudied deep learning problems, such as modeling thousands of interconnected, non-stationary, high-dimensional time series with extremely low signal-to-noise ratios.
- A typical intern project involves adapting a newly published frontier LLM paper to complex, adversarial financial markets.
- No prior finance background is required—Jane Street provides financial domain training to candidates with solid machine learning foundations.
---
`01:08:15` Centralization of Compute and the AI Labor Explosion
Dwarkesh shifts to a mind-bending scaling metric: the exponential growth of effective AI labor within individual frontier labs. Compute available at the frontier grows 4x–5x annually in FLOP terms, while the compute required to achieve human-level task execution decreases by roughly 3x per year due to algorithmic efficiency. Compounding these factors means that a lab's internal "effective AI population" grows by roughly **10x year-over-year**.
While today's models cannot operate fully autonomously, this trajectory implies that a lab going from 10 million AI labor equivalents in one year will reach 100 million the next, 1 billion the year after, and eventually surpass the entire human population of Earth before the end of the decade—even without factoring in recursive self-improvement.
```
+-------------------------------------------------------------------------------+
| EFFECTIVE AI POPULATION SCALING WITHIN A SINGLE LAB |
+-------------------------------------------------------------------------------+
| Compute FLOP Scaling (~5x/yr) + Algorithmic Efficiency (~3x/yr) |
| = |
| ~10x Annual Expansion of AI Labor Force |
| |
| Year 1: 10 Million AI Workers |
| Year 2: 100 Million AI Workers |
| Year 3: 1 Billion AI Workers |
| Year 4+: AI Population > Total Human Population on Earth |
+-------------------------------------------------------------------------------+
```
This structural shift concentrates the vast majority of future economic labor and intellectual output into just two dominant labs: Anthropic and OpenAI. Dwarkesh notes that public discourse focuses heavily on nationalization, but misses the immediate threat: world power is rapidly centralizing into these tiny corporate hubs. If these central models become misaligned, the majority of the world's functional intelligence becomes misaligned simultaneously.
> "Even if compute scaling slows down, it doesn't take many more years before each company individually has more labor equivalents than there are people on Earth."
- Scaling out AI labor yields massive centralization forces: amortizing training across billions of users gives early leaders unmatched unit economics.
- Compute scarcity allows the leading lab to charge higher markups, reinvest profits, and outbid competitors for hardware.
- Gavin Baker's recent critique—joking that Dario Amodei acts like there is only one company in the world—reflects real structural economic drivers pulling the industry toward total centralization.
---
`01:12:00` Centralization vs. Decentralization: The Post-AGI Dilemma
The brothers explore whether any force can counter this relentless march toward centralization. Every key mechanism—economies of scale in pre-training, deployment data loops, continuous learning, compute bidding dynamics, and RSI—pulls authority directly into the hands of the top lab. Dwarkesh calls for an intellectual crusade to conceptualize a decentralized, broadly empowered future after AGI, though both acknowledge how difficult it is to engineer against these raw economic incentives.
```
+-------------------------------------------------------------------------------+
| FORCES DRIVING TOTAL AI CENTRALIZATION |
+-------------------------------------------------------------------------------+
| 1. Pre-training Amortization --> Fixed costs spread over billions of users. |
| 2. Hardware Bidding Power --> Highest-performing models earn top markups. |
| 3. Data Feedback Loops --> More deployment yields richer real data. |
| 4. Recursive Self-Improvement --> Better models directly build better models.|
+-------------------------------------------------------------------------------+
```
Dylan points out the historic irony: standard capitalist economies outpaced centralized command economies because decision-making and capital were decentralized under the rule of law. However, AI flips this dynamic on its head by making hyper-centralized capital ownership infinitely more efficient. Currently, AI infrastructure accounts for around 2% of US GDP (roughly $1 trillion of a $30+ trillion economy), but a handful of entities—NVIDIA, Anthropic, OpenAI, and hyperscalers—control almost all of it.
- Without heavy state intervention or deliberate algorithmic slowdowns, the future defaults to an extreme concentration of resources where humanity must blindly trust a single private entity.
- The state alternative—government nationalization and control—is equally uncomfortable, with Dylan admitting: "I don't trust the government, and I don't trust Dario, and I don't trust Sam."
- Market dynamics are driving us toward an existential choice between hyper-centralized corporate power, draconian state slowing, or unprecedented regulatory intervention.
---
`01:14:45` Value Capture and Internalizing Compute
In closing, Dylan offers one small counter-argument to immediate total concentration: currently, frontier labs do not capture all the value they generate. Anthropic may pay $13 million per megawatt for compute that produces $100 million in gross economic value, because external end-users capture the surplus. For instance, quantitative trading firms like Jane Street generate $300 million to $500 million per megawatt leveraging advanced algorithms and credit market insights.
However, both brothers agree this value leakage is a temporary anomaly. As labs realize that internal R&D yields far higher returns than selling API tokens to external entities, they will stop exporting compute. Instead of letting external firms profit off options trading or financial arbitrage, labs will hoard their FLOPs internally for model training, self-improvement, and direct execution.
> "Why would Anthropic allocate compute to that? If the end monetization that Jane Street has per megawatt is $200, so they're willing to pay Anthropic $100, well, what if Anthropic can just generate hundreds of millions of dollars per megawatt by using that compute internally? And that's what's happening."
- Internal returns on AI labor inside frontier labs are rapidly outpacing the value generated by external business deployments.
- The economic logic forcing labs to shift compute from external inference to internal R&D inevitably closes the door on downstream ecosystem value capture.
- Dwarkesh and Dylan end on a somber note, agreeing to meet for their next podcast when Recursive Self-Improvement has officially kicked off.
## Notable quotes
> `13:08` Anthropics and OpenAI, we could make a trillion dollars right now, but we're just bottlenecked on the mirrors that go into the ASML machines. — Dwarkesh Patel
> `00:00` Don't tell the people this. We'll destroy the myth — Dwarkesh Patel
> `01:14:45` Why would Anthropic allocate compute to that? If the end monetization that Jane Street has per megawatt is $200, so they're willing to pay Anthropic $100, well, what if Anthropic can just generate hundreds of millions of dollars per megawatt by using that compute internally? And that's what's happening. — Dylan Patel
> `01:12:00` I don't trust the government, and I don't trust Dario, and I don't trust Sam. — Dylan Patel
> `01:08:15` Even if compute scaling slows down, it doesn't take many more years before each company individually has more labor equivalents than there are people on Earth. — Dwarkesh Patel
## Mentioned
Dylan Patel, Dwarkesh Patel, Anthropic, OpenAI, SemiAnalysis, Codex, GPT-4, GPT-5.6, Opus 5, Mythos, Fable 5, NVIDIA, Hopper, GB300, Vera Rubin, SpaceX, Google, TPUv7, AWS, Trainium3, ASML, Carl Zeiss, Kimi, OpenRouter, SGLang, vLLM, Meta, Amazon, Elon Musk, SMIC, CXMT, ByteDance, Damon Binder, Paul Volcker, Basil Halper, Berkshire Hathaway, Sholto Douglas, Gavin Baker, Dario Amodei, Sam Altman, Jane Street, Grok Bot, Antithesis, FluidStack
Dylan Patel argues that Anthropic and OpenAI will capture the vast majority of global usable compute by 2028, driven by dramatic jumps in hardware performance and outsized revenue generation per megawatt. This rapid concentration threatens to destabilize global debt markets and centralize economic and intellectual output into two private labs.
Unit Economics and Compute Outbidding (0:00)
Frontier AI labs are shifting from loss-making research entities to high-margin operations. While base infrastructure costs $10 million to $15 million per megawatt, Dylan Patel notes that modern deployments like GPT-5.6 and Opus 5 generate up to $50 million per megawatt for Anthropic. Anthropic reached profitability in Q2 2026, with OpenAI expected to follow in Q3. Dylan projects total annual AI infrastructure capex will expand from over $1 trillion in 2026 to past $2 trillion by 2028, with the two leading labs taking 40% to 50% of new capacity additions in 2027.
Hardware Compounding and China Bottlenecks (5:00)
Hardware efficiency gains amplify the disparity in raw power. Accelerators like NVIDIA GB300s, Google TPUv7s, and AWS Trainium3s deliver 3x to 5x higher performance per watt than prior generations, allowing 2027–2028 gigawatts to produce far more compute than earlier deployments. Upstream, $6 billion in wafer fab equipment capex yields 1 GW per year of Vera Rubin-class systems, unlocking over $1 trillion in downstream revenue over five years, though supply chains remain constrained by ASML and Carl Zeiss tooling lead times.
The Divergence Between US Labs and China (33:27)
Dylan Patel emphasizes that export controls and aggressive Western capital spending have cut China's share of incremental AI compute from 30%–35% in 2022 to under 10% by 2026. Global compute is mapped to surpass 200 GW by late 2028, with China expected to hold 30 GW or less, predominantly on less efficient domestic chips from SMIC and CXMT. As training transitions from fragmented, human-driven experiments to automated AI research that consumes multi-gigawatt clusters, the performance gap between leading American labs and compute-starved Chinese competitors is projected to widen significantly.
