Fervo Targets 100GW Geothermal Pipeline for AI Data Centers
Ai4
Enhanced geothermal serves as the only scalable zero-carbon baseload power capable of satisfying the massive 24/7 electricity demands of next-generation AI infrastructure.
## Background
Published on YouTube by Ai4 on August 19, 2026, this talk features Tim Latimer, Co-Founder and CEO of Fervo Energy. The page description outlines his argument that energy is the primary bottleneck for scaling AI, and that enhanced geothermal energy provides the necessary 24/7, zero-carbon baseload power for future infrastructure.
## TL;DR
Tim Latimer presents the case for enhanced geothermal systems as the essential power solution for the rapid expansion of AI data centers. He details how Fervo Energy adapts oil and gas drilling techniques to tap into deep subsurface heat, achieving dramatic improvements in well performance and construction speed. Latimer also explains how Fervo utilizes modular plant designs, behind-the-meter colocation, and massive subsurface datasets combined with AI to build redundant, cost-effective, and scalable clean energy.
## Atmosphere & style
The presentation carries an authoritative, visionary, and pragmatic tone, blending corporate milestone celebration with technical optimism. Latimer delivers his talk with engaging humor—joking about public company disclosures—while grounding high-level historical energy arcs in rigorous engineering data. The atmosphere feels like an energetic, forward-looking tech keynote that connects physical industrial realities directly to the future of computing.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=LL2abpmLhT4&t=4s) Fervo Energy was founded eight years ago as a Stanford University spin-out and went public in May with a $2.2 billion IPO.
- [01:22](https://www.youtube.com/watch?v=LL2abpmLhT4&t=82s) AI commercialization and global productivity gains are physically constrained by energy access and power availability.
- [01:22](https://www.youtube.com/watch?v=LL2abpmLhT4&t=82s) Data center power requirements have grown exponentially over 15 years, expanding from 30 MW in 2010 to planned 2 GW campuses today.
- [04:29](https://www.youtube.com/watch?v=LL2abpmLhT4&t=269s) The U.S. power grid suffers from structural stagnation, with total U.S. electricity consumption in 2023 almost identical to 2000 levels.
- [04:29](https://www.youtube.com/watch?v=LL2abpmLhT4&t=269s) Over 500 U.S. counties have introduced restrictions or bans on data centers due to concerns over electricity rates, pollution, noise, and water use.
- [07:37](https://www.youtube.com/watch?v=LL2abpmLhT4&t=457s) Traditional clean and baseload power alternatives (natural gas, wind/solar, nuclear, hydro) face severe supply, speed, geographical, or intermittency bottlenecks.
- [07:37](https://www.youtube.com/watch?v=LL2abpmLhT4&t=457s) Fervo Energy transforms geothermal from a niche geographic option into a scalable technology by utilizing horizontal drilling and hydraulic stimulation in deep granite.
- [10:16](https://www.youtube.com/watch?v=LL2abpmLhT4&t=616s) Fervo's engineered geothermal system uses a closed-loop Binary Organic Rankine Cycle plant, resulting in zero carbon emissions and zero water loss.
- [12:28](https://www.youtube.com/watch?v=LL2abpmLhT4&t=748s) Fervo's 500 MW Cape Station project in Utah can scale to 4 GW of continuous capacity, supported by a 600,000-acre land portfolio representing nearly 50 GW of potential projects.
- [12:28](https://www.youtube.com/watch?v=LL2abpmLhT4&t=748s) Technological improvements increased Fervo's power output per well from 3 MW to 15 MW while reducing drilling time from 70 days down to 21 days.
- [16:17](https://www.youtube.com/watch?v=LL2abpmLhT4&t=977s) Fervo structures multi-gigawatt facilities into independent 50 MW modular power blocks to deliver high system redundancy and 99.9%+ reliability.
- [16:17](https://www.youtube.com/watch?v=LL2abpmLhT4&t=977s) By collocating data centers behind the meter at geothermal production sites, developers can bypass lengthy grid interconnection delays and avoid local rate spikes.
- [17:45](https://www.youtube.com/watch?v=LL2abpmLhT4&t=1065s) Fiber-optic and microseismic sensors at Fervo's Utah site generate over 3 TB of subsurface data daily (over 3 PB total) to map fluid flow miles underground.
- [19:14](https://www.youtube.com/watch?v=LL2abpmLhT4&t=1154s) Fervo partnered with NVIDIA and Pacific Northwest National Laboratory to build AI-driven subsurface digital twins to optimize subsurface geothermal operations.
## Detailed walkthrough
Here is the detailed, faithful walkthrough of part 1 of the transcript.
---
[00:04](https://www.youtube.com/watch?v=LL2abpmLhT4&t=4s) AI's Unforeseen Energy Bottleneck and Fervo Energy's Public Debut
Tim Latimer, Co-Founder and CEO of Fervo Energy, opens his talk at an AI conference by reflecting on his company’s unexpected journey. Eight years prior, when he and his co-founder spun Fervo out of Stanford University, it felt as though every other entrepreneur on campus was launching an AI startup. As they parted ways with their peers, Latimer felt like they were the "odd child" choosing to pursue clean energy. Eight years later, however, Fervo finds itself acting almost like an AI company because energy and artificial intelligence have become inextricably intertwined. Commercializing AI and unlocking global productivity gains now depends entirely on continued, massive energy access.
Latimer highlights a major corporate milestone: Fervo held its IPO in May, raising $2.2 billion—a tangible reflection of how critical new power generation buildout has become to the modern economy. He jokes about the downside of being a public company: he is now legally required to show a detailed financial disclaimer slide in every presentation for the rest of his life, prompting the audience to quickly read through the required legal disclosures.
- **Founding Story:** Spun out of Stanford University eight years ago.
- **Capital & Market Validation:** Raised $2.2 billion in an IPO in May.
- **Key Realization:** AI development and commercialization are physically constrained by power availability.
---
[01:22](https://www.youtube.com/watch?v=LL2abpmLhT4&t=82s) Historical Energy Waves and Tech Industry Realities
Latimer frames the current momentum around AI within the broader arc of human history, arguing that human progress and prosperity have always been tied to access to energy. Every major industrial revolution has been underpinned by advances in energy technology that yielded cleaner, cheaper, and more accessible power. Whenever economic stagnation or a pause in energy expansion occurs, it is simply because the next key technological innovation has not yet arrived. AI represents the next massive wave of human productivity, and like every revolution over the past 200 years, it demands a dramatic expansion of energy resources.
To demonstrate that this is an industry-wide consensus, Latimer shares quotes from prominent tech leaders who recognize that energy is becoming their primary operational constraint:
- **Sam Altman (OpenAI):** Famously told Congress that the cost of intelligence will eventually collapse to the cost of energy.
- **xAI:** Stated that the next major technical shortage will be electricity.
- **Satya Nadella (Microsoft):** Noted that operations will soon be power-constrained rather than chip supply-constrained.
- **Jensen Huang (NVIDIA):** Highlighted that using energy for intelligence is "the best use of energy we can imagine."
Latimer traces how data center power needs have evolved over the last 15 years since the rollout of cloud computing. In 2010, the largest data center in the world was a 30-megawatt facility—a footprint that seems quaint by modern standards. That baseline doubled by 2013, grew to 650 megawatts at a single site by 2017, and today developers are planning 2-gigawatt data center campuses. To put that in perspective, 2 gigawatts equals the electricity consumed by a city of 2 million people. Finding locations worldwide where that amount of power can be deployed quickly, cheaply, and reliably is becoming nearly impossible.
---
[04:29](https://www.youtube.com/watch?v=LL2abpmLhT4&t=269s) The Grid Stagnation Crisis and Community Backlash
The surge in demand from AI comes at a time when the U.S. electrical grid has suffered from a multi-decade period of structural stagnation. Remarkably, the total electricity consumed in the United States in 2023 was almost identical to the amount consumed at the turn of the century in 2000. Because the power grid did not expand for over two decades, the industrial muscle required for growth—building transmission lines, manufacturing large transformers, maintaining specialized supply chains, and training skilled craftspeople—atrophied significantly.
Restarting this growth engine overnight is extraordinarily difficult. Making matters worse, the amount of 24/7 reliable "baseload" power on the U.S. grid is actively declining due to plant retirements. While data centers require "triple-nine, five-nine" (99.999%) reliability around the clock, the grid is shedding baseload capacity. Over the next decade, forecasts project that out of a total U.S. grid capacity of roughly 1,000 gigawatts, an additional 100 to 250 gigawatts of new electricity must be added solely to accommodate AI and data center growth.
