Signals Inbox·September 2, 2026·Grid Tech

Why is Google buying so much geothermal power?

Google is buying geothermal because AI is turning dependable 24/7 electricity into a bottleneck, and Fervo can now plausibly deliver hundreds of megawatts on the timeline Google needs.

Q1What did Google actually buy from Fervo?

Google has just committed to 396 MW of enhanced geothermal power from Fervo Energy in Utah, making this a real hyperscale power deal rather than another geothermal experiment.

Fervo’s latest SEC filing shows that Google Energy signed a 15-year power purchase agreement for electricity from Cape Station in Utah. The capacity is due to arrive in four 99 MW blocks starting in 2028. Google also gets the opportunity to add roughly another 600 MW, which could push the total close to 1 GW by 2030.

That is a huge jump from where the Google-Fervo relationship started. Their first commercial project in Nevada, which began delivering power in 2023, was roughly 3.5 MW. A later deal involving Google, Fervo and NV Energy expanded that relationship to 115 MW. Google is now moving into projects measured in hundreds of megawatts.

The latest contract is also tied to a potential Google data center in Utah. Fervo’s SEC filing says the project may eventually use a private generation system that could deliver electricity directly to data-center load, subject to engineering and regulatory approvals. Google is effectively reserving a large block of future computing power supply while the geothermal plant is still being built.

Q2Is Google really buying that much geothermal power?

Yes. Google’s geothermal purchases are still small next to its entire clean-energy portfolio, but they are enormous compared with the geothermal industry that exists today.

Google signed more than 12 GW of net-new clean-energy agreements in 2025 alone, according to its latest Environmental Report, and nearly 35 GW between 2010 and 2025. Against those numbers, a 396 MW geothermal contract does not transform Google’s overall energy mix.

Now compare it with geothermal itself. The latest U.S. Geothermal Market Report from the National Laboratory of the Rockies puts total U.S. geothermal nameplate capacity at only 3.97 GW in 2024. Google’s new Fervo contract by itself equals almost 10% of that installed base. If the Utah expansion reaches roughly 1 GW, one Google-linked project would equal around one-quarter of all U.S. geothermal capacity that existed in 2024.

The geothermal market had already started moving before this deal. The same national-lab report counted 27 geothermal PPAs signed from 2021 through mid-2025, compared with only nine between 2015 and 2019. Google is arriving just as procurement starts moving from scattered projects toward much larger commitments.

Google-Fervo deal versus U.S. geothermal capacity

Q3Why does Google suddenly need so much 24/7 power for AI?

Google’s electricity demand is currently growing at a pace that makes reliable power supply a business constraint, especially as the company builds more AI infrastructure.

Google reported a 37% increase in electricity demand in 2025, the largest annual increase in its history. The previous year was already up 27%. Its data centers consumed more than 42 million MWh in 2025, versus about 30.6 million MWh one year earlier.

So we are looking at roughly 11 million additional MWh of data-center electricity consumption in a single year. Google can improve chip efficiency and data-center design, but efficiency has clearly failed to stop total electricity use from surging.

The wider market points in the same direction. Lawrence Berkeley National Laboratory’s latest U.S. data-center study estimates that data centers could consume 11.8% of all U.S. electricity by 2030 in its central case, with scenarios ranging from 9.5% to 15.3%.

This explains why Google is becoming interested in technologies that would have looked niche five years ago. Securing enough electricity for AI data centers has become part of infrastructure planning, rather than something handled after the computing investment has already been decided.

Q4Why can’t Google solve its power problem with solar, wind and batteries?

Solar, wind and batteries remain central to Google’s energy strategy, but they have not closed Google’s hour-by-hour clean-power gap.

Google already buys extraordinary amounts of renewable power. It signed more than 12 GW of new clean-energy agreements in 2025, its biggest annual procurement ever. Yet Google says its global carbon-free-energy score across data centers and offices was only about 65% on an hourly basis, slightly below the 66% achieved a year earlier.

That gap is easy to miss. Google has matched 100% of its annual electricity consumption with renewable-energy purchases for nine consecutive years. Annual matching means Google can buy enough renewable electricity over a year to equal what it consumes. It does not mean a Google data center is actually running on carbon-free electricity at 2 a.m. on a windless night.