Macroeconomic Risks and Debt Market Strains (48:48)
Dylan's research group projects an $11 trillion cumulative AI capex requirement globally between 2024 and 2029, with cash flows covering roughly $6 trillion and leaving a $5 trillion credit deficit. As tech companies issue high-yielding debt to finance infrastructure, borrowing costs across the economy could surge. Dwarkesh Patel warns that a 500-basis-point increase in interest rates would severely elevate US debt-servicing costs, which currently absorb roughly 20% of federal spending, while squeezing corporate debt and raising default risks for heavily indebted emerging markets.
Shift to Internal R&D and Centralization (68:15)
Contrary to the industry view that inference will dominate workloads, Dylan expects frontier labs to direct a larger proportion of compute to internal R&D and recursive self-improvement rather than external API monetization. Dwarkesh Patel estimates that combining a 4x–5x annual increase in FLOPs with a 3x annual improvement in algorithmic efficiency expands a single lab's effective AI labor force by roughly 10x per year. This creates intense centralization, concentrating intellectual capacity and economic leverage inside two private companies while bypassing traditional distributed market structures.
The discussion concludes that economic incentives heavily favor the centralization of frontier compute within Anthropic and OpenAI. However, unresolved bottlenecks—such as supply chain delays, public and regulatory backlash, and strains on global debt markets—remain substantial friction points that could disrupt this trajectory.
[13:08] Dwarkesh Patel: Anthropics and OpenAI, we could make a trillion dollars right now, but we're just bottlenecked on the mirrors that go into the ASML machines.
[0:00] Dwarkesh Patel: Don't tell the people this. We'll destroy the myth
[74:45] Dylan Patel: Why would Anthropic allocate compute to that? If the end monetization that Jane Street has per megawatt is $200, so they're willing to pay Anthropic $100, well, what if Anthropic can just generate hundreds of millions of dollars per megawatt by using that compute internally? And that's what's happening.
[72:00] Dylan Patel: I don't trust the government, and I don't trust Dario, and I don't trust Sam.
[68:15] Dwarkesh Patel: Even if compute scaling slows down, it doesn't take many more years before each company individually has more labor equivalents than there are people on Earth.
Original transcript
[0:00] Okay, I'm back with Dylan Patel, founder of SemiAnalysis. Our version of a family Thanksgiving dinner is a regular yearly podcast. We are not actually related. Don't tell the people this. We'll destroy the myth. Basically, where the world economy is headed is more and more becoming a function of where lab economics are headed, where the compute market is headed, etc. So I want to understand where the crazy future ends up within a few years. But let's start with just where we are today. So walk me through lab compute, and lab revenue right now, and maybe projecting out a year or two? Yeah, so when we go back to last year, even at the end of the year, most of GDP growth in America was just AI infrastructure. And as we look towards this year, about a third of the compute coming online is for the labs, for Open Ananthropic. Now, it may be built by others and then rented to them, but it's at the end customer, it's them. As we go forward into the future, the numbers for computer ballooning, right? We're at a little bit over a trillion dollars of capex this year. As we go out into 28, it's going to be more than two trillion dollars. The labs are also taking an increasing percentage of this. And so ultimately, you've got a very interesting situation where the labs are going from companies that spend tens of billions of dollars a year to hundreds of billions of dollars a year to forecasting to spend trillions of dollars a year even towards the end of the decade. And this is at least some of the contracts they've begun signing with their partners. And so this requires a big reshaping of what happens with their economics. Now, they have been companies that mostly lost money. Anthropics started turning a profit in Q2. It's believed at some point in Q3, OpenAI could potentially start turning a profit even with the big rise of Codex and 5.6 and all this. But if we go back a year ago. Everything that they all the money they had was venture funded losses. If we go back to even the beginning of this year, those venture-funded losses, they've now turned the corner and are actually starting to profit. Now, that doesn't mean they're not taking in new capital. The new capital is still coming in to accelerate the growth further. But ultimately, there's more and more of their business is being funded off of their own revenue rather than capital injections into them. Over the last year and a half, their margins have really skyrocketed. The base cost of compute tends to be around $10 or $13 or $15 million per megawatt. The most interesting aspect about what's happening now is before, again, they were generating, if they served a model, GPT-4 being served on… you know, NVIDIA Hopper GPUs was generating negative gross margin for OpenAI. But now when OpenAI serves GPT 5.6 or Anthropic serves Opus 5 or Mythos, Fable 5, their revenue generation has passed well beyond the sort of incremental $10, $15 million per megawatt. In the case of Anthropic, the revenue has gone as high as $50 million per megawatt. And what that now enables them to do is, hey, if I spend $10 on inference capacity, I actually generate $50 of revenue, and then I can turn around and incrementally spend… all of that profit on training. - One thing I'm very interested in understanding is how you see the centralization of compute happening in the labs or the relative ratio of compute that goes to the world. versus goes to the labs. Where if you say right now, a third of marginal compute is going to the labs, by when is it-- over half. of the incremental compute in the world is going to the labs and by what point do the labs have basically a vast majority of the world's compute? Yeah, so earlier this year, you know, the beginning of this year, Anthropic OpenAI started at two for OpenAI and less than two for Anthropic. End of this year, they're both above five. So they have 3, 4x compute as a whole. Yeah. When you look at the incremental compute added, that's about 30% of the compute added this year. And as we step forward to next year, Given what's already been signed and penned and inked, you've got something even more dramatic, right? You've got Enthropic OpenAI are taking as much as 40% to 50% of compute next year. And this centralization doesn't look like it's slowing down or stopping. In fact, it looks like it's only accelerating. Now, who's building that compute for them will change. Next year, bigot new insurance, for example, SpaceX is building a ton of compute. And they're actively going to lease quite a bit of it to Anthropic and OpenAI, most likely, because they're the ones who have the marginal capability to pay the highest price. In addition, OpenAI and Anthropic are also starting to build their own compute. OpenAI with their own chips, Anthropic with TPUs that they're purchasing from Google and deploying with FluidStack. And so when you ask, hey, when does half of the world's incremental data?
[5:00] compute go to just open-air anthropic i mean it's it's really by the end of next year It's already half of the incremental compute is going to Anthropic and up to the high. Because computers growing so fast. incremental compute is going to be basically most of compute. So it's very soon, you're saying maybe within a year and a half or two years, and most of the world's compute is owned by two labs, or at least is serving the demand from two labs. How long do you think, so there's this trend where maybe world compute in gigawatts doubles every year, but the compute at the frontier laps triples every single year. But if you keep the current trend going, it goes from like two at the beginning of this year to close to like six at the end of this year, just multiplying out by three. 18 by the end of 2027, 54 by the end of 2028. Are you like, okay, at that point, they simply can't continue tripling given the amount of world compute? Or how do you see the world compute situation over the next few years? Yeah, so if the incremental compute adds this year 30 gigawatts, next year 50 gigawatts, and the year after that 70, roughly, you end up with this really interesting phenomenon, which is... okay, well, a new Watt deployed this year is significantly more efficient than the Watt's deployed two years ago. So actually, you know, a humongous percentage of the world's compute was deployed this year. Even though it didn't double the number of Watts deployed, I'm deploying GB300s and TPV7s and Tranium-3s, which are way, way, way more efficient, you know, 3x, 5x more performance per Watt than the prior generation chips. And so ultimately, you've got a huge ladder here. on 45% of compute next year. You've got them in, let's say, December 27. They have taken on half of the world's incremental new compute. But that half of the world's new incremental compute is actually at a higher performance than everything else before it. So you've got another multiplier on that. So by the time you're in towards the end of 2028, if this trend continues, which I see nothing that's stopping it. Um... You've got them just controlling most of the usable flops in the world. on their own. The thing I'm confused about is why you think we only add 80 gigawatts in 2028 if we enter in a world in which the price, the value of computing increases so much. That's the upper bound, by the way. That's the, that's the like, I'm so fucking bullish. Right. Okay. So let's, let's do some chain of thought here. So when I interviewed a few months ago, you said in order to make a gigawatt of, I think, Vera Rubens, you need, you need 55,000 N3 wafers 6k N5 wafers and 170k DRAM wafers I don't know if those numbers might have changed I'm going to troll you but the way you said wafers was so fucking you didn't pay for it by the way when we first moved to the US I had the VW thing pretty bad and I was a vegetarian vegetarian I remember you told me about this in North Dakota I was in elementary school and I'd be like can I get some wedgies can I get some wedgies Anyways, so that's where one gigawatt, right? - Yeah. - Now I, How did LLM run your wafer fab equipment model? and figure out how much tooling, how much the tooling costs to produce a gigawatt of compute basically every single year. And it was at like three to four billion dollars. Now suppose you add in clean rooms and Shell and everything else at the Fab. So six billion dollars of like Fab CapEx produces every single year. a gigawatt and a gigawatt produces right now, $100 billion of revenue. But also that 6 billion in CapEx is producing a gigawatt every single year. And that gigawatt is producing $100 billion every single year. So even over the course of five years, So, you know, So the first gigawatt is generated five years of profits. The second gigawatt that the fab has produced has generated four years of profits and so on. Six billion of capex at the fab level will have generated over a trillion dollars of end AI revenue. Yeah. There's a lot of OPEX along the way. There's a lot of other CAPEX, like the data center, the power. And you had to pay the OpenAI for the R&D. There's a lot of different people who need money here. But yeah, there's a huge... Take away half of it for all these middlemen. That so means there's 100x discrepancy between fab CAPEX and end revenue generated. More than that, actually, really, but we're just being very conservative. And as a result, this is capitalism, right? figure out like we're going to be you're you're you have this huge discrepancy where you can turn one dollar into a hundred dollars and And they're not going to figure out a way to make more mirrors. I mean, they are. Right. It's just these mirrors take some time to bake, right? But the emergency are so big, we're like... Anthropocan Open Area, we could make a trillion dollars right now, but we're just bottlenecked on the mirrors that go into the ASML machines. How can we make more mirrors if we spend $100 billion on this? That's the situation we're going to be in pretty soon. And I'm just like...
