Geothermal Energy for Data Centers: Why Big Tech Is Going Underground

Something quiet is happening underneath the AI boom. While most of the conversation about data center energy focuses on solar farms, grid strain, and whether nuclear can be revived fast enough, a handful of the world's largest tech companies have been quietly signing deals that point in a very different direction — straight down.

Geothermal energy, for most of its history, has been a niche technology. Useful in Iceland. Great near volcanic hot spots. Not something you'd expect to solve the electricity needs of a hyperscale data center in Nevada or New Mexico.

That picture is changing fast.

Google signed a deal with geothermal startup Fervo Energy in 2021 — the first corporate power purchase agreement for next-generation geothermal in history. That project went operational in 2023. In February 2026, Google signed a follow-on deal for up to 150 MW of geothermal power through Nevada utility NV Energy. Meta has inked two separate agreements for a combined 300 MW. Microsoft has plans for a geothermal-powered data center campus in Kenya.

These aren't small pilots. They're serious procurement deals at serious scale, and they're happening because something has gone quietly wrong with the energy math of running AI.

The Cooling Crisis Nobody's Talking About

Here's a number that doesn't get nearly enough airtime: according to the IEA, cooling accounts for somewhere between 7% and more than 30% of a data center's total electricity consumption, depending on how efficiently the facility is designed. For less-optimized enterprise facilities, cooling alone can eat up nearly a third of total power draw.

Now layer on what AI is doing to the heat problem.

Traditional servers generate heat. AI servers — the GPU-dense racks running large language models — generate a lot more heat, packed into much smaller spaces. Rack-level power densities that used to top out at 10-15 kilowatts are now routinely hitting 50-100 kW for AI workloads, and some next-generation liquid-cooled configurations are pushing higher. When you pack more heat into less space, the cooling systems have to work proportionally harder.

The DOE's Lawrence Berkeley National Laboratory reported that U.S. data centers consumed about 4.4% of total U.S. electricity in 2023, with projections of 6.7% to 12% by 2028. Globally, the IEA estimated data center electricity demand at 415 TWh in 2024 — roughly 1.5% of global electricity — and projects it could nearly double to around 945 TWh by 2030 in a base-case scenario.

That's not a gradual trend. That's a structural shift in global electricity demand driven almost entirely by AI.

And the companies building and operating these facilities know it. Which is why the energy procurement strategies of Google, Meta, and Microsoft have gotten dramatically more interesting in the last three years.

PUE: The One Number That Explains Everything

To understand why geothermal is attractive to hyperscalers, you first need to understand PUE. Power Usage Effectiveness is a simple ratio: total facility power divided by IT equipment power. A PUE of 1.0 would be theoretically perfect — every watt going straight to computing with zero overhead. A PUE of 2.0 means for every watt running the servers, you're burning another watt on everything else — cooling, lighting, power distribution losses.

The industry average isn't great. Uptime Institute puts the 2023 average at 1.58 — and that number has been largely flat since around 2020.

Google reports a 2024 fleet average PUE of 1.09. The gap between those two numbers is enormous in practice. Do the math on a 100 MW IT load: at PUE 1.58, you're drawing 158 MW total — 58 MW overhead. At PUE 1.09, you're drawing 109 MW total — 9 MW overhead. That's a difference of 49 MW of continuous overhead load, running 24 hours a day, 365 days a year.

49 MW. Continuously. The scale of that overhead gap is why data center operators will spend enormous amounts of capital on anything that might move the needle on cooling efficiency.

Two Paths Underground: Power vs. Direct Cooling

Geothermal can serve data centers in two fundamentally different ways, and it's worth being precise about the distinction because they're at very different stages of commercial maturity.