Geothermal helps with those hours. A geothermal plant can produce around the clock instead of following sunlight or weather. Batteries can move renewable electricity across time, but very large data-center loads would require enormous amounts of storage when renewable output stays low for longer periods.

Google still needs cheap solar and wind in huge quantities. Geothermal gives the portfolio something different: continuous clean generation that can sit underneath those variable sources.

Q5Why is geothermal more useful to Google than nuclear or fusion right now?

Enhanced geothermal currently gives Google a plausible route to hundreds of megawatts of clean, firm power before its advanced nuclear and fusion projects are expected to reach comparable scale.

Google has already signed agreements across all three technologies. Its partnership with Kairos Power covers up to 500 MW of advanced nuclear capacity, but the first project with TVA is expected to deliver up to 50 MW around 2030. Google has also agreed to buy 200 MW from Commonwealth Fusion Systems’ first ARC fusion plant, which CFS is targeting for the early 2030s.

Fervo is working on an earlier clock. Cape Station’s first roughly 100 MW phase is currently being commissioned. Its following 400 MW phase is scheduled for 2028, the same period in which Google’s new contracted capacity is supposed to begin arriving.

Geothermal fills a very specific window. Solar and wind can supply enormous amounts of electricity today. Advanced nuclear could become much more important during the 2030s. Fusion remains a longer bet. Enhanced geothermal sits between them, with commercial plants being built now and the possibility of reaching several hundred megawatts before the decade ends.

Google-backed firm-power technologies by disclosed scale and timing

Q6Why doesn’t Google just use natural gas for reliable data-center power?

Natural gas can absolutely power AI data centers, although today even gas does not guarantee fast access to new generating capacity.

Demand for gas turbines has exploded alongside data-center construction. GE Vernova reported 116 GW of gas-power equipment backlog and slot reservations in its latest quarterly results, up from 100 GW only one quarter earlier. Mitsubishi Power told S&P Global that industry demand had risen above 100 GW annually and that delivery times for some new installations had stretched to five years or more.

That changes the comparison with geothermal. Gas still has major advantages: the technology is mature, plants can be very large, and the U.S. has abundant fuel infrastructure. But a developer ordering turbines these days may already be looking toward the end of the decade or beyond.

Geothermal also gives Google protection from two things gas cannot remove: direct combustion emissions and long-term fuel costs. Those advantages become more valuable inside a 15-year contract.

Google probably will not avoid gas entirely. The company is exploring several power structures as AI load grows. The interesting part is that next-generation geothermal can now compete for the same scarce resource gas developers are fighting over: dependable megawatts delivered quickly enough to matter.

Q7Has Fervo actually proved that enhanced geothermal works?

Fervo has proved the core enhanced-geothermal technology well enough to sell commercial electricity; Cape Station will determine whether the same approach can be repeated reliably across hundreds of megawatts.

The breakthrough came with Project Red in Nevada. Fervo drilled horizontally into hot rock, stimulated the reservoir using techniques adapted from shale drilling, circulated water through it and generated roughly 3.5 MW. The project began supplying the grid in 2023 and subsequently accumulated hundreds of production days.

Cape Station is a much harder test. Phase I is roughly 100 MW, split into three modular GeoBlocks. In its latest quarterly operating update, Fervo said the first two GeoBlocks had reached mechanical completion. Wells had already been connected to the first plant, geothermal fluid was moving through its heat exchangers and the turbines had begun spinning during commissioning.

Phase II adds another 400 MW through eight larger GeoBlocks. Fervo has secured long-lead equipment and is already drilling those wells ahead of expected delivery in 2028.

There are still real risks around reservoir performance, construction, permits, water sourcing and induced seismicity. Cape Station has experienced small seismic events during stimulation and uses a monitoring system that can trigger pauses in operations. We know enhanced geothermal works today. We still need evidence that the industrial version can keep working across dozens of wells for years.

Q8Is Fervo really getting much faster at drilling geothermal wells?

Yes. Fervo’s drilling speed has improved dramatically, and that is probably the most important reason enhanced geothermal looks commercially different today than it did a few years ago.