[10:00] Not going to be able to solve that supply constraint, that just seems quite hard to imagine. No, there's definitely – you've seen people do funny arbitrages here where they buy turbines and then they try and resell them. Right. Because the value of a turbine is way more because it's the thing bottlenecking a data center. I think if anyone had $400 million and the ability to convince ASML to sell them an EV tool, they should totally just go buy one and wait, wait, wait, and then sell it for north of – billion dollars, right? But ultimately, yes, capitalism will cause these things to expand, but it's a whip, right? It takes a long time for the whip signal to get to the tail end. Um... of that. And so the supply chain doesn't react immediately. In fact, You go to talk to someone at Carl's Ice, they're like, yeah, yeah, yeah. We need to make 100 EUV tools by the end of the decade. I think when we first had our – when we had our episode – earlier this year, They didn't even think they needed to make that many, enough mirrors to make 100 EUV tools a year. And so now they're like, OK, we need to do that. But in reality, because of all the economics of what's going on, it should be even more. But it takes so long to pill. Suppose every single company in the firm – sorry, in the stack got private equitied. Like somebody came in who was super HGI-pilled and was like, we're going to maximize production. How fast... What do you think the physical constraints on making more things would be? Because the reason I ask is we're pretty soon going to be in a world where... The lab revenue or just AI cash flows, because obviously the accelerators also have these huge cash flows, will be so big that you can just fund. extreme expansion of all this production from cash flows themselves. Yeah, I do agree. Generally, there's obviously some physical constraints. The way the supply chain is expanding currently, the hundred is roughly still the right number. For 2030? A hundred ASML tools for 2030. But, you know, if you said... Carl Zeiss, here's $10 billion. Please fucking just expand production. That would change things. And you would have to do this with every company in the supply chain. I don't think it'll happen this year. I don't think it'll happen next year. I don't think it'll happen the year after. Because the world is capital constrained. But in a world where, say, the top labs... are generating, let's say even combined, a trillion dollars in revenue next year. They're not able to say 10%? I don't think they're going to do that, but... Yeah, or hundreds of billions at least, right? It seems like they realize where the world is headed. I feel like they could just make... So the thing is the labs can spend hundreds of billions. They're going to generate hundreds of billions of revenue next year. But ultimately, CapEx next year is like $2 trillion. So you've got this big mismatch, right? The wafer fabrication equipment supply chain will do something on the order of $200 billion. The data center market supply chain will do even more. The accelerator supply chain will do even more. The energy supply chain will do a number. You sum all this up. It's going to be well north of $2 trillion of CapEx. So the labs have not yet... gotten to the point where their cash flows can fund this stuff. Of course, yeah. I mean, obviously they will never get to that point, right? Because they want to keep... Yeah, you reinvest. You want to make your capex higher than your returns. But... The key question I really want to understand is if the current content continues to be like north of 50 gigawatts, per lab by the end of 2028. So between them, they'd have 100 gigawatts. Those gigawatts, as you're saying, drive many-fold more throughput or more performance by 2028 than they are now, right? Because the hardware's gotten better. So not only have like flops per watt increase, but also the hardware gets better at working with AI workloads. Okay, so 100 gigawatts for the lab's end of 2028. How much is like world compute? I think that may be a little difficult given 2028 you start to have... They've taken 70, 80 percent of incremental compute, and I'm not sure what happens to markets then. How much does the price of compute skyrocket for them to actually be able to buy 70, 80 percent of compute? is Google or Meta or Amazon willing to sell even that much. Also, one caveat when we're sort of talking about these gigawatt numbers is when Amazon is serving bedrock anthropic models – that counts as Anthropic compute in sort of our worldview because it is effectively at the end of the day counted as revenue for Anthropic, even though like there's a revenue share and credit back all that. In 2028, if they get to 100 gigawatts combined, they have done... really disruptive things to the market because anyone can make money off of 10 to 15 million dollars per megawatt compute today. You literally, like I kid you not, it's not that hard. Go get a GB300 rack. Go download the Kimmy weights. Go download VLMRSGLang. Set it up. you know, Codex and Fable can actually help you do this. It's pretty simple. I mean, it's not like it's, It's not trivial, but it's not like rocket science. And go put it on open router. It's very simple. and you'll start generating more revenue than you're paying for the compute. And
[15:00] So this is sort of already led to this compute pricing $10 to $15 million per megawatt start to inflect up. And to get to that 100 gigawatts in 2028, you have to believe... that the labs can outpay for compute because anyone can make money at 10 to 15. You know, does compute now get to $25 million a megawatt? Does it get to $40 million a megawatt? But as you're saying, it's already the case that the labs are generating way more revenue per megawatt than everybody else. If they stay as far ahead as they are currently, you'd expect that to be the continue of the case. If there's like some kind of recursive self-improvement where the AI labs are like relatively uplifted or they have models internally, they're not releasing externally that are helping them make the next model better, you'd expect that to be even more the case. this or like SpaceX or whoever's like slightly further behind will just sell compute to the highest bidder if they can't internally monetize it as well as the labs so you I feel like it's continue expecting them to be able to gobble up like bid for larger and larger shares of the I think that is my worldview that they will continue to gobble up more of the compute but ultimately they can't do it at at current pricing or anywhere close to it for sure They do have to start paying 25, 30, 50 million dollars a megawatt to really gobble up 70 percent of the world's compute in 2028 to get to that 100 gigawatts by 2028, which is a very sort of aggressive goal. The other aspect of this that's really challenging is we've already seen a huge slowdown for the AI labs. This regulation that they advocate for is actually slowing down the labs a lot more than it slows down sort of the open source Chinese language models. OpenAI stopping training for two weeks, Anthropic not releasing what their safety assessment said is Model 2, which is widely believed to be the next version of Mythos. They're clearly not releasing their best models, in which case their revenue per megawatt stalls or even can start to decline again because other models are competitive again. So it's not that they're falling behind. It's just that they're not releasing their best stuff. What if there is some regulatory impact that prevents them from releasing their best models? Now, their revenue per megawatt does not climb as fast. Then their ability to buy that incremental compute for a higher price than everyone else starts to diminish. And then maybe they can't get to that 100 gigawatts. In a world where safety doesn't matter, I do believe that's exactly what happens, right? They can start generating $100 million per megawatt or more. And they can pay $50 million a megawatt. take everything off of my hands. But... There are, you know, forces at play that which we cannot describe that that would potentially slow this down. Yeah. Yeah. I mean, I think a good intuition pump is just what if the models were literally as good as a fully automated software engineer? They're not currently there yet. Right. Like I think they're far from just being able to fully automate the job of like a full white collar worker. But white collar workers earn six figures or north of that a year. And if you have a gigawatt that can sustain a population of like, say, a million of white collar workers, let's say roughly, right, that's like you could then off the back of that. That would be $100 billion. That's actually surprisingly low. Yeah, $100K per person, million population, yeah? Yeah, yeah, yeah. I don't know. But it would be many hundreds of billions of dollars if you get, like, full AGI per gigawatt. I think the other aspect of this is, and we've continued to see this, most of the value capture... is not happening, right? Like, most of the value that these models generate does not get given to OpenAI Anthropoc. Thankfully, so far, it is mostly just being given to the users. Yeah, yeah. Right? Jane Street, with their exclusive contract with OpenAI for GPT 5.6 Ultra Fast Mode, or Jane Street, where they're like... one of Anthropic's biggest customers is generating way, way, way, way more value out of the tokens they're paying for than Anthropic is generating in terms of profit, right? Because they get to, you know, Yeah. than, than anthropic. And so the ultimately, you know, and that's, that's what's required. So sure. If you had a million new software engineers, you know, the cost for software engineer would also fall. One thing I'm confused about is does the market come in equilibrium? And if it comes in equilibrium, would you just expect the price of compute to equal whatever Anthropoc and OpenAI can generate from it, or be very close to it, with like a small amount of markup for Anthropoc and OpenAI. Like right now it's really weird that there is a 4x or more difference between what compute sells for and how much money Anthropoc can make from it. And in a world where the revenue per gigawatt continues to increase, if Anthropic's ability to monetize a gigawatt