Attempting to scale conventional power sources creates severe friction:
- **Fossil Fuels (Gas & Coal):** Suffer from fuel price volatility, long supply chain lead times, pollution, and severe community resistance.
- **State Interventions:** Latimer notes that even in Texas—famously known as the "wild west" of business-friendly energy deregulation (drawing a nod and shout-out from a Texan in the audience)—Governor Greg Abbott announced a moratorium on new data center construction. New York instituted a similar pause months earlier, showing that pushback spans the entire political spectrum.
- **Local Resistance:** Over 500 counties across the United States have implemented local bans or restrictions on data centers due to community fears over rising local electricity rates, pollution, noise, traffic, and heavy water consumption.
---
[07:37](https://www.youtube.com/watch?v=LL2abpmLhT4&t=457s) Evaluating Energy Alternatives and the Next-Gen Geothermal Breakout
When analyzing the existing zero-carbon and baseload generation options, tech companies find themselves facing severe structural bottlenecks across every traditional category:
- **Natural Gas:** Ordering a new gas turbine today requires waiting in line for up to six years due to global equipment backlogs.
- **Solar and Wind:** While valuable to the broader power system, their intermittent nature does not match the flat 24/7 generation profile that a high-uptime data center requires.
- **Nuclear:** Large-scale commercial nuclear deployments are a decade or more away. Latimer notes that the last nuclear plant built in the U.S. ran significantly over budget and took about a decade longer to construct than originally planned.
- **Hydropower:** Highly clean and reliable, but strictly constrained by geographic topography and unable to scale to meet new regional demands.
This bottleneck is where geothermal energy enters as the ideal solution. Geothermal power is not fundamentally new—the world's first geothermal power plant was built in Larderello, Italy, around the turn of the 20th century in the early 1900s. However, conventional geothermal power has historically accounted for less than 1% of global electricity generation because it relied on rare geologic anomalies where volcanoes or tectonic faults brought intense heat directly to the surface.
Fervo Energy's core breakthrough is transforming geothermal from a niche geographic phenomenon into a globally scalable infrastructure technology. By leveraging horizontal drilling and hydraulic stimulation techniques adapted from the modern oil and gas industry, Fervo can harvest heat from hot rock deep underground anywhere in the world, freeing the industry from its historical volcanic constraints.
---
[10:16](https://www.youtube.com/watch?v=LL2abpmLhT4&t=616s) Technical Deep Dive: How Fervo's Engineered Geothermal System Operates
Latimer presents an instructional video detailing Fervo's underground engineering process. The journey begins long before a drill bit touches the earth, using subsurface data mapping and computational modeling to locate deep underground heat reservoirs.
<Sequence>
<Step title="Vertical and Horizontal Drilling">
Drill rigs bore vertically downward over 10,000 feet into hot, dense basement rock (such as granite). The drill bit then turns horizontally, navigating over a mile laterally through the hot formation. A parallel "sister well" is drilled alongside it, creating a paired system consisting of an injection well and a production well.
</Step>
<Step title="Groundwater Protection and Well Isolation">
To protect shallow freshwater aquifers, the wells are encased and sealed behind multiple layers of heavy steel casing and solid cement, completely isolating the fluid system from the surrounding environment.
</Step>
<Step title="Multi-Stage Hydraulic Stimulation">
Engineers execute multi-stage hydraulic fracturing between the horizontal wellbores. This creates a highly permeable network of micro-fractures in the rock, dramatically expanding the subsurface surface area so circulating water can maximize heat exchange with the surrounding hot granite.
</Step>
<Step title="Closed-Loop Surface Power Generation">
Cool water is pumped down the injection well, heats up as it traverses the engineered fracture network, and rises to the surface through the production well as high-temperature geothermal brine. At the surface, the brine enters a Binary Organic Rankine Cycle (ORC) power plant. Heat exchangers transfer the thermal energy to a secondary working fluid with a lower boiling point, which vaporizes to drive a generator turbine. The working fluid is condensed back into liquid via air-cooled condensers, while 100% of the geothermal brine is continuously reinjected underground in a fully closed loop—losing zero water to evaporation and producing zero carbon emissions.
</Step>
</Sequence>
---
[12:28](https://www.youtube.com/watch?v=LL2abpmLhT4&t=748s) Project Scale, Technological Acceleration, and Cape Station
Fervo's technology meets every critical criteria required by data center operators: rapid execution, 24/7 reliability, competitive costs, minimal land footprint, zero water loss, and zero emissions. Fervo proved out its concept with a successful commercial pilot in Nevada built in partnership with Google, which has been reliably feeding zero-carbon electricity into the grid using horizontal geothermal wells for nearly three years.
Building on that success, Fervo is now constructing **Cape Station**, a massive 500-megawatt geothermal project located in southwest Utah (roughly a three-hour drive from Salt Lake City):
- **Rapid Construction Speed:** Fervo received its initial ground-clearing permits for the first power plant block at Cape Station in October 2024. Exactly one year later, construction on that plant was fully completed.
- **Expansion Potential:** While initial construction focuses on 500 megawatts, the broader Cape Station subsurface field can scale to 4 gigawatts of total continuous power capacity.
- **Land Portfolio:** Fervo holds over 600,000 acres across the western United States, representing nearly 50 gigawatts of potential geothermal project pipeline under its direct control.
```
Fervo Well Architecture & Performance Evolution
====================================================================================
Design Iteration Vertical Depth Lateral Length Drill Time Power Output
------------------------------------------------------------------------------------
2023 Nevada Pilot ~9,000 ft 3,200 ft 70 days 3 MW
Cape Phase 1 ~10,000 ft 5,000 ft -- --
Design 3.0 ~12,000 ft >7,920 ft (1.5 mi) 21 days 15 MW
====================================================================================
```
By making these technological leaps, Fervo achieved a **5x increase in power output per well** (from 3 MW to 15 MW) while simultaneously **reducing drilling time by nearly 70%** (from 70 days down to 21 days), even when drilling deeper into hotter, harder rock. To make these depths tangible, Latimer notes that the Empire State Building serves as a useful height reference point: Fervo's initial 9,000-foot deep well reached depths equivalent to roughly six Empire State Buildings stacked end-to-end underground.
To maximize manufacturing efficiency, Fervo deploys standardized, modular **50-megawatt power blocks**, allowing the company to rapidly move down the learning curve and lower the levelized cost of energy with each new deployment.
---
<ElicitationsGroup message="Explore related energy and infrastructure topics:">
<Elicitation label="Analyze the economics of geothermal vs nuclear for data centers" query="Compare the levelized cost of energy (LCOE), construction timelines, and land footprint of enhanced geothermal systems versus small modular reactors (SMRs) for powering AI data centers."/>
<Elicitation label="Detail Fervo Energy's closed-loop Binary Organic Rankine Cycle" query="Explain how a Binary Organic Rankine Cycle (ORC) works in low-to-medium temperature geothermal power generation, including the working fluids used and efficiency optimizations."/>
</ElicitationsGroup>
Here is the detailed walkthrough of the transcript section:
### [16:17](https://www.youtube.com/watch?v=LL2abpmLhT4&t=977s) Modular Redundancy and Colocation Strategy for AI Data Centers
Fervo Energy achieves "triple 9" (99.9%) reliability or higher for data centers by structuring large projects into modular, independent power plants rather than relying on single, massive generation units. For example, a 1-gigawatt project is designed as 20 independent 50-megawatt power plants. This layout provides inherent redundancy: if one turbine is taken offline for maintenance or suffers an unplanned outage, the remaining 19 units continue operating smoothly. This modular architecture makes Fervo’s multi-gigawatt clusters a key differentiator in the energy market.
To bypass lengthy grid interconnection delays, Fervo is pivoting directly toward collocating data center loads behind the meter with their geothermal projects. This approach speeds up deployment while ensuring energy remains affordable and sustainable. Moreover, behind-the-meter colocation avoids shifting extra grid upgrade costs onto local consumers—a growing political friction point that has slowed down conventional energy buildouts. Over the past several years, Fervo has contracted 650 megawatts of power, representing over $7 billion in revenue, and removing grid constraints will allow them to scale up even faster.
- **Project Redundancy**: A 1 GW project consists of 20 separate 50 MW power plants to ensure high availability.
