AI and Geothermal: Why Startups Are Suddenly Talking About Baseload Power
Something unusual has been happening in Silicon Valley pitch meetings. Founders who a few years ago would have been talking about SaaS metrics and API integrations are now talking about drilling depths, reservoir temperatures, and capacity factors. The reason is simple: artificial intelligence has made energy a startup category again — and geothermal energy is emerging as one of the most compelling solutions to AI's insatiable demand for clean, reliable, always-on power.
On TBPN, John Coogan and Jordi Hays have been tracking this convergence closely. The thesis is straightforward: AI workloads have fundamentally different energy requirements than traditional computing, and those requirements are creating a massive opening for geothermal startups. This post unpacks why baseload power matters, how AI and machine learning are actually improving geothermal technology, and where the startup opportunities lie.
The Baseload Problem: Why AI Cannot Run on Sunshine Alone
To understand why geothermal matters for AI, you first need to understand the baseload problem.
Traditional office buildings and web servers have variable power demand. Traffic spikes during the day, drops at night. People go home. Servers can be powered down or consolidated. This demand pattern is well-matched to solar energy, which peaks during the day, and wind energy, which is variable but can be forecasted and balanced with battery storage.
AI training workloads are fundamentally different:
- 24/7 operation: A large training run operates at maximum power draw around the clock for weeks or months. There is no "off-peak" period. GPUs running at full utilization consume the same power at 3 AM as they do at 3 PM.
- No tolerance for interruption: If power drops — even briefly — a training run can lose hours of progress. Checkpointing helps, but the recovery overhead is significant. Intermittent power sources create risk.
- Massive scale: A single large training cluster consumes 50 to 150 megawatts continuously. At that scale, you need a power source that delivers consistently, not one that requires 4x overbuild to average out to your needs.
- Growing demand: Every generation of AI model is larger and requires more compute to train. Power demand is not plateauing — it is accelerating.
The Intermittency Gap
Solar panels generate electricity about 20 to 30 percent of the time (capacity factor) in the best US locations. Wind turbines manage 30 to 45 percent. To provide 100 MW of continuous power from solar alone, you would need approximately 400 MW of installed capacity plus 200 to 300 MWh of battery storage — and even then, multi-day weather events (cloudy, still periods) can deplete reserves.
Batteries are improving rapidly, but lithium-ion storage at the scale needed for multi-day backup at 100+ MW is extremely expensive — billions of dollars — and poses supply chain risks. Long-duration storage technologies (iron-air, flow batteries, compressed air) are promising but not yet commercially proven at the scale AI requires.
Nuclear: The Obvious Solution That Takes Too Long
Nuclear power is the gold standard for baseload generation. It runs at 90+ percent capacity factor, produces zero carbon emissions during operation, and a single plant can generate 1,000+ MW for 40 to 60 years. Tech companies know this — hence Microsoft's Three Mile Island restart, Amazon's utility partnerships, and Google's exploration of small modular reactors (SMRs).
The problem is time. Building a new nuclear plant in the US takes 10 to 15 years from conception to operation, factoring in licensing, environmental review, construction, and commissioning. SMRs promise faster timelines, but no commercial SMR is operating in the US as of 2026. For AI companies that need power in 2 to 5 years, nuclear is a long-term solution, not a near-term one.
Enter Geothermal: The Dark Horse
Geothermal energy taps heat from the earth's interior to generate steam and drive turbines. It has been used for electricity generation since the early 20th century, primarily in volcanically active regions (Iceland, Indonesia, the Philippines, parts of California). What makes it suddenly relevant to the AI conversation is a combination of technological advances and demand-side urgency.
Why Geothermal Fits AI Perfectly
- 24/7 baseload power: Geothermal plants operate at capacity factors of 90 to 95 percent — comparable to nuclear, far superior to solar or wind. The earth's heat does not stop at night or during cloudy weather.
- Zero carbon emissions: Geothermal plants produce minimal to zero greenhouse gas emissions, depending on the technology. This aligns with tech companies' climate commitments.
- Small physical footprint: A 50 MW geothermal plant occupies a fraction of the land required for an equivalent solar farm. This matters when data centers need power close to the facility.
- Scalability: Modern enhanced geothermal systems (EGS) are not limited to volcanic regions. With advances in deep drilling and reservoir engineering, geothermal can theoretically be deployed anywhere — you just need to drill deep enough to reach hot rock.
- Speed of deployment: A geothermal well can be drilled and brought online in 1 to 3 years — much faster than nuclear, competitive with natural gas plants, and without the emissions.