[20:00] doubles or triples or something. It'd be weird if then the gap continued to increase. And so Anthropic, just by having some software, having some weights, can take something that costs them $10 and then turn it into $100. Yeah, so there's a bit of a – it's always a fun question, right, which is where does the value go in AI? AI is generating all this value. The end user, which I think we all agree is generating more value than anyone else, hence they're paying a lot for these models. But then you have the app layer. Well, so far, the app layer has generated very little value. And you've got the model layer, which, again, up until a year ago was generating negative gross margins and is now generating massive positive gross margins. And it looks like it's on the path to generating $100 million per megawatt, so turning $10, $15 into $100, as you said. But if we go back again a year ago, the hardware supply chain was generating all this gross margin while literally everyone else was losing money on it. Open Ananthropic were just plowing VC money in. And as were many other startups and many of these hyperscalers are building infrastructure without knowing if there was going to be a payoff. So ultimately, you had this negative value being created on the model layer almost, if you will, because they were selling the tokens for less than it cost them on the infra side. initially in 2023, the memory guys were making no money. off of, you know, HBM or memory for AI, even though theoretically their value they were delivering was humongous. Now you've got, well, actually, QSMC makes way less value than the memory guys. Is that actually how much, you know, they're capturing less value, you know? So the value captors shifted around a lot, which is very fun for people tracking the market or participating in the market, like Jane Street as an example. There's a lot of, you don't know how to plug them that hard. Yeah. So, What happens going forward? Does Anthropic and OpenAI, they've slowly started to balloon in value capture. Do they balloon and take all the value capture? Well, that was a thought. And then Elon showed, actually, no, I can sell my compute for $25 million a megawatt or $40 million a megawatt to Anthropic and Google. Even if it's a short-term thing, I've sold it for this price, and I'll recoup my entire CapEx in a year. or whatever that sold for 40 B a gigawatt. The space I sold for 40 B a gigawatt to Google. What does that sell for at the end of next year? I think most compute will still continue to transact at sub $20 billion a gigawatt. Even at the end of next year? Because all of it has to be financed. For compute that you can build without financing, right? If Meta can build compute, Microsoft, Amazon, SpaceX can build compute without finding a customer just saying, fuck it, I'm going to build this compute. and then turn around and wait until it's already built, they now control what's going on. So most compute is contracted well before it's built. Yeah, yeah, yeah. And so this is sort of what Elon took advantage of in the market is, he actually had all this compute, and he was like, hey, Anthropic, I know you're making like $60 plus billion per gigalot, why don't you just buy my stuff for a crazy amount of money? And obviously the, you know, it's not like Elon decided this or Anthropic decided this sort of markets figured itself out. Um, Other people, you know, you go to a random cloud, They're like, okay, I'm going to build a gigawatt of compute. or 100 megawatts at compute, I'm going to spend the CapEx. I need to turn around and find a customer. If I want to find a customer, I need to find the capital. Who's going to give me the capital on the customer? The customer has to sign a deal. And then I take the customer's commitment to the credit markets and I raise the capital. And so there's this sort of like completely different power structure where Meta, who is effectively hoarding compute them in spacex are plausibly the like number three and the only plausible number three is because they're hoarding all this compute they're using their balance sheets and capabilities to build compute to build compute without an customer that's monetizing at a huge degree and they have an actual balance sheet so they can go to the credit market and being like. Hey, guys. You build a gig a lot. You can make your margin, not a crazy margin, but you can make a good margin, and I now have all this compute. And now Meta and SpaceX have this optionality of looking around and being like... is my internal use case going to make me more money, or should I go out there and sell it to Anthropic OpenAI at crazy margins? Yeah. So now we've sort of entered a regime where SpaceX and Meta are saying, actually, I'm going to build the compute... And I can start to rent it out for not 13. I can sell it for 25, 50, and more. So as I've been doing video essays and other formats, I've been looking to hire a new editor for the podcast. But actively searching for editors has been quite time-consuming because the vast majority of candidates don't fit the profile that I'm looking for. So I created a recruiter in Grokbot to see if it would help. I give it a huge context dump where I monologue basically everything that I wanted, and then it spun out four other bots to narrow in on different parts of the search. One went through the last year of my email for Relevant Inbound. One searched my Xfeed and DMs.
[25:00] One went through the end credits on various documentaries I liked, and the last one looked for editors who work for some of the YouTubers that I follow. Grokbot then took all these different candidates that these sub-agents had found, it filtered them against my criteria, and then delivered for me a final short list to review. To be honest, the first batch had a few good candidates I wanted to see, but mainly a bunch of duds. But after I gave Grokbot some more feedback about what it was missing, it came back with a new list of candidates that I'm actually extremely excited about. inbound email and ex-DMs for promising new candidates to potentially interview. If you want to try GrokBot yourself, go to x.ai.com. What do you think their revenue per gigawatt is by the end of 2027? Like for Anthropical OpenAI by the end of 2027? I think it's highly dependent on who has the best model, if they're allowed to keep releasing their best models. But I don't see why it wouldn't be 50 plus million dollars a megawatt. By the end of 2027. Oh, by the end of 2027. Yeah. That's where it gets more challenging. But I think it could get to higher than that. It's like 70, 80 million dollars a megawatt blended across the company. If not higher. Yeah. And so I think, If that's the case, right? Then what happens to the price of compute? Well, if I'm anthropic, incremental compute is worth it. Maybe I spend $40 million a megawatt on SpaceX compute. And if I'm SpaceX, you know, I look to the supply chain, I'm like, well, you know, I've struck this deal with Jensen where he's now all of a sudden using Twitter. And, you know, Elon's saying they're exclusive to NVIDIA, but why doesn't Jensen raise his prices? And then, you know, And SK Hynix and Micron and Samsung looked at Nvidia and were like, well, why don't they raise their price? So I think the value capture, there's a bullwhip effect here, right, where just because someone has risen the prices doesn't mean the entire supply chain rebalances immediately. But over time, the supply chain will rebalance and things will cost more and more. And, you know, to get that incremental capacity, you sort of have to, right? So TSMC raising prices very slowly, but memory companies raising prices very quickly. You know, substrate companies raising prices very quickly. Elon wouldn't have sold if it was $15,000. but he's selling because it's 25 plus. So obviously he rose his prices really quickly. Yeah. I'm sort of surprised you think revenue per gigawatt doesn't increase way more than even like 100 per gigawatt by the end of next year. When does RSI happen? When does it take off, right? Even if RSI doesn't happen, the current rate of progress continues. If you just look at how much progress have you made in, let's say, the last... A year and a half. What was the model from a year and a half ago? My problem with this is the best model that exists in the world was trained in February. Okay, you're saying maybe we just won't be allowed to release the best models. And OpenAI says they're not training models for two weeks, man. What the hell? Yeah, yeah, yeah. I mean, there's one thing, like, internally, are they getting enough use for it? It's just a little bit of the price of compute. Or in other words, like, does AI progress as a whole slowdown because of regulation? Yeah, but they're not even allowed to use this, like, new model. Like, Astro's not widely deployed internally, man. Right, right, right, yeah, yeah. But still, I don't know. If you have, like, a model that is, what was the model released, like, let's say, the beginning of last year? Like GPT... 4-0? Is that 4.0? Yeah. That's like you're talking about a 4.0 to Fable size or Mythos 2 size leap by this point. Again, by the end of 2027. Yeah, but Mythos 2 is not out. Or like even Mythos, right? Like that leap again. Even Mythos is not allowed to be out, right? They've neutered it. Yeah, yeah, yeah. Like we can't use it to optimize inference performance. We can't use it to optimize all sorts of things. Right. Yeah, maybe there's like some slowdown in AI progress or the deployment of AI that means that the revenue per gigawatt can be lower. only 100 per megawatt. by the end of next year. Yeah, I mean, as long as the model gets better, the value generated out of it gets better. Obviously, who captures the value is still up for debate, but ultimately, everyone's going to raise their prices. Yeah, yeah, yeah. Because they can. Yeah. And it's super inflationary, especially if the method of regulation is right now so far, it's just don't release the models. But more and more, the method of regulation is New York's banning data centers. Texas is holding memoratoriums. Ohio saying you have to or at least trying to say you have to like pay everyone's property tax in a certain radius. These sorts of things are going to decrease supply and increase cost. And that's going to get passed on as well. So you start to end up in a spot where progress does slow, at least in the external sense. keep getting better and better. I see no reason why, like, you know, again, like in a takeoff scenario, why would Anthropic not have their best model six months ahead of what is externally available because of safety and regulation, but also... The competitive advantage. And then that six month difference, if progress accelerates, is actually a bigger differential. So that's the thing that would cap revenue per megawatt gains to a much lower growth than we've seen in the first half this year. Here's something I'm very interested in. as these companies go public and they're accountable to investors. And let's say end of next year they have I don't know, close to 20 gigawatts. So like 10% of the compute, 2 gigawatts. Let's say they want to go from 60% compute to training to 70% compute to training.