- **Contracted Power**: 650 MW contracted, representing over $7 billion in long-term revenue.
- **Key Advantage**: Behind-the-meter colocation eliminates reliance on lengthy grid interconnection timelines and avoids passing costs to retail consumers.
### [17:45](https://www.youtube.com/watch?v=LL2abpmLhT4&t=1065s) Utilizing Microseismic Monitoring and Massive Subsurface Data
Beyond supplying energy to AI infrastructure, Fervo actively leverages internal AI tools to accelerate its own engineering and operational advancements. A central challenge in geothermal energy for over a century has been understanding subsurface fluid flows occurring two miles underground. Fervo solves this using microseismic monitoring and a fiber-optic sensing platform deployed across its well fields to continuously map fluid flow across dozens of wells.
This sensing technology generates massive volumes of data: at Fervo's site in Utah, the fiber-optic sensors capture over 3 terabytes of data per day, accumulating over 3 petabytes of data since the operation began. While this data volume was once incomprehensible, AI tools now process petabytes of subsurface information to solve long-standing geothermal mysteries. Fervo also applies AI and automation across software simulation, predictive maintenance, drill bit damage monitoring, and water supply optimization to drive down costs.
- **Data Generation**: Over 3 TB of data generated daily from fiber-optic sensors at the Utah project site.
- **Total Data Volume**: Over 3 PB of subsurface data collected to date.
- **AI Applications**: Microseismic subsurface monitoring, software simulation tools, predictive maintenance, drill bit damage detection, and water supply optimization.
### [19:14](https://www.youtube.com/watch?v=LL2abpmLhT4&t=1154s) Partnerships for Digital Twins and the Future of Geothermal AI Power
Fervo recently announced a partnership with NVIDIA and Pacific Northwest National Laboratory (PNNL). This initiative utilizes exascale computing at PNNL alongside NVIDIA’s chip capabilities to construct high-fidelity "digital twins" of subsurface geothermal environments miles below the earth's surface. These digital simulations allow Fervo to rapidly optimize operations and drive down cost curves.
Geothermal energy is positioned as the primary catalyst for unlocking the next wave of AI advancements. To maintain public support and political viability, AI expansion must be sustainable, fast, reliable, and affordable for local communities. Fervo envisions the enhanced geothermal industry scaling to over 100 gigawatts, delivering the baseload power required for the AI revolution.
- **Partners**: Partnership between Fervo Energy, NVIDIA, and Pacific Northwest National Laboratory (PNNL).
- **Technology**: Building high-fidelity subsurface digital twins powered by exascale compute and AI chips.
- **Industry Scale Target**: Scaling enhanced geothermal power to 100+ GW.
## Notable quotes
> [01:22](https://www.youtube.com/watch?v=LL2abpmLhT4&t=82s) the cost of intelligence will eventually collapse to the cost of energy. — Sam Altman
> [01:22](https://www.youtube.com/watch?v=LL2abpmLhT4&t=82s) the next major technical shortage will be electricity. — xAI
> [01:22](https://www.youtube.com/watch?v=LL2abpmLhT4&t=82s) the best use of energy we can imagine. — Jensen Huang
## Mentioned
Tim Latimer, Fervo Energy, Stanford University, Sam Altman, OpenAI, xAI, Satya Nadella, Microsoft, Jensen Huang, NVIDIA, Greg Abbott, Google, Cape Station, Empire State Building, Pacific Northwest National Laboratory
## Video description
Tim Latimer - Co-Founder & CEO, Fervo Energy
The AI revolution has an energy problem. Every model trained, every inference run, every data center scaled demands massive, reliable power. I'll make the case that the companies who solve the energy problem will be the ones who win the AI race, and that geothermal is the only source that delivers the always-on, zero-carbon baseload that AI infrastructure actually requires. Attendees will leave with a new lens on AI's physical limits and a clear view of why enhanced geothermal is the most consequential energy bet of the next decade. If you're building AI and you haven't thought seriously about where the power comes from, this is the talk to attend.
Fervo Energy CEO Tim Latimer argues that electricity availability has replaced chip supply as the defining bottleneck for artificial intelligence. He contends that enhanced geothermal systems, adapted from oil and gas horizontal drilling, provide the scalable, zero-carbon baseload power required to circumvent an atrophied electrical grid.
AI Demand Collides with an Atrophied Grid (1:22)
Latimer notes an emerging consensus among technology leaders that power availability will dictate the pace of AI deployment. Data center footprints have expanded from 30-megawatt facilities in 2010 to planned 2-gigawatt campuses, each consuming as much electricity as a city of 2 million residents. Over the next decade, projections indicate that U.S. data center growth will require adding 100 to 250 gigawatts of capacity to a national system that currently totals roughly 1,000 gigawatts.
Meeting this surge is complicated by decades of structural stagnation. Total U.S. electricity consumption in 2023 was virtually flat compared to 2000, leaving supply chains, equipment manufacturing, and skilled labor unprepared for rapid expansion. Alternative sources face distinct bottlenecks: natural gas turbines carry backlogs of up to six years, commercial nuclear additions remain a decade away, and solar and wind cannot provide flat baseload power. Simultaneously, data center developers face severe community resistance, including building moratoria in Texas and New York alongside restrictions across more than 500 U.S. counties over utility rates, water consumption, and land use.
Adapting Drilling Tech for Geothermal Heat (10:16)
Traditional geothermal power historically accounted for less than 1% of global generation because it relied on rare volcanic areas where heat reached the surface naturally. Fervo bypasses this geographic limitation by utilizing horizontal drilling and hydraulic stimulation techniques developed by the oil and gas sector to engineer geothermal reservoirs in hot basement rock such as granite.
The system drills vertically over 10,000 feet, turns horizontally for more than a mile, and pairs with a parallel sister well. High-pressure hydraulic fracturing creates an underground network of micro-fractures between the wells to maximize heat exchange. Water pumped down an injection well absorbs subsurface heat and returns to the surface through a production well. In a closed-loop Binary Organic Rankine Cycle power plant, the thermal brine vaporizes a secondary working fluid to drive generator turbines before being fully reinjected underground, resulting in zero greenhouse gas emissions and no evaporative water consumption.
Modular Deployment and Rapid Drilling Gains (12:28)
Following a multi-year commercial pilot in Nevada with Google, Fervo began developing Cape Station in southwest Utah. The site has an initial 500-megawatt phase with an underlying subsurface capacity that Latimer estimates can support 4 gigawatts of continuous generation. Construction on the initial power plant block was completed in October 2025, one year after initial ground permits were issued. Across the western United States, Fervo controls more than 600,000 acres, which it projects could support roughly 50 gigawatts of geothermal capacity.
Operational metrics show rapid technological learning. Between its early Nevada wells and its latest design iteration, Fervo increased per-well electricity output by 5x (from 3 megawatts to 15 megawatts) while cutting drilling duration by nearly 70% (from 70 days down to 21 days), even as vertical depths reached 12,000 feet with 1.5-mile horizontal laterals. To maintain cost predictability and scale manufacturing, the company builds standardized 50-megawatt generation blocks.
Behind-the-Meter Colocation and AI Digital Twins (16:17)
To satisfy data center demands for 99.9% reliability, Fervo partitions large installations into multiple independent 50-megawatt units—such as twenty plants for a 1-gigawatt project—to ensure continuous power if individual turbines go offline. Fervo also favors collocating data centers behind the meter. This strategy avoids lengthy regional grid interconnection queues and prevents the political backlash of passing grid upgrade costs to local retail ratepayers. The company has secured 650 megawatts of contracted power, representing over $7 billion in long-term revenue.
To optimize subsurface performance, Fervo incorporates fiber-optic sensors and microseismic monitoring across its fields, generating over 3 terabytes of flow data daily at Cape Station and accumulating more than 3 petabytes to date. In partnership with NVIDIA and Pacific Northwest National Laboratory, Fervo uses AI accelerators and exascale computing to run high-fidelity digital twins of underground reservoirs. Latimer projects that enhanced geothermal can scale past 100 gigawatts nationally to support computing growth.
While Fervo has demonstrated operational feasibility and secured major revenue contracts, the broader geothermal transition depends on sustaining rapid cost reductions across varied geologies and successfully executing multi-gigawatt, behind-the-meter colocation without unforeseen long-term reservoir depletion.