Path one is geothermal power generation. Geothermal plants — including next-generation enhanced geothermal systems — generate electricity that gets delivered through the grid or directly via power purchase agreements. The data center buys that electricity the same way it buys any other electricity, except the source is a geothermal field instead of a gas turbine or wind farm. The specific advantage here is firmness: unlike wind and solar, geothermal power runs 24 hours a day, 365 days a year, regardless of weather. For a hyperscale data center that never turns off, that's not a minor detail — it's the whole value proposition. This is the pathway where the big deals are happening right now.

Path two is direct geothermal cooling. Instead of generating electricity, you use geothermal resources — or underground thermal storage — to directly offset cooling loads. The DOE has described concepts like Cold Underground Thermal Energy Storage (Cold UTES) and mine-water cooling for data centers, where subsurface thermal resources are used to shift or reduce peak cooling demand rather than generate power. This pathway is genuinely promising but it's earlier-stage — more demonstration projects and pilot programs than the MW-scale commercial deployments we're seeing on the power generation side.

The companies making headlines right now — Google, Meta, Microsoft — are primarily pursuing path one. They want firm, clean megawatts to run their operations. The direct cooling pathway may eventually layer on top of that, reducing what little overhead load remains, but it's a different story with a different timeline.

What Enhanced Geothermal Actually Is

Traditional geothermal power plants require very specific geology — you need naturally occurring hot water or steam at reachable depths. That limits the technology to places like Iceland, the Geysers in California, or parts of Nevada where volcanic heat is close to the surface.

Enhanced geothermal systems (EGS) break that constraint. The basic idea: drill deep enough anywhere on earth and you'll find hot rock. EGS involves drilling two or more wells into that hot, dry rock, then creating or enhancing fracture networks between them so that water injected down one well heats up as it moves through the rock and comes back up the other well as steam or hot fluid, which can drive a turbine.

The challenge is engineering. Creating the right fracture network without triggering unwanted seismic activity, maintaining flow rates, keeping the wells productive over years — these are hard problems. They're the reason EGS has been a research project for decades without becoming a commercial reality.

What changed is that the oil and gas industry got very good at horizontal drilling and hydraulic fracturing for tight formations. Companies like Fervo Energy realized that the same techniques — and critically, the same workforce and supply chain — could be applied to EGS. Horizontal well pairs, fiber-optic sensing in the wellbore, advanced subsurface analytics: tools that the oil patch refined over 20 years of shale development are now being pointed at hot rock instead.

That's the technology leap that made Google's 2021 deal possible. And it's what's enabling the scale-up happening right now in Utah.

If you're new to geothermal generally, our guide on how geothermal heat pumps work covers the fundamentals — but the EGS story for power generation is a different application of the same underlying principle: the earth holds enormous heat, and the question is just how to access it.

Google Goes First: The Fervo Partnership

In 2021, Google announced what it called the first corporate power purchase agreement for next-generation geothermal energy — a deal with Houston-based Fervo Energy to develop a project in Nevada. The framing was future-oriented: this was about proving the technology worked at commercial scale, not just producing megawatts.

Two years later, it worked. In 2023, Google reported the project operational, delivering carbon-free electricity to the local Nevada grid that serves its data centers.

That might sound underwhelming — "the project worked" — but the significance was hard to overstate. This was the first time a next-generation EGS-linked project had gone from corporate deal to operational status. It de-risked the technology in a way that no amount of lab research or pilot demonstrations could. Google had put real money on the line, and the power came out the other end.

The implications for every other hyperscaler were obvious. If Google could pull this off in Nevada, the question was no longer "can this work" but "how fast can we scale it."

Google also hasn't stopped at Nevada. In 2025, Google signed what it described as the first geothermal corporate PPAs in Taiwan, with initial 10 MW projects through Baseload Capital, opening up the Asia-Pacific market for this kind of deal structure.

The Scale-Up: Nevada's 150 MW Bet

The Fervo proof-of-concept was one thing. What happened in February 2026 is something else entirely.

Ormat Technologies announced the signing of a geothermal portfolio PPA of up to 150 MW through NV Energy's Clean Transition Tariff — specifically to support Google's Nevada data center operations. Target online window: 2028 to 2030, subject to regulatory approval.