[30:00] And their investors are like, well, if you're able to generate $100 billion per gigawatt, you're basically saying no to like $200 billion of revenue in order to increase your training compute. As investors are like, what the fuck? You're already spending so much on training. Why are you spending even more on training? As a public company, do you think... Yeah, what do you think would happen if they're just like, no, we will keep increasing the share of compute we spent on training to offset the increase in revenue that each gigawatt of compute is giving us? Yeah, so this is sort of what I personally believe that the labs are going to allocate less and less compute to inference over time, which I think is very non-consensus, right? Everyone sort of… The standard belief of most people is, oh, most compute will go to inference. Most of it will go to forward passes. for training, not maybe necessarily revenue-generating inference. But ultimately, you end up with... if they're generating, you know, 30, $40 million per megawatt today, you allocate 40% to inference. If you now get to generating 60, $70 million per megawatt, Do you... still allocate 40% to inference and generate all this profit, And then Do dividends and share buybacks? Or do you go build AGI? And I think the... obvious answer from Anthropoc and OpenAI, and not just at the executive level, but also their board is go build AGI because it's way more profitable. Um, And so ultimately... you're going to see them ratchet up their percentage of compute dedicated to training. While each increment of compute is getting more and more profit generating, if they had dedicated to inference. Right, and so the whole point is, well, okay, if I'm selling tokens... is OpenAI releasing ultra-fast mode for just external, or are they doing it internally too? And it turns out, no, actually, I'm going to allocate it to internal and external because my internal... Value that I'm generating from super fast AI or the best AI model is way more than what someone externally is. And so ultimately, sure, I could generate $100 million per megawatt. But if I turn that towards AI research, what is the incremental progress that I get? And then what does that do towards my future earnings potential, the discounted cash flows of whatever the hell I've done, right? Yeah. And so, you know, they're not going through that calculation, but ultimately it's It makes more sense to dedicate more and more compute internally. And the only reason to have inference compute be so large is so you can grow your training fleet. Right, right, right. I think this is an interesting economics question that I feel like we can have the models digest. What would have to be true about a world where they reduce fraction of compute spend on inference? I think they have been over the last three months already. Interesting. Interesting. I think... parts of this year, they were increasing fraction of compute. So let's just take month by month. you would agree that every month, Anthropic has added more compute than the prior month. There might be some noise when they sign a SpaceX deal or whatever, but in general... the amount of compute, is a curve up. And so in January, they added less compute than December. And yet their revenue ads skyrocketed, and then they've sort of plateaued. They're only adding – they're not adding $25 billion ARR every month now. And so that means the marginal megawatt they're getting is going higher percentage to R&D than is inference. Interesting, yeah. And so they are factually increasing their compute towards R&D today. Yeah, yeah, yeah. Yeah, I think this is like... self-evident um if you if you like look at what they're doing enough yeah So if I look at the numbers you said of how fast world compute grows, here's some things I want to understand. So it seems like if I added the numbers you just said, it would be over 200 gigawatts of world compute by the end of 2020, right? Yeah, globally. Okay. And how fast can that continue growing? Like, globally, I compute after 2028. Yeah, so 30 this year, 50 next year, 70 and 28. 29 should be, like, on the order of 90 to 100. Yeah. - Like then just 100 more every single year or something. - I think the slope can continue to go upwards, and it's hard to predict anything more than four years out, given who knows, are we in RSI regime, or when is the world economy growing at 10% a year? 'Cause if you're at 100 plus gigawatts a year, you're at absurd GDP growth. - Right, if you think there's 200 gigawatts globally, in 2028, how much is in China by that point. And how does Chinese compute continue increasing through this whole trend? Because if If the RSI stuff kicks off in the West before China has a large amount of compute, maybe we're living in a different world than when it doesn't. Yeah, so China today... So if we sort of level set back to 2022... The US was adding about 45 to 50% of the world's compute. China was adding about 30 to 35% of the world's compute and the rest being taken up by the rest of the world. Since 2022, we've had big regulations against China and a dramatic increase in America.
[35:00] of watts are being deployed in America. And China is really a very small number. It's sub 10% of watts being deployed for data center AI compute is in China. And as we step forward, they're still at a very small number. Their domestic production is quite small. They're purchasing from NVIDIA. is still quite small. And a lot of that ends up in other places as well, right? You know, Malaysia or what have you. So ultimately, China domestically still continues to have sub-10% of incremental new compute. So in 2028, it might start to inflect up, I think. But it's pretty easy to say China will have like 30 gigawatts of AI compute. or less by 2028 yeah in 2028 okay and then how fast does their hockey stick go up I do think in 2028, they have a big uplift in what compute they're able to deploy. 2026, they're still mostly relying on a lot of the smuggled chips, you know, a lot of the chips that TSMC made for companies that they thought weren't Huawei but ended up being Huawei or a lot of HBM that Samsung is shipping, you know, sort of. But in 27, fabs start to go up. In 28, especially, fabs start to go up from SMIC and CXMT and such, where domestic production Yeah. Nvidia will have in 28 or Google will have in 28 or opening. I will have in 2028 so even even the gigawatt number overstates things you're saying It's like 30 gigawatts, but it's really much worse chips But then how yeah, how does it if you think the world is gonna add 100 gigawatts the following year? I? I know you said you can't really say that far out. How much is China able to have the subsequent year? Basically, I want to know, did they just hockey stick at the point at which they are able to start shipping large amounts of compute? Or is it still going to be less than U.S. plus allies? be able to domestic produce domestically. Um, But ultimately, you know, China is definitely going to hockey stick if there's anything China's really good at is scaling manufacturing really, really quickly. And I imagine China's China will start to be able to extract more and more purchasing of even foreign ships into domestic China or at least close the gap in what the U.S. is allowing NVIDIA to sell them or what have you. Do you think China could do... Adding 50 gigawatts by 2029. marginal incremental gigawatts in 2029. I think that's, I think that's completely reasonable. Yeah. Um, And part of that could also be purchased from foreign. But yeah, I think it's completely reasonable that China in 2029 can do 50 gigs. But if most of those are domestic chips, there is some factor there where that 50 gigawatts is really worth as much as 20 gigawatts in America or from American chips. Right, right, right. So yeah, you're actually projecting a world where maybe the leading lab in 2028... has more compute than China will have in like All of China will have in 29 or even 30. If you weighed gigawatts by their quality. It's implying that there's nothing done to slow down the U.S. labs. Yeah, that's right. Clearly, the government is starting – and politicians are starting to do that. Yeah, yeah, yeah. Whereas China is not going to slow down AI. In fact, the only thing they're going to do is accelerate it. So honestly, I – When Andrew Jensen asked about expert controls, I am a libertarian person. I'm like, I wasn't like genuinely sure what I thought about this issue. I was steel manning. what is like the opposite view that he has, because I think it's important to hash out ideas. But I'm like, yeah, maybe there's a world where if we just cooperated with China, it would be better for us, especially since they control so much of the supply chain and the other things that we needed for robotics and other things. But I didn't realize the compute situation was as fucked as you're saying. Like actually, the export controls do seem to have like really, if they ship the amount that you're saying, that's a huge difference. compute stock. And so if that ends up being the case, that would have worked. I think that's actually a notable success. I would say the only caveat there is some of it is export controls, but some of it is also just... uh, financial systems, right? American financial systems are more willing to YOLO into startups than Chinese financial systems. But once Chinese financial systems choose an industry to focus on, they'll subsidize it a hell of a lot more. Um, and so the Chinese semiconductor industry has significantly more subsidies than the rest of the world's semiconductor industries combined. Um, which, which points to like, you know, if takeoff is not as fast as sort of you're implying, but actually it takes longer than ultimately China,
[40:00] will catch up drastically on the semiconductor side, which then is compute at some point. The other aspect of this that I would think is like, I think is like noteworthy is Chinese companies today are not that far behind in AI models, at least percept perceivably by the public relative to the amount of compute they have. Right. The leading Chinese labs have. 100, 200 megawatts total of compute at most. ByteDance Seed being the one outlier where they have significantly more than that. But Kimi is not running a gigawatt or anywhere close to it. Whereas Anthropic is nearly five gigawatts by the end of the year, right? Or more, sorry. And so the question is sort of, well, does it matter? And I think right now it doesn't matter that much this difference in compute because, you know, when we break down the compute ratio or budget of a lab Historically, it's So far it's been like 60% training, 40% inference, but that training gets broken down further. And it's actually like 50% of the compute is research data. Like 10% of the compute is development, and then 40% is inference. And what I mean by research and development is researchers are generating ideas, testing new architectures, testing new data mixes, testing new hyperparameters, whatever it is they're doing, new attention techniques, blah, blah, blah. But ultimately, when they do the training run, when Anthropic trains Mythos, it's sub 200 megawatts. The pre-training or the whole thing? The pre-training. It's sub 200 megawatts for call it two months. And then the RL is even less. But you think the RL was less to compute than the pre-train? At least in terms of single-side inference. I mean, single-side of your training app. But Total Computer was probably higher, right? But it's like sequential, right? So at most, the most they ever used at one point in time was maybe 200 megawatts. And then in reality, they had multiple gigawatts. So most of their compute was going to the research, not the development of a model. And there's reasons for this, right? It's hard to coordinate all these clusters. It's hard to co-locate all of them. It's hard to do multi-site training. It's hard to do RL. Generating even more rollouts during RL does not necessarily make it better. all two gigawatts that you have for training onto training. Actually, I can only leverage 200 