## Background
该视频由 Ai4 于 2026 年 8 月 19 日发布在 YouTube 平台上。视频主要记录了 Fervo Energy 联合创始人兼 CEO Tim Latimer 的主题演讲，阐述了 AI 发展面临的严峻能源瓶颈，并介绍了增强型地热技术如何为 AI 数据中心提供全天候、零碳排放的基荷电力。
## TL;DR
随着 AI 训练和推理对电力的需求爆炸式增长，传统电网与清洁能源难以满足数据中心对高可靠性基荷电力的渴望。Fervo Energy 借用油气行业的深井水平钻探与水力压裂技术突破了传统地热的地缘限制，成功研发出增强型地热系统（EGS）。公司在犹他州的旗舰项目 Cape Station 展现了惊人的建造效率与降本曲线，不仅能通过模块化设计提供高冗余的“表后”电力，还深度结合光纤传感与 AI 数据分析优化地下流体建模。演讲表明，地热能将成为支撑下一代 AI 基础设施突破的核心能源动力。
## Atmosphere & style
整场演讲充满前沿科技感与工业实干主义色彩，Tim Latimer 的言辞自信、专业且逻辑严密。他以幽默轻松的开场拉近与观众的距离，随后用清晰的数据对比和行业巨头的观点勾勒出 AI 能源危机的紧迫感。从技术原理的娓娓道来到吉瓦级项目的宏大构想，现场氛围既有对现实电力瓶颈的冷静剖析，又洋溢着对清洁能源未来发展的蓬勃雄心。
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=LL2abpmLhT4&t=4s) Fervo Energy 于 2026 年 5 月完成 IPO 融资 22 亿美元，AI 模型训练与推理对电力的庞大需求使其跃升为 AI 生态系统的核心环节。
- [02:21](https://www.youtube.com/watch?v=LL2abpmLhT4&t=141s) 数据中心单体规模从 2010 年的 30 MW 暴增至如今规划的 2 GW，科技巨头普遍认同电力已取代 GPU 成为 AI 发展的主要瓶颈。
- [02:21](https://www.youtube.com/watch?v=LL2abpmLhT4&t=141s) 美国电网在过去二十年间几乎零增长，缺乏应对暴增用电需求的基础设施，未来十年仅满足 AI 需求就需新增 100 至 250 GW 电力。
- [06:30](https://www.youtube.com/watch?v=LL2abpmLhT4&t=390s) 传统能源方案（天然气、核电、风光）因交货周期长、成本超支、间歇性或政策监管阻力等原因，难以满足数据中心 24/7 超高可靠性的基荷电力需求。
- [08:49](https://www.youtube.com/watch?v=LL2abpmLhT4&t=529s) Fervo 通过引入石油与天然气领域的深层水平钻井与水力压裂技术，打造增强型地热系统（EGS），摆脱了传统地热极端依赖自然火山的限制。
- [08:49](https://www.youtube.com/watch?v=LL2abpmLhT4&t=529s) EGS 采用双井闭环循环与有机朗肯双循环系统（ORC），将地热流体 100% 重新注入地下，实现全程零蒸发损失与零碳排放。
- [13:19](https://www.youtube.com/watch?v=LL2abpmLhT4&t=799s) 从内华达州试点到 Cape Station 旗舰项目，Fervo 钻井耗时从 70 天降至 21 天，单井发电能力提升 5 倍至 15 MW。
- [13:19](https://www.youtube.com/watch?v=LL2abpmLhT4&t=799s) Fervo 拥有超 60 万英亩地热产权并储备近 50 GW 潜在产能，利用 50 MW 标准化功率模块推进吉瓦级规模化扩展。
- [16:17](https://www.youtube.com/watch?v=LL2abpmLhT4&t=977s) 采用 20 个 50 MW 小型电厂组合成 1 GW 项目的模块化架构，提供极高固有冗余，确保单点故障不影响 overall 连续供电。
- [16:17](https://www.youtube.com/watch?v=LL2abpmLhT4&t=977s) Fervo 进军数据中心“表后”供电与同址共建模式，直接绕过漫长的电网互联审批流程，规避了因电价上涨引发的政治与社会阻力。
- [17:45](https://www.youtube.com/watch?v=LL2abpmLhT4&t=1065s) Fervo 部署光纤传感器平台进行微震监测，日增超 3 TB 数据，借助 AI 工具解开了困扰地热学界百年的地下流体流动之谜。
- [18:50](https://www.youtube.com/watch?v=LL2abpmLhT4&t=1130s) 联合 NVIDIA 与太平洋西北国家实验室（PNNL），利用外围级计算能力构建地下高保真数字孪生模型，加速推动地热向 100+ GW 规模扩展。
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=LL2abpmLhT4&t=4s) AI 革命下的能源瓶颈与 Fervo 的崛起
在大会现场，Fervo Energy 联合创始人兼 CEO Tim Latimer 讲述了公司独特的创业历程。八年前，他与合创始人将公司从斯坦福大学拆分独立时，身边几乎所有创业者都在涌向 AI 领域，而选择搞能源的他们反倒成了同辈眼中的“异类”。然而，八年后的今天，由于 AI 模型的训练、推理以及数据中心的大规模扩展对电力产生了近乎贪婪的需求，能源与 AI 的命运已紧密交织在一起。Fervo 不仅没有脱节，反而跃升为 AI 生态系统中至关重要的一环。全球商业化、AI 生产力的持续释放，完全取决于能否获得稳定且充足的能源支持。
随着电力基础设施建设的重要性被全球市场高度认可，Fervo 迎来了资本市场的重磅突破——公司于 5 月成功完成 IPO，一举筹集了 22 亿美元资金。 Tim Latimer 在现场打趣道，作为一家上市公司，虽然获得了海量资金，但代价就是今后演讲都必须展示那张带有免责声明的法律合规 Slide。
- **核心观点**：人类社会的每一次重大进步与工业革命，本质上都与更清洁、更低廉、更容易获取的能源突破息息相关。
- **关键数据**：Fervo 在 2026 年 5 月完成 IPO，成功融资 22 亿美元。
---
### [02:21](https://www.youtube.com/watch?v=LL2abpmLhT4&t=141s) 从 30 兆瓦到 2 吉瓦：数据中心爆发式增长与电网停滞
回顾云计算过去十多年的演进，数据中心对电力的需求经历了一场惊人的蜕变。2010 年，全球最大的数据中心容量仅为 30 兆瓦（MW）；到了 2017 年，这一数字飙升至 650 兆瓦；而如今在 AI 浪潮驱使下，科技巨头们动辄规划建设 2 吉瓦（GW）级别的单一超大规模数据中心园区。2 吉瓦的用电量相当于一个拥有 200 万人口的大型城市的总耗电量，这直接导致全球可承载此类超大项目的场地极度匮乏。
科技行业的领袖们早已对此达成共识：
* **Sam Altman (OpenAI)**：曾在国会听证会上表示，智能的最终成本归根结底将折算为能源的成本。
* **xAI 团队**：警告称下一波严苛的短缺将不再是芯片，而是电力。
* **Satya Nadella (微软)**：明确指出当前的瓶颈已从 GPU 供应转向了电力供应。
* **Jensen Huang (英伟达)**：强调将能源转化为“智能”是目前所能想象到的最佳能源利用方式。
然而，供需矛盾的核心在于美国电网基础设施的长期停滞。过去二十年间，美国整体电力消耗几乎保持平稳（2023 年的用电量与 2000 年基本持平），导致整个供应链、变压器制造以及工程技术人才队伍严重萎缩。面对突然爆发的 AI 供电需求，电网“增长肌肉”早已退化，无法一夜之间重启。
- **用电量对比**：2010 年最大数据中心为 30 MW，2017 年为 650 MW，当前规划达 2 GW（2,000 MW）。
- **电网需求**：未来 10 年，美国电网（现有总量约 1,000 GW）仅为了满足 AI 与数据中心的增长，就需要额外新增 100 至 250 GW 的电力供应。
- **基荷痛点**：传统基荷电力（24/7 不间断供应）的比重在过去 15 年间持续下滑，与数据中心对“99.999% 超高可靠性”的要求背道而驰。
---
### [06:30](https://www.youtube.com/watch?v=LL2abpmLhT4&t=390s) 传统能源的困局与电网监管“刹车”
在追求 24 小时稳定电力的过程中，现有能源方案均面临难以逾越的障碍。化石能源（天然气与煤炭）受制于燃料价格剧烈波动、碳排放压力以及严峻的社区环保阻力；风能和太阳能虽然发展迅速，但其间歇性发电机理无法直接匹配数据中心的全天候运行；新建核电站动辄需要十年以上的建设周期且屡屡超支；水电则受到极其严格的地形限制。
即便是在被誉为能源“自由天堂”、行政审批极度宽松的德克萨斯州，近期也出现了政策剧变。德州州长 Greg Abbott 宣布暂停新的数据中心建设审批。此前纽约州等地区出台类似限制时，业内尚将其归因于当地政治风向；但连德州也按下“暂停键”，无疑给整个科技与能源行业敲响了警钟。各地居民对电费上涨、噪音污染、水资源消耗及碳排放的抗议越发强烈，已有超过 500 个县出台了针对数据中心建设的禁令。
- **燃气轮机瓶颈**：当前订购一台新的天然气发电机组，排队等待交货的时间已长达 6 年。
- **核电困境**：上一座在美国建成的核电站不仅严重超支，且工期比预期整整延误了近十年。
---
### [08:49](https://www.youtube.com/watch?v=LL2abpmLhT4&t=529s) 突破地缘枷锁：增强型地热（EGS）的技术革命
地热能虽然拥有 24/7 连续稳定输出、零碳排放等天然优势，且早在 1900 年代初就在意大利建成了全球首座地热电站，但传统地热极端依赖天然火山或浅层地热资源。这种天然地理限制导致地热能占全球总发电量的比例长期不足 1%。
Fervo 的突破性创新在于将石油与天然气行业积累的高级钻井技术引入地热领域。通过深层水平钻井与水力压裂技术，Fervo 能够人工打造地下热交换网络，从而摆脱了“只能在火山上建地热站”的宿命，使地热发电具备了在广袤地区大规模复制的条件。
<Image alt="Fervo Energy 地热发电厂示意图，展示了双井系统和地下热交换网络" caption="Fervo 增强型地热系统（EGS）原理图" src="image_agent_tag_1"/>
---
#### 增强型地热技术运行机制
1. **地质建模与精准钻探**：在开钻前利用数据构建地下模型定位热源。垂直向下钻探超过 10,000 英尺（约 3,000 米）后转向，在高温岩层中沿水平方向继续钻进超过 1 英里（约 1.6 公里）。
2. **双井平行布置**：并行钻入一对姐妹井，分别作为注入井（Injection Well）和生产井（Production Well）。井壁采用多层重型钢制套管与水泥固井封闭，彻底隔绝浅层地下水。
3. **多段压裂连通**：对地下硬岩进行水力压裂，建立庞大的微裂缝网络，大幅增加水与热岩石的接触表面积，从而最大化热量回收效率。