Project Red’s first-generation well design required about 70 days to drill roughly 11,000 feet. Fervo’s latest Sawtooth 7 well at Cape Station reached almost 19,500 feet in only 21 days, despite being longer and technically more complex. Fervo says its drilling rate has improved 143% since the first Cape Station well.

The company is also drilling much longer horizontal sections. That lets each well contact more hot rock, increasing the amount of heat that can potentially be extracted without multiplying the number of surface locations.

The same learning curve shows up outside Fervo. The latest U.S. Geothermal Market Report highlights drilling work at the Department of Energy’s Utah FORGE site where drilling time fell from 310 hours in 2020 to 110 hours in 2023.

That pattern is central to the geothermal bet. Conventional geothermal depended heavily on finding unusually convenient natural reservoirs. Enhanced geothermal adds a manufacturing element: drill, learn, standardize and repeat. If those learning curves continue, the economics can improve much faster than they would in an industry where every new project remained completely bespoke.

Q9Is enhanced geothermal getting cheap enough for Google?

Enhanced geothermal is already cheap enough for companies such as Google to sign long-term contracts, although Fervo still needs another large drop in construction costs before geothermal becomes broadly competitive everywhere.

Fervo currently expects Cape Station Phase II to cost about $5,500 per kilowatt. Its long-term target is around $3,000 per kilowatt. That is a meaningful gap, but the direction has been consistent as drilling gets faster and each new well design produces more potential power.

The financing market is also starting to treat these projects more seriously. Fervo secured $421 million of non-recourse project financing for Cape Station Phase I, meaning lenders were willing to finance the project largely against its own economics. Fervo then raised roughly $2.2 billion in its IPO, giving it far more capital to fund the next wave of construction.

Google does not need geothermal to beat the cheapest solar farm on headline cost per MWh. The comparison includes what the electricity can actually do. A continuous megawatt available overnight and during low-wind periods has more system value to a 24/7 data center than another intermittent megawatt arriving when the grid already has plenty of renewable generation.

The remaining cost question is therefore tougher: can Fervo keep cutting costs while moving from one flagship development to many projects? We do not have that answer yet.

Q10Can geothermal help Google avoid years of waiting for the grid?

Geothermal can help Google get data-center power faster when projects are designed around direct delivery, although it cannot magically remove every transmission and permitting bottleneck.

The grid problem remains severe. Berkeley Lab’s latest queue data show about 2,061 GW of generation and storage projects waiting for U.S. grid interconnection at the end of 2025. That backlog has come down from its peak, but projects that eventually get built are still spending much longer in the connection process than they did historically.

Fervo has started designing around that constraint. The company says its GeoCluster model can support behind-the-meter systems where geothermal generation serves a nearby large customer directly, alongside other energy resources. The generator can potentially connect more fully to the wider grid later.

Utah has also created a regulatory route for closed private generation systems under SB 132. Fervo’s latest SEC filing says its Google project is being developed to use those flexible delivery options, including possible front-of-meter or behind-the-meter configurations.

For an AI company, time has an economic value of its own. A slightly more expensive power source that lets a multibillion-dollar data center operate years earlier can beat a cheaper project stuck in an interconnection queue.

Q11Is Google starting to build data centers where geothermal power is available?

Google’s Utah plans suggest that access to large blocks of electricity is starting to influence where new computing infrastructure gets built.

The new Fervo contract is supposed to provide foundational power for a potential Google data center in Utah. The data-center project still needs engineering work, commercial decisions and government approvals, so we should not treat the site as finalized.

Still, the sequencing is interesting. Google has secured the power relationship while Fervo develops one of the country’s richest enhanced-geothermal clusters in southwest Utah. Cape Station already has 500 MW under development, and Fervo has identified additional multi-gigawatt geothermal resources across its wider Utah portfolio.

This reverses the way hyperscale data centers were often discussed a few years ago. Companies traditionally chose attractive data-center markets based on fiber, land, tax incentives, customers and labor, then worked with utilities to obtain more electricity. These days, electricity itself can decide which sites remain viable.

Computing is starting to move toward abundant power. Utah geothermal is one version of that shift; other developers are clustering data centers around gas generation, nuclear plants, large solar-storage projects and regions with spare grid capacity.

Q12Is Google basically helping create the enhanced-geothermal market?