150 megawatts. Firm, baseload, carbon-free power. For context: that's enough to power a small city. And it's structured through a utility clean energy tariff rather than a direct corporate PPA, which means it uses an existing regulatory mechanism designed to let large customers procure clean energy without forcing costs onto other ratepayers. The model is, in Ormat's framing, repeatable.

That last word — repeatable — is probably the most significant thing about this deal. The first Fervo project was a proof of technology. This is a proof of market structure. If a utility tariff pathway for large-scale geothermal procurement can work in Nevada, it can work in other states with active geothermal resources.

Meta Joins In — Twice

Meta hasn't gotten as much coverage for its geothermal moves as Google, but the scale of what the company has committed to is comparable — and arguably more aggressive, given that both deals are still in development rather than partially operational.

In August 2024, Meta announced a partnership with Sage Geosystems for up to 150 MW of geothermal baseload power, with the first phase targeted for 2027. Sage focuses on a closed-loop pressurized geothermal design — a somewhat different technical approach from Fervo's horizontal EGS wells, but aimed at the same fundamental goal: turning hot rock into firm, dispatchable electricity.

Then in 2025, Meta signed a second agreement — this time with XGS Energy, for 150 MW in New Mexico, structured in two phases projected to come online by 2030.

Add those up: 300 MW of contracted geothermal capacity across two developers and two states, from a company that rarely announces energy deals loudly. Meta's AI infrastructure buildout is enormous, and the company has quietly decided that geothermal needs to be a meaningful part of how it powers that infrastructure.

The geographic diversification here matters too. Nevada, New Mexico — these aren't just random locations. They're states with known geothermal resources, favorable permitting environments, and increasingly, the growing supply chain and workforce to support development. The pattern looks less like a few speculative bets and more like a deliberate portfolio strategy.

Microsoft, Kenya, and the Olkaria Wildcard

The most geographically unusual piece of this story involves Microsoft, its partner G42, and the Olkaria geothermal field in Kenya's Rift Valley.

In May 2024, Microsoft and Abu Dhabi-based AI company G42 announced a $1 billion digital ecosystem initiative for Kenya. Buried in the announcement was a detail that caught geothermal observers' attention: the planned data center campus at Olkaria would run entirely on renewable geothermal energy, with a new East Africa cloud region targeted within 24 months of definitive agreements.

Olkaria is one of the most productive geothermal fields in Africa. Kenya generates more than 40% of its electricity from geothermal — a remarkable figure for any country — and Olkaria has been the backbone of that supply for decades. For a data center, it offers something rare in Africa: abundant, carbon-free, baseload power with decades of operational history.

The timing of the project remains execution-dependent — the announcement described an initiative and LOI framework rather than finalized agreements. But the strategic logic is clear: geothermal-powered data centers in geothermal-rich regions aren't just about sustainability optics. They're about reliable power supply in markets where grid reliability is a real operational risk.

Kenya is an extreme version of a broader argument: for data center siting decisions in geothermal-rich regions globally, the power supply advantage of geothermal isn't marginal. It can be the whole reason the project is viable.

Cape Station: The World's Biggest EGS Project

If Google's Nevada project was the proof-of-concept for next-generation geothermal, Fervo's Cape Station project in Utah is the scale-up everyone has been waiting for.

Fervo describes Cape Station as the world's largest EGS development. The numbers have been growing steadily: originally announced at 400 MW, it was upsized to 500 MW, with Fervo stating Phase I at 100 MW operational in 2026, Phase II adding 400 MW by 2028, and permitting work underway for a potential 2 GW total site capacity.

In April 2025, Fervo announced a 31 MW, 15-year power purchase agreement with Shell Energy for Cape Station output — a deal notable not just for the megawatts but for the counterparty. Shell isn't a tech company hedging its sustainability commitments; it's an energy major that has made calculated bets on where power markets are going. When Shell takes a 15-year position on an EGS project, that's a signal about commercial viability that carries weight independent of any single tech company's procurement strategy.