4. **闭环循环发电**：
* 将冷水从注入井泵入地下，冷水吸收高温岩石的热量升温成为地热流体；
* 高温地热流体经生产井抽取至地面，进入**有机朗肯双循环系统（Organic Rankine Cycle, ORC）**；
* 地热流体在热交换器中将热量传递给低沸点的工作介质，驱动汽轮机发电；
* 发电后的地热流体经过空冷凝汽器降温，**100% 重新注入地下**，全程零蒸发损失、零碳排放。
---
### [13:19](https://www.youtube.com/watch?v=LL2abpmLhT4&t=799s) 商业化落地：从 Cape Station 到吉瓦级扩展
技术的可行性已在实践中得到验证。三年前，Fervo 与 Google 合作在内华达州启动了首个横向地热试点项目，目前已稳定运行近三年。基于试点经验，Fervo 正全力推进位于犹他州西南部的 **Cape Station** 旗舰项目（距离大会举办地约 3 小时车程）。
Cape Station 展现出了惊人的建设速度：2024 年 10 月拿到首批施工许可并清理场地，仅仅一年后，首座地热发电厂即告竣工。该项目初始规划容量为 500 兆瓦（MW），但单一 Cape Station 园区的潜在扩容极限可达 4 吉瓦（GW）。
- **资源储备**：Fervo 在美国西部已锁定超过 60 万英亩的地热产权，储备了近 50 GW 的潜在发电产能。
- **降本增效曲线**：
- **第一代试点井（2023）**：深度约 9,000 英尺，水平段 3,200 英尺，耗时 70 天，发电能力 3 MW。
- **Cape 第一阶段**：深度近 10,000 英尺，水平段 5,000 英尺。
- **3.0 架构设计**：深度达 12,000 英尺，水平段超过 1.5 英里（约 2.4 公里），单井发电能力飙升至 **15 MW**（较一代提升 5 倍），而钻井周期从 70 天骤降至仅需 **21 天**。
- **模块化构建**：项目采用 **50 兆瓦标准化功率模块（Power Block）** 推进，既加快了工程建造复用，又极大地压低了单位度电成本。
<FollowUp label="想了解 Fervo 的有机朗肯循环（ORC）具体是如何实现零水耗发电的吗？" query="请详细解释 Fervo Energy 使用的有机朗肯双循环（ORC）系统的工作原理及其实现零水耗的技术细节。"/>
### [16:17](https://www.youtube.com/watch?v=LL2abpmLhT4&t=977s) 地热能的固有冗余与“表后”数据中心布局
Tim Latimer 详细阐述了 Fervo Energy 如何通过独特的项目设计实现极其苛刻的“3个9”（99.9%）甚至更高级别的供电可靠性。当公司向客户提供 1 吉瓦（GW）规模的项目时，实际上并非依赖单一的巨型电厂，而是由 20 个相互独立的 50 兆瓦（MW）小型电厂组合而成。这种模块化的架构带来了无与伦比的固有冗余：即使其中一台透平机因例行维护而关停，或者某个电厂发生突发的意外停机，其余 19 个电厂依然能够稳定运行，确保整体电力供应不受实质性影响。
基于这种独特的冗余优势，Fervo Energy 正直接进军数据中心“同址共建”（Collocation）与“表后”（Behind-the-meter）供电市场。这种部署模式不仅能够实现经济、可持续且快速的建设，还能直接绕过耗时冗长的电网互联审批流程。更重要的是，这种模式不会将额外的电网升级成本转嫁给普通消费者，从而避免了当前因电价上涨而日益尖锐的政治争议和建设阻力。通过打破电网互联的时间瓶颈，公司正以更快的速度推动 AI 能源基础设施的落地。
- **冗余架构**：1 吉瓦的项目由 20 个独立的 50 兆瓦电厂组成，单点故障不影响整体供电。
- **商业合同**：过去几年中，Fervo Energy 已签订了 650 兆瓦的供电协议，代表着超过 70 亿美元的未来收入。
- **表后部署优势**：摆脱对电网互联（Grid Interconnection）等待时间的依赖，降低合规与政治风险，加快落地速度。
---
### [17:45](https://www.youtube.com/watch?v=LL2abpmLhT4&t=1065s) AI 反哺地热：微震监测与地下数据挖掘
Fervo Energy 不仅是在为 AI 提供电力，同时也在内部深度应用 AI 技术来突破地热开发的技术瓶颈。近百年来，地热行业最大的难题之一在于所有核心反应都发生在地表两英里（约 3.2 公里）深处，人们很难准确掌握地下井网的实际流体流动情况。为了攻克这一难题，Fervo 在整个井网中部署了光纤传感器平台，通过微震监测等手段实时捕捉地下流体动态。
这一光纤传感系统会产生海量的监测数据。仅在犹他州（Utah）的项目基地，光纤传感计划每天就会产生超过 3 TB 的数据，自项目启动以来已累计收集了超过 3 PB 的数据。在几年前，面对如此庞大的数据集，行业几乎无法有效处理；而如今借助现代 AI 工具，技术人员能够从数 PB 的数据中提取深层洞察，彻底解开了地热学界困扰了一个多世纪的地下流体流动之谜。
- **犹他州项目数据量**：每日新增超过 3 TB 监测数据，累计数据量已达 3 PB。
- **核心技术**：光纤传感器平台配合微震监测（Microseismic Monitoring），精准映射数十口井之间的流体流动。
---
### [18:50](https://www.youtube.com/watch?v=LL2abpmLhT4&t=1130s) 数字孪生与未来百吉瓦级地热蓝图
除了地下流体监测，Fervo Energy 还在将 AI 融入其业务的方方面面，包括建立自主开发的软件模拟操作系统、实现高级自动化、进行钻头损坏预测性维护（Bit Damage Monitoring），以及优化水源供应管理。 Latimer 强调，AI 的应用正在大幅拉低地热能的成本曲线，助力公司早日实现成为“地球上最便宜电力来源”的终极目标。
在合作创新方面，公司于上月宣布与英伟达（NVIDIA）及太平洋西北国家实验室（PNNL）达成三方合作。该项目将利用 PNNL 的外围级计算（Exascale Compute）能力与 NVIDIA 的强大芯片算力，构建高保真的全数字孪生（Digital Twin）系统，对地表数英里之下的复杂地质过程进行精确建模。在演讲的最后，Latimer 展望道，当前地热产业正处于早期爆发阶段，未来将扩展至 100 吉瓦以上的超大规模。他总结指出，AI 革命要持续推进，就必须具备可持续性、经济性、高可靠性与快速构建能力，而增强型地热正是解锁下一波 AI 飞跃的关键钥匙。
- **前沿合作**：联合 NVIDIA 与 Pacific Northwest National Laboratory（PNNL），利用外围级计算构建地下高保真数字孪生模型。
- **全流程 AI 赋能**：覆盖软件模拟操作系统、预测性维护、钻头损耗监控及水资源优化。
- **远景目标**：驱动增强型地热能向 100+ 吉瓦（GW）规模跨越，打造全球成本最低的清洁基荷电力。
## Notable quotes
> [02:21](https://www.youtube.com/watch?v=LL2abpmLhT4&t=141s) 智能的最终成本归根结底将折算为能源的成本。 — Sam Altman
> [02:21](https://www.youtube.com/watch?v=LL2abpmLhT4&t=141s) 下一波严苛的短缺将不再是芯片，而是电力。 — xAI 团队
> [02:21](https://www.youtube.com/watch?v=LL2abpmLhT4&t=141s) 当前的瓶颈已从 GPU 供应转向了电力供应。 — Satya Nadella
> [02:21](https://www.youtube.com/watch?v=LL2abpmLhT4&t=141s) 将能源转化为“智能”是目前所能想象到的最佳能源利用方式。 — Jensen Huang
> [00:04](https://www.youtube.com/watch?v=LL2abpmLhT4&t=4s) 人类社会的每一次重大进步与工业革命，本质上都与更清洁、更低廉、更容易获取的能源突破息息相关。 — Tim Latimer
> [18:50](https://www.youtube.com/watch?v=LL2abpmLhT4&t=1130s) AI 革命要持续推进，就必须具备可持续性、经济性、高可靠性与快速构建能力，而增强型地热正是解锁下一波 AI 飞跃的关键钥匙。 — Tim Latimer
## Mentioned
Tim Latimer, Fervo Energy, Sam Altman, OpenAI, xAI, Satya Nadella, 微软 (Microsoft), Jensen Huang, 英伟达 (NVIDIA), Greg Abbott, Google, Cape Station, 太平洋西北国家实验室 (PNNL)
## Video description
Tim Latimer - Co-Founder & CEO, Fervo Energy
The AI revolution has an energy problem. Every model trained, every inference run, every data center scaled demands massive, reliable power. I'll make the case that the companies who solve the energy problem will be the ones who win the AI race, and that geothermal is the only source that delivers the always-on, zero-carbon baseload that AI infrastructure actually requires. Attendees will leave with a new lens on AI's physical limits and a clear view of why enhanced geothermal is the most consequential energy bet of the next decade. If you're building AI and you haven't thought seriously about where the power comes from, this is the talk to attend.
[Source](https://www.youtube.com/watch?v=LL2abpmLhT4)
Source quotes
[2:21] Sam Altman: 智能的最终成本归根结底将折算为能源的成本。
[2:21] xAI 团队: 下一波严苛的短缺将不再是芯片，而是电力。
[2:21] Satya Nadella: 当前的瓶颈已从 GPU 供应转向了电力供应。
[2:21] Jensen Huang: 将能源转化为“智能”是目前所能想象到的最佳能源利用方式。
[0:04] Tim Latimer: 人类社会的每一次重大进步与工业革命，本质上都与更清洁、更低廉、更容易获取的能源突破息息相关。