How AI and Machine Learning Are Improving Geothermal
In a beautiful feedback loop, the AI industry is not just creating demand for geothermal energy — it is also providing tools that make geothermal exploration and development dramatically more effective.
Seismic Data Analysis for Site Selection
Identifying viable geothermal sites has historically been expensive and uncertain. It requires analyzing seismic data, geological surveys, and temperature gradient measurements to predict where subsurface heat is accessible. Traditional methods rely on manual interpretation by geologists, which is slow and limited by human pattern recognition.
Machine learning is transforming this process. Companies are training models on historical geothermal exploration data — thousands of wells, seismic surveys, and geological assessments — to predict geothermal potential with far greater accuracy and speed. AI can identify subtle patterns in seismic data that human analysts miss, reducing the number of expensive exploratory wells needed to confirm a site.
Drilling Optimization
Drilling is the largest cost component of geothermal development — typically 40 to 60 percent of total project cost. Geothermal wells must reach depths of 2 to 10 kilometers, through hard, hot rock formations that destroy conventional drill bits and damage equipment.
ML-driven drilling optimization uses real-time sensor data (torque, temperature, pressure, vibration) to predict rock formations ahead of the drill bit, optimize drilling parameters (weight on bit, rotation speed, mud flow), and anticipate equipment failures before they cause costly downtime. Companies report 20 to 40 percent reductions in drilling time using these techniques.
Reservoir Modeling
Once a geothermal reservoir is identified, engineers must model how heat will be extracted over time — flow rates, temperature decline, and optimal well placement. Traditional reservoir modeling uses physics-based simulations that are computationally expensive and limited in the scenarios they can explore.
AI-powered reservoir models combine physics simulations with data-driven approaches, enabling engineers to run thousands of scenarios in the time it previously took to run a handful. This leads to better well placement, optimized extraction rates, and more accurate predictions of long-term reservoir performance.
Key Companies Driving the Geothermal Renaissance
Fervo Energy
Fervo Energy is arguably the most prominent geothermal startup, having secured a power purchase agreement with Google for its project in Nevada. Fervo uses horizontal drilling techniques borrowed from the oil and gas industry to create enhanced geothermal systems (EGS). Rather than relying on natural hot springs or steam reservoirs, EGS creates artificial reservoirs by injecting water into hot rock formations and extracting the heated water through a second well.
Fervo's approach is significant because it dramatically expands the geographic range of geothermal. Traditional geothermal is limited to regions with naturally occurring hydrothermal resources. EGS can work anywhere the subsurface temperature is high enough — which, given enough depth, is essentially everywhere.
Quaise Energy
Quaise Energy is taking a radically different approach to deep drilling. Instead of conventional mechanical drilling (rotating a bit against rock), Quaise is developing millimeter-wave drilling — using directed energy from a gyrotron (the same technology used in plasma physics) to vaporize rock. This could enable drilling to depths of 20 kilometers, where temperatures exceed 500°C regardless of location. At those temperatures, the energy potential is enormous — a single deep well could generate tens of megawatts.
Quaise is earlier-stage than Fervo, but the implications of their technology are transformative. If millimeter-wave drilling works at commercial scale, geothermal becomes a truly universal energy source.
Eavor Technologies
Eavor Technologies has developed a closed-loop geothermal system that functions like a giant underground radiator. Rather than injecting water into open rock formations (which raises concerns about induced seismicity and water contamination), Eavor drills a sealed loop of connected wells through hot rock. A working fluid circulates through the loop, absorbing heat and returning to the surface to generate electricity or provide direct heating.
The advantage of closed-loop systems: no fracking, no water consumption, no induced earthquakes, and predictable performance. The trade-off: lower power output per well compared to EGS, because heat exchange is limited to conduction through the pipe walls rather than direct contact with the rock.
Startup Opportunities in the Geothermal-AI Nexus
The convergence of AI demand and geothermal innovation creates several categories of startup opportunity:
Geothermal Exploration Software
Building AI-powered platforms that analyze geological data, seismic surveys, and well logs to identify geothermal prospects. Think of it as "Palantir for geothermal" — integrating disparate data sources into a unified model that predicts where to drill and how much energy to expect. The market for this is growing as developers seek to reduce exploration risk.
Co-Location: Data Center + Geothermal Plant
The most capital-efficient configuration for AI power may be building a data center directly adjacent to a geothermal plant. This eliminates transmission losses (which can be 5 to 10 percent over long distances), reduces grid interconnection costs, and provides dedicated, reliable power. Several startups are exploring this co-location model, particularly in geothermal-rich regions like Nevada, Utah, and Oregon.