megawatts. As we get further and further down, implement automated coding, automated researcher, I actually... expect the percentage of the compute budget that goes to research versus training to start to and to start to like become a lot more fuzzy or even higher for training. Also things like continual learning, right? All of these things start to mean that more and more is actually going to training the model. If you end up in a world where you're doing 100 gigawatts a year, at current prices, that would be 5 trillion of CapEx every single year. And then stack on the fact that you have to build the power plants way before then slash it's a 30-year asset. You stack on the fact that the data centers are a 15, 20-year asset, and you have to build that then too. So the $5 trillion, you have to account for future years growth, so it's actually going to be more like $7 or $10 trillion. Huh. I understand because you're not including the fact that like that doesn't include the fact that there's not the infrastructure for the power generation or whatever in the data center itself. Right, exactly. And the data center itself... is... So when you talk about AI CapEx, people are saying $40, $50 billion, but that's really just the critical IT, right? The servers, the networking, the fiber, the transceivers, optical communications, all this sort of stuff. It doesn't account for the data center itself. Or the power plants themselves, which are being built ahead of time. Yeah, yeah, yeah. If I'm building 100 gigawatts this year and 150 gigawatts next year, well, then all of the buildings for that 150 gigawatts need to be built in CapEx this year. And if I'm building 200 gigawatts the year after that, all those power plants need to be spent. You have to buy the turbines this year. Yeah. Right. And so you've got this like this like actually it's much bigger than even five trillion dollars if you're building 100 gigawatts. Right. So very plausibly. incremental capex every year is - Getting close to $10 trillion. By the end of the decade. Right, which is going to be close to a tenth of the world economy. And a third of it. If all of it's going up in the U.S., it's like, well, the U.S. economy will have grown as well. But still, the current size of the U.S. economy will be like a third to a quarter of the U.S. economy. would just be going towards data centers. And as I say that out loud, I'm like, maybe you're right, and we just won't allow it. And that's the reason this doesn't happen, right? Because for this exponential continue, just like a quarter of America's economy is just building data centers. Yeah, I mean, I believe in capitalism and reallocation of resources towards the most profitable thing, but at the same time, politics exist and credit markets exist and capital markets exist. So,
[45:00] to enable let's say that 100 gigawatts by 2030, or let's even like, let's even like pair it down to 2028, where it's like three or $4 trillion of CapEx across all of these items. You know, a couple, you know, over, you know, two and a half towards IT CapEx, and then another one to two on data center and energy, and all the supply chain downstream, like semiconductors and all that stuff. So if you're at, If you're at $3 or $4 trillion of CapEx, Where does all this cash come from? No one is generating that much cash from the business yet. Right? So hyperscalers, they funded all of the growth up until now, Google, Microsoft, Amazon, Meta. They funded a huge percentage of it. They were more than half of compute. But they now... Don't generate cash. They actually spend everything on CapEx. And in addition, they raise debt and spend everything on CapEx, right? You've seen Meta do it, even Amazon, even Google, um... Microsoft will be there soon. Everyone is raising debt to pay for their CapEx. So now who is the incremental person to pay for this that was not doing it before? In the case of Google, it was pretty simple for them to stop doing buybacks. or Meta stopped doing buybacks and turn around and buy computer infrastructure. And that doesn't have a huge effect on the market, but it does have some effect. But as you step forward to 2028, where the hyperscalers are now raising hundreds of billions of dollars of debt, and then all of their supply chain is raising hundreds of billions of dollars of debt, who pays for this? And so there's a few different ways. You know, there's some semiconductor companies like NVIDIA and Broadcom and the memory companies turning around and deciding to fund some of this CapEx. There's... the traditional infrastructure investors who are turning around and gathering capital and investing in infrastructure and instead of bridges its data centers. And then lastly, there's everyone in the economy who's realizing – Maybe I shouldn't buy a home. Or maybe I shouldn't invest in credit for a home that's helping people buy homes, or maybe I shouldn't buy government debt. I should just buy hyperscaler debt, or I should just buy this data center's debt, or I should buy Anthropik's debt because Anthropik's willing to pay 20% you know, rates for, you know, the incremental billion dollars to build their capacity because they know their revenue from it's going to be huge. And they're going to pay 20 percent because it's still better than renting it from SpaceX for $50 billion a gigawatt. So you've got all of this contention. But if you now do this, the whole world economy is like really shifted around. Antithesis is a deterministic software testing platform that enables perfect reproducibility. It also unlocks some pretty insane approaches to debugging, like time travel. With Antithesis, you can jump to any point in the trajectory and start from there. So when there's a crash, you can rewind to the exact moment that something went wrong and freeze the entire system. the application, the database, even the environment itself. This lets you do something that would otherwise be impossible, which is to observe every part of a distributed system at the exact same instant. Time travel also allows you to add telemetry and logging to an event that has already happened. For example, you can rewind to five seconds before a crash and decide to capture all the network traffic. Most powerfully, Antithesis gives you a live terminal into your system that hit play and see what happens. Then go back and try something else. In production, you often only get one shot on goal with this sort of destructive analysis. If you restart a deadlock service, for example, the exact deadlock you needed to study disappears. But with antithesis, the original timeline is always reproducible, so you can test as many hypotheses as you need. And if you don't want to do all this time traveling yourself, you can just have your agents do it for you via the antithesis API. Go to antithesis.com slash the work cache to learn more. So you and I have been debating off air for the last few days whether there will be a sovereign debt crisis as a result of AI. And the logic is this. AI... is you have a situation where, as we were mentioning, very little investment. turns into a lot of money right so the rate of return what a fucking problem dude can't believe it no it is a huge problem for everybody else who can't turn a little money into a lot of money right um so the rate of return is incredibly high even at the data center level you know if you're built if you like build a data center and you're like trying to get rented out to anthropic and urban ai for like 10x what it costs you on a depreciated basis to build it it's And so you turn $1 into like $2 or $10 or something at the end of the year. That raises the rate of interest higher. Now, if the rate of interest goes higher, And if it does that for the entire economy and people are borrowing more and more money, they're competing against the other lending that the government would have done or the other companies would have done or that you as a consumer or a mortgage buyer would have done. Then that's just making it basically more expensive for everybody else to borrow.
[50:00] There's huge implications. for tons and tons of people. Sorry, I'm going to go on a bit of a monologue here. But we've been thinking about this together. So I think the US will be fine at the end of the day because they can if the data centers are built in America, you can fundamentally just like tax the data centers. But the way the current tax system is set up, you know, corporate income is like less than 10 percent of federal revenues and 80 percent plus is payroll taxes and income taxes, which as more and more automation happens will shrink. At the same time, on the spending side, currently 20% of tax revenue spending goes towards servicing the debt, basically, paying interest payments on the debt. Now, a lot of the debt is short duration, so it refurbishes every five years it rolls over. Why are you fucking laughing? Because, you know, it's like things you've learned in the last month. Yeah, like it's any different for you. Like you got a degree in fucking financial economics. the internet thinks i'm a beekeeper a few months few months this is our business dylan i know i know i'm sorry sorry um and uh so i'm self-conscious fuck no it's good you're doing good i just think it's funny A million people! Listen to this guy who just learned about death this month! The amount of the fraction of tax revenue that goes towards servicing the debt basically goes from 20 percent to 25 percent. If there are five percentages, that would go towards like north of 40 percent. But if you take into account the fact the government is borrowing two trillion dollars every single year, then that goes from like 40 percent to like north of 60 percent. So 60 percent of tax revenue basically just goes towards paying interest payments on the debt. Now, I think the U.S. is going to be fine because also the tax base will increase if we let data centers get built in America. Other countries are absolutely fucked in my opinion. I was just looking at which countries have a lot of debt have very little tax revenue, and also a lot of their debt is serviced quite often. And those countries, like Pakistan or Nigeria or something, I think are just going to be very fucked in this new interest rate regime. So this crowding out effect is actually the thing that I've... is the reason it's not like YOLO 1 billion gigawatts. Yeah, yeah. Right? You've got... You've got all these industries and countries that use a lot of debt, whether it's, you know, all these impoverished countries that you mentioned earlier that are just going to default. You've got like consumer packaged goods, right? Like all of these like companies that make things you see at Trader Joe's or wherever. Use a lot of debt. All these telecom companies use a lot of debt. And banks. use a lot of debt. And so if interest rates go up, So in the market, not necessarily the government set interest rate, but in the market, the spread of interest rate between what the government says their federal rate is versus what everyone else is charging because Amazon wants to raise $100 billion of debt next year or whatever the hell the number is – Probably less, but you end up with this really challenging problem of where does the cash come from? There is some level that is funded by cash flows, and cash flows keep going up. But the logical thing to do is to invest way more than your cash flows because then the returns in the future years will be amazing. So you have this delta. And then what's pushing down on the delta is all of these other things, right? There's regulations against data centers, regulations – consumers getting mad, politicians getting mad, regulations against AI. The AI lab is not releasing their latest models because of safety reasons. All of these things – and interest rates going up are an influence on all of these things. So all of these things bend the curve from what does capitalism want? in terms of just pure, simple economics, to what is the complex system that we have want, and bends it lower and lower and lower to where not as many gigawatts as should be built. will be built. Well, the interest rate is part of capitalism, right? Yeah, but like, you know, like in the simple economic model versus like the more complex what we have. What is the rate at which you think Amazon or Anthropica or whatever will be releasing bonds? for debt next year? They do hundreds of billions of dollars in debt. What is the rate? What is the average rate? I don't think Amazon will do hundreds of billions of dollars of debt. Total. Let's say the... The hyperscalers in total will raise – and all the clouds. Yeah, yeah. In the modeling that we do, we have about $11 trillion of CapEx from 2024 to 2029. - Total. - Total. And if you do, you know, if you fund a lot of this with cash flows and as much as you can, you still end up with north of five trillion dollars of credit. that need to be issued for this $11 trillion plus buildup. You don't think the AI revenue continues even 3xing year over year? AI revenue does go up