[18:50] Tim Latimer: AI 革命要持续推进，就必须具备可持续性、经济性、高可靠性与快速构建能力，而增强型地热正是解锁下一波 AI 飞跃的关键钥匙。
Source quotes
[1:22] Sam Altman: the cost of intelligence will eventually collapse to the cost of energy.
[1:22] xAI: the next major technical shortage will be electricity.
[1:22] Jensen Huang: the best use of energy we can imagine.
Original transcript
[0:04] Excellent. Uh, thank you for the introduction
and I'm excited to be here uh at an at an
[0:08] AI conference. Um, I have to say I started the
company about eight years ago with our co-founder.
[0:13] We spun the company out of Stanford and I think
every other entrepreneur there was starting an AI
[0:18] company. And so I think we kind of felt as we were
leaving school, we said our goodbyes, we parted
[0:22] our ways, we were sort of the odd child doing
um, energy and here we are. we find ourselves
[0:28] um eight years later being kind of the AI company
because um energy and AI have become incredibly
[0:34] intertwined uh for reasons we're going to get into
and I I think a lot of our future of successfully
[0:40] commercializing developing AI driving productivity
for the world is going to rely on continued access
[0:46] to energy and uh energy is increasingly
playing a huge role in the AI buildout. Uh
[0:53] one very exciting development of FVO in the last
couple of months is we held our IPO in May. Uh
[0:59] we raised $2.2 billion in that a reflection of how
important new power generation buildout is to this
[1:06] uh world. So that's the upside. The downside is
I have to show this slide in every presentation
[1:10] that I give for the rest of my life. So quickly
read through that and understand the uh all the
[1:16] disclaimers. Um so that's the fun part about being
a public company. Um but energy is vital. Um,
[1:22] this is something that we thought of since uh
the beginning of FVO. Whenever you think about
[1:27] human progress, we believe it has always been very
intertwined with our access to energy. All of the
[1:32] industrial revolutions, all of the buildout of new
prosperity for the world has been underpinned by
[1:39] advances in energy technology that get cleaner,
more affordable, more accessible energy out there
[1:43] to the world. And um uh AI is just the next
wave of of this. It seems like anytime there's
[1:51] a pause in energy buildout, anytime there's a
stagnation in the economy, it's just because
[1:55] that next innovation hasn't happened yet. And
every single new innovation, we find creative new
[1:59] ways for energy to drive significant demand. And
I think that the next next wave of productivity,
[2:05] of improvement, of prosperity for humanity
is going to come from a huge belt out of AI.
[2:11] And just like every other industrial revolution
we've had over the last 200 years, that is going
[2:15] to have to be penned with an expansion of access
to better energy resources. And I'm not the only
[2:20] one saying this. We have kind of a cross-section
here of a lot of the leaders in AI talking about
[2:26] how important the world of of of energy access is
to their their um their business and operations.
[2:33] And so there's a couple things that have happened
here. uh going back to the world of cloud getting
[2:38] rolled out 101 15 years ago people started
realizing that if you put enough servers together
[2:42] it wasn't a small amount of energy it was a large
amount of energy and so the world of cloud and
[2:46] data centers started taking off about 15 years ago
but the largest data center in the world in 2010
[2:52] was a 30 megawatt data center which seems quaint
by today's uh um uh by today's standards you know
[2:59] that kind of doubled in size by 2013 in 2017 a
650 megawatt data center came on um and Now, we're
[3:06] talking about building 2 gawatt data centers at a
single campus. And this is where you're starting
[3:10] to run into constraints because there are very
few places in the world where you can plop down
[3:14] something that has uh you know 2 gawatts is kind
of the equivalent of the amount of energy that a
[3:19] city of 2 million people would use. And there are
very few places in the world where you can plop
[3:23] down that amount of energy quickly and cheaply and
affordably and and expand. And so the next phase
[3:28] of growth is going to be stuck on our ability to
access AI. And some of the thought leaders here,
[3:33] you know, Sam Alman famously told Congress that
the cost of intelligence will collapse to be the
[3:39] cost of energy eventually. Um, XAI stated that
the next big shortage will be electricity. Um,
[3:45] Satcha Nadal Satcha said that we're going to
we're going to be power constrained, not chip
[3:50] supply constrained. And Jensen talked about how
energy for intelligence is the best use of energy
[3:55] we can imagine. So, the demands on the energy
system are getting bigger and bigger and bigger.