Energy Trading Platforms
As data centers become major energy consumers, they need sophisticated tools to manage power procurement: forecasting demand, negotiating PPAs, trading renewable energy credits, and optimizing across multiple power sources. Startups building energy management platforms specifically for data center operators are addressing a growing need.
Advanced Drilling Components
Geothermal drilling requires specialized equipment that can withstand extreme temperatures and pressures. High-temperature electronics, advanced drill bit materials, and downhole sensors are all areas where startups can innovate. The oil and gas industry's supply chain provides a foundation, but geothermal's unique requirements (harder rock, higher temperatures, deeper wells) create opportunities for purpose-built solutions.
Heat Reuse Systems
Data centers produce enormous amounts of waste heat. Geothermal plants can be designed to capture and reuse this heat for district heating, industrial processes, or agricultural applications (greenhouses). Startups building heat reuse systems that connect data center waste heat to local demand are creating value from what would otherwise be wasted energy.
Why Energy Literacy Is Becoming a Founder Skill
As TBPN has emphasized repeatedly, the days of founders blissfully ignoring energy are over. If you are building an AI company in 2026, understanding energy is as important as understanding software architecture. Here is why:
- Cost: Energy is now the single largest variable cost for many AI operations. Founders who can optimize energy procurement save millions and improve unit economics.
- Availability: Compute is limited by power, not by GPU supply. Founders who can secure power — through creative PPAs, co-location, or alternative energy sources — gain a competitive advantage in compute access.
- Sustainability: Customers, investors, and regulators are increasingly asking about the carbon footprint of AI products. Founders who can demonstrate clean energy sourcing have an advantage in enterprise sales and fundraising.
- Long-term planning: Energy infrastructure decisions have 10- to 20-year consequences. Founders who understand energy markets can make better strategic decisions about where to locate, how to scale, and when to invest in on-site generation.
If you are the kind of founder who digs into the details — the full stack, from silicon to substation — you are our kind of people. Rep the builder mindset with TBPN mugs and stickers for your laptop, workstation, or hard hat.
The Bigger Picture: Energy as the AI Battleground
Zoom out, and a clear picture emerges: the AI race is increasingly an energy race. The companies and countries that secure abundant, cheap, reliable power will train the largest models, deploy the most inference capacity, and capture the most value from artificial intelligence.
Geothermal is not the only answer — nuclear, natural gas, solar-plus-storage, and even fusion will all play roles. But geothermal occupies a unique sweet spot: clean, baseload, scalable, and deployable on a timeline that matches AI demand. The startups and investors who recognize this early will be well-positioned as the AI infrastructure buildout continues to accelerate.
The earth has been generating heat for 4.5 billion years. We are just now learning to use it to power the intelligence revolution.
Frequently Asked Questions
Is geothermal energy actually scalable enough for AI data centers?
Traditional hydrothermal geothermal (tapping natural steam reservoirs) is limited to specific geographic regions. However, enhanced geothermal systems (EGS) and deep drilling technologies are dramatically expanding the addressable market. The US Department of Energy estimates that EGS could provide over 100 GW of baseload power in the US alone — more than enough to support projected AI data center demand. The challenge is not resource availability but drilling cost and speed, both of which are improving rapidly.
How does geothermal compare to nuclear for powering AI?
Both offer zero-carbon baseload power with high capacity factors. Nuclear provides higher power density (1,000+ MW per plant vs. 5-50 MW per geothermal site) but takes 10-15 years to build and faces regulatory challenges. Geothermal can be deployed in 1-3 years at smaller scale and faces fewer regulatory hurdles. For AI data centers, geothermal is a near-term solution (2026-2030) while nuclear is a long-term solution (2030+). Many operators will likely use both, with geothermal bridging the gap until nuclear capacity comes online.
What are the environmental risks of enhanced geothermal systems?
The primary concern with EGS is induced seismicity — small earthquakes caused by injecting water into rock formations at high pressure. Several EGS projects have been paused or modified due to seismic events. However, closed-loop systems (like Eavor's) eliminate this risk entirely, and EGS operators are developing protocols for seismic monitoring and pressure management that significantly reduce risk. Water usage is another consideration, though most EGS systems recirculate water in a closed loop, consuming far less than evaporative cooling towers.
Can startups actually compete in the energy space, or is it dominated by incumbents?
The geothermal space is unusually open to startups because the technology is evolving rapidly and incumbents (large utilities, oil and gas companies) have been slow to invest. Fervo Energy, Quaise, and Eavor are all venture-backed startups that have secured major partnerships and projects. The AI demand driver is accelerating timelines and bringing in capital from tech investors who understand the opportunity. That said, energy is a capital-intensive, regulated industry — startups need patient capital and regulatory expertise to succeed.