[55:00] think it can go up forever. I don't, you know, like just like without like certain constraints being hit, I think certain labs will have certain incentives. Um, And labs are not the ones building all the compute in many cases, even though they're increasingly trying to go that way. They'll have all these cash flows. Like if their revenue keeps increasing, whatever, that's fine. But how much did you say their revenue will be? You think they'll not have that much revenue? No, I'm just saying until 2029, there's something on the order of $11 trillion of CapEx, and six of that is funded with cash. And five of that is funded with debt. And if that's the case, $5 trillion of debt being raised across the whole ecosystem does make interest rates go up. And then what prevents that? You know, there's a couple of things. One, do labs increase their revenue per megawatt more and keep inference allocations large, in which case they're... taking all this profit, they're accumulating all the profit across the S&P 500 because everyone's paying to, you know, reduce their costs. Of course, their profits will also go up, but, you know, cash has to come from somewhere. Um... So there's an upper limit on how fast their revenue can grow versus the value they deliver into the world. And there's a diffusion aspect of the technology. But ultimately, labs revenue keep going up. They can't cash flow fund everything. The optimal scenario is you actually use credit as much as you can to fund because even if cash flows from the labs fund a lot of stuff, you want to build more than that. And so there is some amount of credit that gets built. Our current modeling has $5 trillion of credit and $6 trillion of cash-funded infrastructure investments through 29. And when you take that – You've sort of got... this is not enough compute relative to what this demand growth is from the AI models. And so you've got the obvious answer, which is revenue for megawatt keeps growing up. Yeah, that makes sense. How much do you think interest rates will increase by 2029 as a result of all this? dude you know it's just vibing a number but if you're vibing a number out you know Growth in the world and economy is growing up a lot, so why wouldn't interest rates for Amazon – go from, you know, from where they are today. I think meta... Okay, let's take... So this is going to be extremely vibed out, but recently Meta's raised at like 5% to 6%. I don't see why they wouldn't pay 8%. Because they would happily pay 8% because the return from the compute that they're going to build is humongous. Right. And the market won't want them to. But if they want to pay 8%, the flip side is if they pay 8%. Versus the five they do, five and a half, six they do today, you know, 250 bps increase. That makes everyone else in the economy also pay 250 BEPS more, which then – causes a lot of things, right? Banks will scream because if there's credit spread goes up, then their assets don't repri... their debt themselves reprices faster than their assets reprice. And you ultimately end up with... they're losing tons of money if their credit spread blows up. The other consequences of this are... this is a point you made, but if interest rates rise... the discount rate increases, which means that the discount cash flows of all equities increase. Yeah. Crater, which means that even though the stock market as a whole might be doing fine, like S&P 500 will be fine. Yeah. Any individual stock will probably have just like cratered in value, especially the the Buffett like Berkshire type, you know, pay good cash flows for 30 year type. Like why the fuck would I invest that much if my discount rate isn't 3%? Or 5%, it's now 8% or 10%. And for developing countries, Basil Hopper, who's a good friend and he's an economist, he made this point that we'll see a second Volcker shock. So in the 80s, to fight inflation, FAC chair Paul Worker raised interest rates like more than 5%, or it's like something like 8%, real interest rates 8%. And that caused some 40 different countries, mostly in Latin America, to default in that decade. And I think that will probably happen again. In fact, okay, now we're getting into like singularity talk. So we've been talking about what happens if interest rates... I think this all happens before singularity. Yeah, that's what I'm saying. That's what I'm saying. That's what I'm saying. So we were talking about before singularity, interest rates rise 2%, 3%, etc. At some point, I think it's very likely that the world economy will be doubling every single year. Okay, this is not happening in five years, but it'll happen eventually. It's just like there will be – there's a researcher, Damon Binder, who's done great work on this. But basically, if you look at input-output tables in a fully automated economy, Yeah, if the economy grows at 3% a year, then it's like, you know, rule of 70, it's like 20-something years. Right, but he was like, "Okay, well, right now we're bottlenecked by the fact that there's people, and you can't, like,
[60:00] people every single year. But in a world where you can also double labor force every single year, how fast can the economy grow? And I think it could double every single year. At the very least, it would be like tens of percent every single year. Okay. The rate of interest should be pretty close to the growth rate. It won't be exactly that because of consumption, but it should be pretty similar. So then we'll go into a world, I think in the 2030s. where the rate of interest is like tens of percent. um and like i don't know part of my brain is like it might be hundreds of percent but like okay let's say it's at least tens of percent i'm just like okay every country that is not involved in the production of ai defaults um every stock that is not an ai stock is like worth basically zero because discounted cash flows are worth nothing if the federal government can't figure out a way to tax ai you know the the servicing the debt is more than the current tax revenue. All these other effects that I'm sure we're not even pricing in, like you can't get a mortgage, et cetera, et cetera. Because fundamentally, what is happening in this world? This is all nerd speak, right? But let's step back. What's happening? Just now it started the nerd speak. We'd be entering a regime where we're in a totally different growth regime, basically. And the economy is basically saying, hey, the government borrowing money to pay people pensions The opportunity cost of that is extremely high now because that money could be spent building a robot factory that builds a robot factory that builds a robot factory. And so the opportunity cost of capital is going to increase significantly. a ton and that just like that's fundamentally what the cause of all of these things we're talking about Yeah, so as interest rates go up, equity markets get pummeled. Yeah. And even AI companies, right? People are like, you know, some people who really believe in AI are like, why does Micron or Hynix or Kyoxia trade at two or three times earnings? And it's like, Well, if you're really AI-pilled, everything in the economy should trade at like two or three times earnings. And if you're not AI-pilled, then sure, they're over-earning. So it's sort of like an argument for why I think memory is going to do great, but memory stocks shouldn't 10x or whatever again because – If we're in the market where there's that much demand for memory, which means AI causes drastic change in the economy – then everything should trade at like two or three X multiples and the stock market should fucking crash. Right. Right. And so in a sense, like meta trading, I don't know. I think meta trades at like something, they're like $1.5 trillion company. It's like, what? Silly? They're worth way more than that, at least in a logical sense. You just look at their cash flows. And like all the infrastructure they're hoarding and all the compute that they're going to be able to sell for crazy amounts of dollars per watt, either as tokens because their lab works or just Anthropik and OpenAI. Yeah. Ultimately becomes a question of like, You have to reallocate all the capital to the AGI. And you do that by pricing everyone else out. And so the limiter on AGI is not how fast can the research engineers, like our roommate Shoto, can crank the gears. It's actually just like... How much does the rest of the world let that happen? Right. Because they're going to regulate. They're going to obviously increase interest rates. They're going to say no data centers. They're going to say stop building fabs. They're going to say, oh, shit, every company's equity value is tanking. So how can I pay for AI? you know, To increase my business, well, then, you know, like, okay, then Anthropoc and OpenAid have to start, like, building their own stuff. And obviously, they're going to eventually focus on, you know, they're building their own chips already, or at least designing their own chips, and it'll expand out their, you know, their contracting their own data centers and building their own infra in the next couple years. So there's sort of like how does this reallocation of the economy happen? But there's a lot of downward pressure on it not being just straight takeoff. even if the models were capable of it. which I think you and I believe we're in a world where models are capable of that. But... Slow takeoff is, at least my hope, possible because everything in the economy and regulatory world, like governments saying don't release your models, governments saying actually you can't even use your models internally that much because that's going to happen soon. They're already saying you can't release your models. Which is actually... The thing I'm most worried about is, you know, a singularity. Which external deployment is actually helping, right? So the fact that we're preventing external deployment... Well, does that prevent singularity? I mean, right now it leads to more revenue because the models aren't capable of ours. Yeah, yeah. But I'm worried about a world where it's 2030 and the government's like, we're going to wait six months before you can release your model to the public. Six months, 100x, let's go! Yeah, and that six months, they do recursive self-reformment internally. They just have... all kinds of crazy shit happening in the company. Meanwhile, the rest of us are stuck with models that are like, at current pace, years behind. - Yeah. - So here's my thought. Okay, suppose, that the whole world gets in on this conspiracy to try to slow down AI. I don't think it's a conspiracy. It's outwardly written from every politician. Suppose they basically prevent an entire... they slow down AI by a year. If computers increasing two to three