[4:00] the growth is increasing and um and a lot all the
major players in the tech industry are starting
[4:07] to converge on this realization that if we want
to build out all the great applications you see
[4:11] when you walk through this exhibit floor, if we
want to make major advancements in the models,
[4:15] if we want to make major advancements in in
inference and the ability for users to actually
[4:19] get fast responses and competent responses from
AI and truly revolutionize the society we want,
[4:25] it will have to be underpinned by a massive
expansion in energy. And that's where we're
[4:29] running into constraint problems because
actually in the United States up until about
[4:33] uh two or three years ago our energy system had
been in a multi-deade period of stagnation. So
[4:39] if you actually look we have almost used the we
almost used the exact same amount of electricity
[4:44] in 2023 as we did at the turn of the century. Um,
and one of the things that means is that actually
[4:51] all of the muscle of growth, the ability to
build new infrastructure projects, build new
[4:55] transmission lines had really atrophied over the
last 20 years. Um, that's not something that we
[4:59] had figured out how to do in the United States in
a very long period of time. and restarting that
[5:04] growth engine has turned out to be exceptionally
challenging because a lot of the skilled workers
[5:08] needed to do the build the transformers and the
energy generation technologies you know didn't
[5:13] really exist anymore the workforce the supply
chain because there had not been growth for
[5:17] a 20 plus year period and we're now asking that
trend to reverse and restart overnight and it's
[5:22] even more challenging than that because right
when at the point when we're asking the grid
[5:26] to grow at its fastest rate ever we're actually
dealing with the challenge of a declining amount
[5:31] of reliable electricity in this You can see the
chart on the right where we talk about base load,
[5:36] the 24/7 reliable electricity needed to keep the
lights on is actually less now than it was um uh
[5:43] 15 years ago. And that trend is continuing to go
downward. And this is really important because
[5:47] unlike other sources of electricity load, data
centers need to operate 24/7. They need to have,
[5:52] you know, triple 9 59 reliability. And so we
need more 247 reliable electricity on the grid.
[5:59] And we're asking that growth to come at a time
when there has been no growth for a significant
[6:04] amount of time. And this is running into massive
constraints. You know, the the um depending on
[6:09] what forecast you want to believe over the next 10
years in the US. We have a grid that's about 1,000
[6:14] gawatts now. And we need to add somewhere between
100 and 250 gawatts of new electricity to the grid
[6:20] just to account for the growth in AI and data
center load. And that's more growth than we've
[6:24] done in decades. and restarting that engine has
proven to be really really challenging. Um what
[6:30] we see is that that conventional fossil fuel
generation, gas and coal are subject to fuel
[6:35] price volatility. They're subject to increasing
community backlash. You're seeing this over and
[6:40] over again. You know, we are from Texas, which
is famously sort of like the wild west of energy
[6:45] markets. Yes, there you go. I think I've got one
Texan in the room. Given it's an energy talk,
[6:49] there's probably a lot more Texans in the room,
but yeah, there we go. [laughter] But even Texas,
[6:55] which is famous for our ability to build anything,
permit anything, businessfriendly state, Governor
[7:00] Abbott uh just this week announced a moratorum
on new data center construction. You know,
[7:06] this happened in New York a couple months ago
and I think people said generally, you know,
[7:09] that that kind of fits with the political winds of
of what has happened in New York, but I think it's
[7:14] a huge wakeup call to the industry that even Texas
is saying we need to put a pause to data center
[7:20] buildout. And that's because you're running into
challenges where consumer prices are going up,
[7:25] emissions are going up. You're seeing more
and more of these energy resources built uh in
[7:29] communities where people don't think they're good
neighbors because of pollution, electricity cost,
[7:33] noise, and we have to find a better solution
if we're going to get there. And the solutions
[7:37] in front of us today don't really leave a lot of
great options. If our if our bill is to power AI
[7:42] with 247 reliable electricity and do it quickly
so that this revolution can have the impact that
[7:47] it needs to have, what you're seeing is extended
lead time on turbines for fossil fuel generation,
[7:52] as well as pollution problems and community
acceptance problems. What you see is solar and
[7:56] wind resources, while they have a certain value on
the grid, don't match the 24/7 energy generation
[8:01] profile that's needed for a data center operation.
new nuclear buildout is uh is a decade or more
[8:07] away and hydro power is incredibly geographically
limited. And so, you know, two of the solutions
[8:12] that people are turning to are growing right now,
benefiting uh very significantly from this market
[8:17] trend. But if you want to get uh a new gas turbine
today, you have to get in line for what's become
[8:22] a six-year wait time uh in order to get access
to one of those gas turbines. Additionally, new
[8:27] new nuclear buildout, the the last time we built
a nuclear power plant in this country, it went
[8:31] significantly over budget and took about a decade
longer than thought. So, new nuclear is also going
[8:35] to take a while to catch up. Meanwhile, people's
crisis on electricity bills, crisis on pollution,
[8:41] community acceptance challenges, water challenges
are all things that need to be solved in the next
[8:45] couple of years. And this is the advantage that
we see geothermal playing. This is a photo of
[8:50] what of our site in Utah, about a three-hour drive
from here, where we're talking about a new era of
[8:56] energy. Um, geothermal is not new. We've actually
had geothermal for over a hundred years. The first
[9:01] geothermal power plant in the world was developed
in um Italy around the turn of the last century in
[9:06] the 1900s. Um, but historically, geothermal has
been very very limited um to unique places where
[9:14] there's volcanic activity or something else that
brings that heat super close to the surface. And
[9:19] that has prevented the industry from growing. If
it was on the chart before, it would kind of have
[9:23] the same attributes of hydropower. It's 247. It's
carbon free. It has all those great attributes.
[9:27] But like hydropower, conventional geothermal of
drilling simple vertical wells has limited it to
[9:32] only working in very unique niche geologies around
the world. And so even though it's got this really
[9:37] promising all these really promising attributes
in terms of reliable carbon-f free electricity,
[9:42] it counts for less than 1% of global electricity
today because it's too geographically limited.
[9:47] What we've developed at Fervo is a way to
make geothermal work anywhere. And that's
[9:50] through advancements in drilling technology
where we drill deep horizontal wells, connect
[9:55] them through hydraulic fractures. And as a result,
we're not limited to only developing geothermal on
[9:59] top of volcanoes. We can expand the footprint of
geothermal very significantly. And we've unlocked
[10:05] this technology revolution right at the time where
there's been a revolution in the market in terms
[10:10] of new power demand. And so, um, maybe to explain
a little bit more how the technology works, we've
[10:16] got a nice video here. Before a single drill bit
hits the dirt, our work begins with data. We study
[10:21] the subsurface and build sophisticated models to
identify exactly [music] where the heat is. Once
[10:27] we've mapped the geology, we begin the drilling
process. [music] We drill over 10,000 ft down,
[10:34] turn, and continue drilling over a mile
horizontally through hot rock. A sister well
[10:39] is drilled in parallel, forming an injection well
and a production well. To protect shallow [music]
[10:46] groundwater, our wells are sealed behind multiple
layers of solid steel casing [music] and cement,
[10:51] ensuring our system is completely isolated from
the environment. Next, we connect them using
[10:58] multi-stage stimulation. We hydraulically fracture
to create a network of large fractures in the
[11:03] rock [music] so water can move reliably between
the two wells. This creates a massive amount of
[11:08] surface area, allowing the fluid to touch as
much hot [music] rock as possible to maximize
[11:12] heat recovery. And with real-time monitoring, we
can fine-tune the performance, optimizing flow,
[11:18] pressure, [music] and heat recovery. Now,
the cycle begins. We pump cool water down
[11:24] the injection well down hole. It flows through
these engineered pathways where the surrounding
[11:29] rock heats it up. That geothermal fluid returns
to the surface through the production well.
[11:36] At the surface, we turn heat [music] into
electricity using a binary organic rank
[11:41] and cycle system. The geothermal fluid passes
through heat exchangers, transferring its heat
[11:46] to a separate secondary [music] working fluid
selected for maximum efficiency. That working
[11:53] fluid drives a [music] turbine to generate power,
which is then cooled in an air cooled condenser
[11:58] and immediately looped back through the heat
exchangers. And here's the key. The geothermal
[12:04] fluid stays in a [music] closed loop. We reingject
100% of it back underground. No water lost to
[12:10] evaporation. Just a continuous [music] cycle, down
cold, up hot, again, delivering clean, carbon-f
[12:19] free energy from the earth. A process that's
repeatable, scalable, [music] and built to last.