Fervo secured new project financing in 2025 to accelerate Cape Station development. The technical approach — horizontal well pairs using oil-and-gas-derived drilling techniques, fiber-optic downhole sensing, and advanced reservoir analytics — is the same playbook that made the Nevada project work, now applied at a scale the industry hasn't seen before.

If Cape Station delivers on its timeline, geothermal goes from a promising niche to a mainstream resource option for power-hungry industrial loads. Data centers are the obvious first market, but the implications extend to any large electricity consumer that values firm clean power — which is increasingly everyone.

Iceland and the Geothermal Hubs Already Running

Not every geothermal data center story involves next-generation EGS or major corporate deals in development. Some of them are just... running.

atNorth markets its Iceland data centers on a simple proposition: the grid is already powered primarily by geothermal and hydro. You don't need a special deal or a new development project. You locate the data center in Iceland, you plug into the grid, and you're running on clean baseload power with competitive electricity costs and a cold climate that helps with cooling overhead.

Crusoe, a U.S.-based cloud company, has been expanding its Iceland data center capacity on exactly this basis — citing the combination of geothermal and hydro-backed power supply as a key driver. Iceland's PUE numbers at these facilities can be very low by industry standards, partly because the ambient cold helps with cooling and partly because the power supply is clean and reliable enough to run lean on infrastructure overhead.

Iceland's model is interesting precisely because it's not futuristic. It's a working example of what geothermal-powered digital infrastructure looks like at scale, today, without needing new technology to mature. The constraint is geography — Iceland has a small population and limited land for massive campus-scale deployments. But as a proof of concept that geothermal can underpin serious digital infrastructure, it's been running for years.

The Cost Picture

The honest answer to "what does geothermal power for a data center cost" is: it depends, and most of the actual deal terms are confidential.

The commercial agreements Google has signed with Fervo and through NV Energy-Ormat haven't disclosed $/MWh strike prices. Meta's deals with Sage and XGS similarly haven't disclosed pricing. This is pretty standard for large industrial power agreements — the terms are negotiated privately and the specifics stay private.

What we do have are cost targets and benchmarks.

The DOE's Enhanced Geothermal Shot has set a target of $45/MWh for EGS power by 2035 — a roughly 90% cost reduction from the technology's baseline. That target is explicitly meant to make EGS cost-competitive with other clean firm power sources at utility scale. Whether it's achievable on that timeline is debated, but it establishes the direction: the DOE believes EGS can reach prices that make it broadly competitive without subsidies.

For comparison: wind and solar have come down dramatically in cost over the past decade, but their intermittency creates real problems for loads that can't be curtailed. Data centers can't tell their AI training jobs to pause because the wind isn't blowing. The "clean firm" power premium — what you pay for power that's both carbon-free and always on — is real, and it's part of what makes geothermal's cost calculus different from a simple $/MWh comparison with solar.

On the cooling-efficiency side, the economics are clearer even if specific numbers for geothermal pilot projects aren't broadly published. The PUE math is unambiguous: each 0.01 improvement in PUE at a 100 MW IT load is roughly 1 MW less overhead demand, continuously. That's 8,760 MWh per year per 0.01 PUE improvement. At any reasonable electricity tariff, closing the gap between industry-average PUE and best-in-class PUE represents enormous value. Geothermal-assisted direct cooling — if it matures as the DOE expects it can — would add to that picture. But the direct cooling pathway is earlier-stage, and the published numbers at hyperscaler scale aren't there yet.

For homeowners curious how the residential side of geothermal economics works, our geothermal installation cost guide covers the full picture including the 30% federal tax credit. The commercial data center calculus is different, but the fundamental principle — high upfront, low operating, long asset life — rhymes.