[65:00] X every single year, they prevent a whole year of AI deployment, such that you're a year behind where you've otherwise been. During... RSI, you're getting Three to six years of AI progress in a single year. But they don't just limit compute. Right. They also limit the lab's ability to release the model internally. Right. We saw that. Anthropic had to stop giving Mythos to foreign employees for a bit. I didn't know that was true. Like internally as well. I mean, that's what they claimed. I thought it was just a different checkpoint that was not Mythos, but it was basically Mythos. Yeah. But I mean, like stuff like that is not going to be allowed either. Right. Like the government is dumb, but they're not that dumb. Right. Like, you know, I would hope at least, um, You know, governments are going to not want companies, at least the U.S. government has the cards here, there's not going to want Anthropic to use Mythos 4 internally. They're going to be like, hold the fuck on, right? Like, slow down. You know, because because all of these regulatory reasons, everyone who's elected is going to hate AI, even the people who are elected already hate AI. all the constituents, you're going to literally have, like, I bet you at some point, your parents are going to call you and be like, Dwarka Shvita, you're doing a terrible job. You're making every AI progress happen faster. It's going to happen. It's going to happen. I mean, maybe you educate people, right? And maybe if they're smarter, they're progressing AI faster. But anyways, like, you're going to have real world constraints on the progress and development and He... It's like we could tear ourselves apart before we get there. Jane Street is hiring for two separate ML internships right now. One focused on ML engineering and the other focused on ML research. I sat down with Alok, who helps run the research track to learn more about that program. I think this domain is fundamentally understudied. Often we have unanswered questions within our deep learning research team where we don't understand some, say, some market participants' behaviors or certain dynamics of how trading happens. questions make for really good intern projects because they are ultimately topics that we care about and just haven't gotten around to figuring out yet. So even as an intern, you'll be contributing to real research, not working on some sort of contrived exercise. The Jane Street team follows Frontier LLM research closely. A relatively common intern project is adapting a recent paper to financial markets, which come with their own set of gnarly problems. Ultimately, we're trying to model thousands of interconnected, irregular time series. The signal-to-noise ratios are extremely to do the same. So we have this like adversarial, non-stationary, extremely high dimensional problem that we were trying to solve. To be clear, you don't need to know anything about finance in order to be a good fit. As long as you have a background in ML research, Jane Street can teach you the rest. Their 2027 internship applications are open now. Apply at janestreet.com slash dworkash. One thing I find crazy about these scenarios is Just how... much of the world's future labor supply ends up in very few companies and also how fast the labor supply grows year over year. So if like computer the frontier, you know, in flop terms is growing four or five X a year. And further, the computer required to achieve the liquid abilities like decreasing three X a year. So the computer the frontier is Basically, the effective AI population size at the frontier lab is increasing 10x year over year. And so that doesn't really matter that much right now because AIs are not good enough to do full jobs or be like as autonomous as people in their capacity to do work or pull off schemes or whatever. But if the current trend continues, you have a world where open AI goes from having, say, 10 million basically AI laborers this year to 100 million the next year to a billion the year after that. - Yeah. Even if compute scaling slows down, it doesn't take many more years before each company individually has more... Labor equivalents. than there are people on earth. And I think that's like a thing that is very plausible by the end of this decade. There's more... AI labor, more effective population within a single lab, than there are people on Earth. So we talk often about centralization of power because of nationalization or whatever, but we don't think enough about the fact that we're actually moving very fast into a regime where most people, in terms of work output or something, is just concentrated. within two labs, we're consuming more and more of the world's compute. And so if these AI's are misaligned, then most of the world is misaligned, basically, because most of the world's minds are there. But even if they're not, it's just very few companies have like a lot of influence or a lot of control. Yeah, it's sort of there's the whole spat recently where it's like, I think Gavin Baker was like, Dario believes that there's only one company in the world. And then, you know, Sholto and Dario came out and were like, no, no, no, we didn't say that.
[70:00] But ultimately, you know, if you believe in RSI, you believe in the labs are the most effective user of compute and can generate the most value from the compute, then the only thing that's going to happen is centralization of compute. And if you believe in, you know, sort of AI researchers, RSI, AGI, then all of this exists. All of this is the base. This is even true if there's no RSI. The current effective, like… Effective population of the frontier is currently increasing 10x year over year for a given level of capabilities, right? So if you get to the level of capabilities, which is... a human, a very competent remote worker, or, like, a very competent software engineer, or a very competent researcher, that population of those would, like, 10x year over year, the current rate of capability. I see, and without RSI. And then once you have RSI, it's even crazier. Then it's, like, maybe growing, like, 100x a year, or 1,000x a year. Or they're, like, intelligence is increasing, but the population isn't increasing. Or it's a mixture of the two, right? Yeah, I mean, I guess, like... What world do you see, Dwarakash, where everything is not centralized? Because it seems to me that every force... is screeching towards centralization. Right. And that's scary as hell. Yeah. Um, I would love for it not to be centralized completely. But maybe that's the whole point of a machine that loves grace, right? Is... It is everything and it makes our lives great. Yeah, it's so hard to think about the future. But I agree with you that... I think the fundamental problem is that lab... AI training has huge economies of scale because any effort you spend into training in AI for a specific skill or specific set of knowledge gets amortized across billions of sessions or billions of users. So that's like one effect. The other effect is if you're slightly ahead in the AI race and computers in shortage, you can charge a much higher markup because you can better economize this scarce resource. So there's like two effects which are give more and more to the person who's like ahead in the AI race. So if there's models that are learning from deployment, and one model is deployed much more widely than another one, it's getting much more real-world data. Yeah, your point has taken that whether it's user deployment and continual learning, whether it's training, having these economies of scales, whether it's the incremental progress that the best AI model helps you to make the next AI model, RSI, all of these things... Oh, sorry, I didn't even mention RSI. All of these things point to centralization. So I think one of the big intellectual projects, honestly... um that yeah we should spend some time thinking about uh async or at least i'll spend some time thinking about is what is a vision of like a decentralized broadly empowered future after agi that takes these economies of scale seriously The alternative vision is that the government controls it. And maybe you think that you can trust the government more because it's not a private corporation. I don't trust the government and I don't trust Dario and I don't trust Sam. Yeah, that's the problem, right? But there's no, at least, obviously it's very easy to be wrong about the future and you don't anticipate a key effect or something that changes everything. But ex ante... It's very hard to see a reason why there were like how we avoid a scenario where we have to choose what capitalism worked. Right. It's the decentralized decision making and decentralized power and why super centralized capitalistic economies actually grew slower than super decentralized capitalist economies. To some extent, you have to have rule of law and all this. But then AI flips all this on its head. Right. And ultimately, you're like, actually, private ownership is probably not the most efficient economy. It's still private ownership, but it's like, how many firms are really involved in this share of the economy that's not, what, like 5% of the economy or something like that in the US? Sorry, $1 trillion. divided by 30, less than that, sorry. But yeah, maybe 2% of the economy right now. It's like, NVIDIA is a huge share of it, and Anthropic, and OpenAI, and these hyperscalers. And obviously there's other firms involved, but like a large share of the AI servers are just having from very few companies. So it's like, it could be private property, but like very few companies are involved. This is what the structure of the market is doing. What can prevent it? Unless AI progress slows down, unless governments regulate the fuck out of it, this is all that happens, in which case We're headed for a world where either we have super concentration of resources and we pray that that one company gets everything right. Or we have governments slow everything down and people slow everything down and you have a slowdown of progress somehow, hopefully. And there is a more of a balance of power. And even as we go towards an AGI, ASI, RSI. Um... Everything along the way will still lead to someone's going to allocate, going to capture more resources. So it's kind of hard for a framework in which AI doesn't lead to super concentration. Yeah. Now, the one positive thing here is that today, Anthropic does not capture most of the value. So we can talk all we want about, oh, you know, they went from $20 million per megawatt to $100 million a megawatt, but they're still paying $13 million. for a lot of the compute they're buying. Mm.
[75:00] at the end of the day, the reason they've gone to $100 million per megawatt is because Jane Street is capturing... $300 million per megawatt or $500 million per megawatt. Where Dworkesh, from researching his podcast and learning about credit, is capturing, you know, how many dollars per megawatt? Now, how much can you use? Tough. Yeah, yeah, yeah. But, you know, I think that's the, like, one saving grace is that the rest of the economy maybe profits so much more from Anthropik. No, no, but the whole logic you were laying out earlier is... of them reallocating inference to ARND. The whole logic of that is that the returns to labor inside AI labs is much higher than the returns outside. This is my cope. I agree. In all scenarios of the world, there's 80,000 worlds and only one of them, Anthropic, doesn't own the whole world. Is that... Again, power concentrates because I don't want to send the tokens outside. They're more valuable inside. And so it's the same thing. Why would I let Jane Street make all this money off of these degenerate options traders? Hey, there's some odds there. Come on. Jesus Christ. No, I think it's great. I think it's great. It's a good value for the world to make it an efficient market. Jane Street making all this money off of getting the worldview correctly, making money off of degenerate options traders, whatever it is, Why would anthropic allocate compute to that? If the end monetization that Jane Street has per megawatt is $200, so they're willing to pay Anthropic $100, well, what if Anthropic can just generate hundreds of millions of dollars per megawatt by using that compute internally? And that's what's happening. Well, on that somber note, I guess we'll meet again when the RSI is officially kicked off. You're not going to hit me on your podcast again for like two months? All right. Cool. Thanks, dude.