[12:28] All right, that's the tech explanation. And the
key with this is that, you know, is that if you
[12:34] look at conventional geothermal, it has all these
great attributes, but it's limited geographically.
[12:39] And we've solved that geographic limitation
problem and turn geothermal into something that
[12:43] is truly scalable. And this is really important
because every single checkbox on there is really
[12:48] important these days. It's not just about speed to
power. It's not just about 247 reliability. It's
[12:52] not just about cost. More and more sustainability
and stewardship is part of this conversation. You
[12:57] know, why is Texas putting a moratorium on data
centers? Why are there more than 500 counties
[13:01] across the country that have put bans on data
centers? It's because people are concerned about
[13:05] the emissions and the traffic and the water
consumption that come with those. And so,
[13:09] you really need to be thinking about this from a
full picture, cost, reliability, sustainability
[13:14] if you're going to advance forward. And we believe
Fervo has that solution. And we're in a position
[13:19] to achieve multi-gawatt scale. Um, three years
ago we brought our first pilot project online,
[13:24] a project here in Nevada through a partnership
with Google um that has been producing power using
[13:29] this horizontal drilling technology for nearly
three years now. A huge accomplishment. And we've
[13:34] moved forward by building a 500 megawatt project
in southwest Utah, about a three-hour drive from
[13:39] here. So, this is Cape Station. When we talk about
speed to power, this is what we're talking about.
[13:44] This was October 2024. We finally got our permits
to do first ground clearing of the first power
[13:49] plant. And one year later that power plant was
complete. And so this is a very fast solution.
[13:55] You can see the well pads right in the background
where we provide that geothermal brine to this
[13:58] facility. We are starting in Utah with this 500
megawatt project, but just project Cape can scale
[14:04] up to 4 gawatt. A huge project in of itself. We
have an acreage position that consists of over
[14:09] 600,000 acres across the western United States.
Meaning that nearly 50 gawatts of potential is
[14:14] something already under Furbos's control that
we're advancing through our project pipeline
[14:18] to build out. You know, we individually as one
company want to put a huge dent in that 100 gawatt
[14:23] problem that I discussed earlier in terms of data
center uh energy demand and we want to do it in a
[14:28] clean, sustainable, reliable, cost-effective way.
And so we're going to be expanding throughout the
[14:32] entire western United States as we continue to
scale. We're also a part uh on early in a new
[14:38] curve of a technology revolution. Costs are d are
dropping dramatically because of the innovations
[14:43] we brought to the table. I mentioned that pilot
project we did in 2023. For some scale here,
[14:48] we've could put what that well looks like
in terms of its depth and horizontal length
[14:52] and the temperature that we achieved in that
well compared to the Empire State Building,
[14:55] which is a great reference point for height. You
can see that first well that we drilled uh was
[14:59] nearly 9,000 ft deep with a 3,2 foot lateral. That
well was capable of producing about 3 megawatts
[15:05] of electricity. A great starting point for the
world's first ever horizontal geothermal well,
[15:10] but just the beginning of where we wanted to go in
the tech road map. And much like we've we've seen
[15:14] with hydraulic fracturing technology, the oil gas
industry or solar or wind technology or battery
[15:19] technology, we are just early on that learning
curve and driving costs lower and lower. The Cape
[15:24] Phase 1 design, we we went even deeper drilling
nearly 10,000 ft with a 5,000 foot lateral. And
[15:29] recently we finished our design 3.0, a 15 megawatt
project or 15 a well designed to deliver 15
[15:35] megawatts of electricity drilled to 12,000 ft
and over a mile and a half in lateral length.
[15:40] So a huge improvement, a 5x improvement in power
output. The other thing is we're drilling faster
[15:44] and faster as we do this. So the first ever well
that we drilled, it was the first time anybody had
[15:48] drilled these extreme high temperatures, hard
rock granite with a horizontal drilling design
[15:52] anywhere in the world, and it took us 70 days.
This new well, which is nearly double the length
[15:56] of our initial pilot wells, um we're drilling
in just 21 days. And so this is quickly becoming
[16:01] the most cost effective solution for electricity
generation. We also are huge believers in scale
[16:07] up and modularity. So you see here we build our
projects in 50 megawatt power blocks. That's huge
[16:12] because we can get on a learning curve um drive
innovations to drive costs lower. But it's also
[16:17] really important because if you want to talk about
delivering triple 9 redundancy to project sites,
[16:21] this is a key technology that enables that because
if you think about our our projects when we offer
[16:27] a customer a 1 gigawatt project, it's actually 20
independent 50 megawatt power plants. That means
[16:32] if you pull one turbine off for maintenance,
the other 19 are still operating. If there's
[16:36] an unplanned outage at one of the power plants,
the other 19 are still operating. When we work
[16:41] with our customers who are looking for triple 9
reliability or higher to power their data centers,
[16:45] this level of inherent redundancy that's unique
to the way we build out geothermal technology is
[16:50] critically important. And our ability to offer
multi-gawatt clusters of redundant 50 megawatt
[16:55] power plants is a huge differentiator in this
space. And one of the reasons we're moving direct
[16:59] directly into the market of colllocating data
center load with our projects so that we can build
[17:04] them out in a way that is affordable, sustainable,
sustainable, fast, doesn't rely on lengthy grid
[17:10] interconnections, and importantly does not pass
extra cost on to consumers, which is becoming more
[17:15] and more a political touch point that is leading
to a slower buildout. And so this is our our
[17:19] vision. We are working aggressively on expanding
our existing customer relationships. I showed you
[17:24] earlier we've contracted 650 megawatts of power um
over the last several years which represents over
[17:31] 7 billion of revenue with our ability to unlock
behind the meter. We're no longer constrained by
[17:36] grid interconnection times. We have a unique
product that solves all of the challenges to
[17:39] build out AI expansion and we're very excited to
be moving forward at a much faster rate today.
[17:45] Not only are we serving the AI industry through
providing power to data centers, but we also have
[17:49] advanced our technology significantly through
adoption of AI of of AI internally. One example
[17:55] of this is through microismic monitoring.
We have fiber optic sensors throughout our
[17:59] well fields. One of the big mysteries for over a
century that has kept geothermal fields the field
[18:05] of geothermal from advancing is when all of the
action takes place 2 miles below your feet, it's
[18:09] very difficult to figure out where are the actual
wells flowing, what's actually happening. And so
[18:13] our fiber optic sensing platform can actually
determine flow. And what we've mapped here is
[18:18] where the fluid flow is going across dozens of
wells. And the key here is this is something that
[18:23] generates enormous amounts of data. At our field
site in Utah, just the fiber optic sensing program
[18:28] generates over 3 terabytes of data per day. We've
collected three pabytes of data since starting
[18:32] this operation. A few years ago, this this data
would have been incomprehensible. But being able
[18:37] to use AI tools to sift through pabytes worth of
data to actually extract insights has mean we've
[18:43] been able to solve some of the mysteries that have
existed in geothermal for over a century in terms
[18:46] of what's actually happening in the subsurface.
And that's just one of the many areas that we're
[18:50] advancing through things like simulation and
modeling through things like building out our
[18:55] own operating system of software simulation tools
for advanced automations for things like doing
[19:00] predictive maintenance or bit damage monitoring
through AI tools and workflows to monitoring our
[19:04] our water supply optimization. AI not only are is
our role in AI to provide power to it, but we're
[19:10] actually being able to use it to accelerate our
goal to become the cheapest form of electricity
[19:14] on the planet. One of the things we announced just
last month was a three a partnership with Nvidia
[19:19] and the Pacific Northwest National Lab to build
out a full digital twin to actually be able to
[19:23] model using the technologies and exoscale compute
at Pacific Northwest National Labs and the chip
[19:30] um capabilities of Nvidia to actually model with
high fidelity digital twins of what's happening
[19:35] miles below the feet of beneath our feet. So AI is
accelerating our cost curve very dramatically. And
[19:41] what we're looking at here is is the early days
of an industry that will scale to um 100 gigawatts
[19:47] plus. We think that in order for AI to continue
expanding, it must get the right buyin from the
[19:54] communities that they work in by being sustainable
and affordable. It must be reliable and it has to
[19:58] be built fast because this revolution is happening
now. And we think geothermal is the thing that's
[20:02] going to unlock the next wave of advancements
in AI. So, thank you for your time. [applause]