DOE Is Paying Attention

The federal policy environment has shifted noticeably in geothermal's favor over the last few years, driven partly by the same AI-boom energy concerns pushing hyperscalers toward firm clean power.

The DOE's geothermal office has a dedicated page on geothermal and data centers, covering both the direct cooling pathway (Cold UTES, mine-water concepts) and the power generation pathway. That's not a document that appeared by accident — it reflects a deliberate policy recognition that geothermal and data centers are increasingly connected problems with potentially connected solutions.

The Enhanced Geothermal Shot is the flagship initiative: a commitment to drive EGS costs down to $45/MWh by 2035 through a combination of R&D, demonstration projects, and the kind of learning-curve effects that happen when a technology scales. The FORGE (Frontier Observatory for Research in Geothermal Energy) initiative provides the research infrastructure for that kind of learning — a dedicated field laboratory in Utah where researchers can test EGS techniques in real subsurface conditions.

The LBNL data on data center electricity demand growth has added urgency to all of this. When federal labs are projecting data center demand to reach 6.7-12% of U.S. electricity by 2028, geothermal policy stops being a narrow clean-energy niche question and becomes part of the national grid resilience conversation. If AI infrastructure is going to double or triple its share of U.S. power consumption, the grid needs more firm, clean generation — not just more intermittent renewables that require storage and backup to be useful.

Geothermal is, in this framing, a direct answer to a problem the DOE and grid planners are actively worried about.

What Comes Next

The near-term story is pretty clear: more utility-enabled geothermal PPAs, more corporate offtake agreements, and a handful of large EGS projects moving from announcement to operations over the next three to four years. Cape Station's 2026 and 2028 milestones will be closely watched. Meta's 2027 first-phase target with Sage will be closely watched. Every one of these projects is both a commercial proposition and a data point for the rest of the industry.

The medium-term story is more interesting. Direct geothermal cooling — Cold UTES, mine-water concepts, ground-coupled heat rejection — is genuinely promising for data centers in the right geologies, but it needs demonstration projects to generate the performance data that would let operators write it into facility designs confidently. The hyperscalers have the capital and the technical teams to run those pilots. Whether they will, and at what speed, will determine how quickly path two develops alongside path one.

There's also the question of geography. The current wave of U.S. deals — Nevada, Utah, New Mexico — maps onto states with known hydrothermal and EGS resources. But EGS isn't inherently geographically constrained the way conventional geothermal is. Drill deep enough, anywhere, and you find hot rock. The question is whether the economics pencil out at depths that are commercially drillable with current technology. The DOE's cost targets and Fervo's Cape Station project are both bets that the answer is increasingly yes — and that as the industry scales, the acceptable geographies will expand.

Beyond the U.S., Kenya is a template for something broader: in regions where grid reliability is a real operational constraint and geothermal resources are abundant — East Africa, Southeast Asia, parts of Latin America — the data center case for geothermal power isn't just about sustainability. It's about operational viability. For hyperscalers expanding into these markets, "runs on reliable geothermal power" solves multiple problems at once.

The AI boom is accelerating everything. Data center power demand is growing faster than most grid planners modeled, and the gap between "intermittent clean" and "firm clean" has gone from a theoretical concern to an active procurement challenge. Geothermal has spent decades waiting for a moment when "always on" mattered as much as "low marginal cost."

That moment is now.

If you're trying to understand where geothermal fits in the broader energy picture — from a residential heat pump on your property to a 500 MW EGS field in Utah powering AI training clusters — the underlying physics is the same: the earth holds heat. The engineering just keeps getting better at extracting it. For a primer on how the residential side works, see our guide on how geothermal heat pumps work. And if you're evaluating geothermal against conventional options on cost grounds, our geothermal vs. traditional HVAC comparison breaks down the economics in detail.

The data center story is the newest chapter in a technology that's been quietly proving itself for a long time. The difference now is that the companies writing the checks are Google, Meta, and Microsoft — and the checks have a lot of zeros.