Yes. Google has repeatedly committed money and demand before enhanced geothermal became an obvious mainstream power source, which has helped Fervo move from demonstration to utility-scale development.

The relationship began with the small Nevada pilot that eventually supplied Google-linked load. Google then worked with NV Energy on a structure supporting 115 MW of new geothermal capacity. In Taiwan, Google signed 10 MW of geothermal PPAs with Baseload Capital, its first geothermal deal in Asia.

Google later signed a framework with Fervo covering development opportunities of up to 3 GW through 2033. That 3 GW number needs a qualifier: the framework does not force Google to buy 3 GW. Fervo proposes projects and Google decides whether to move ahead. It gives the two companies a pipeline rather than a guaranteed multi-gigawatt order.

That pipeline now has a much stronger commercial proof point. Google has progressed from helping demonstrate the technology to signing the kind of long-duration contract that can support financing for a major power plant.

Google has used a similar playbook with advanced nuclear and fusion: commit demand early enough that developers can build projects that do not yet have a mature customer market. In geothermal, that strategy is already moving into large physical infrastructure.

Q13Are Meta and Amazon buying geothermal power too?

Yes. Meta and Amazon are now making their own geothermal bets, which makes Google’s push look increasingly like the start of a hyperscaler trend.

Meta has two major next-generation geothermal projects under development. Its partnership with Sage Geosystems is targeting up to 150 MW for data centers, while a separate XGS Energy project in New Mexico is also planned at 150 MW. Those projects use different approaches, which tells us hyperscalers are testing the category rather than simply copying Fervo.

Amazon entered the market more recently. Its Nevada energy plan includes 100 MW of geothermal power from Zanskar through NV Energy, alongside 600 MW of solar and 600 MW of battery storage. Amazon describes it as the company’s first data-center project using dedicated geothermal power.

Taken together, Google, Meta and Amazon now have disclosed geothermal projects totaling hundreds of megawatts before counting Google’s optional expansions and long-term development framework. That is a very different market from the handful of small corporate geothermal experiments we saw earlier this decade.

Google currently has the largest individual disclosed enhanced-geothermal PPA among these hyperscalers, while its relationship with Fervo also reaches further into potential multi-project development.

Major disclosed hyperscaler geothermal projects

Q14Could geothermal actually power a meaningful share of AI data centers?

Yes, especially if future data centers are built closer to good geothermal resources instead of forcing every project into today’s biggest data-center hubs.

Rhodium Group modeled next-generation geothermal against projected hyperscale data-center growth and found that behind-the-meter geothermal could economically serve roughly 55% to 64% of expected new hyperscale demand under a scenario where data centers continue clustering in broadly familiar locations. That corresponds to around 15 to 17 GW of geothermal and data-center capacity.

The potential gets much larger if developers become flexible about location. Rhodium found that placing new data centers closer to the best geothermal resources could theoretically cover all projected demand growth in its modeled scenario at substantially lower geothermal costs.

The physical resource is also much larger than today’s industry. The U.S. Department of Energy estimates enhanced geothermal could provide at least 90 GW of U.S. generating capacity by 2050. Existing U.S. geothermal capacity, as we saw above, is still below 4 GW.

That does not mean geothermal will dominate data centers. Drilling capacity, financing, permitting and successful reservoir development will put hard limits on growth. But a 10-to-20 GW data-center geothermal market no longer looks absurd. That would already be several times larger than the entire U.S. geothermal power fleet that exists today.

Q15So why is Google buying so much geothermal power?

Google is buying geothermal because AI has turned continuous electricity into a strategic bottleneck, and enhanced geothermal has suddenly become one of the few clean power technologies that could reach hyperscale quickly enough to help.

Three things have changed at the same time. Google’s electricity demand is growing at a record pace. The U.S. grid is struggling to connect new generation and large new loads quickly. Meanwhile, Fervo and other geothermal developers are applying oil-and-gas drilling techniques to resources that used to be considered too difficult or expensive to exploit.

The timing explains Google's enthusiasm. Solar and wind remain cheaper and will supply far more total energy. Batteries help move that electricity into different hours. Gas remains available for firm generation but is facing its own turbine shortages and gives Google a much worse emissions profile. Advanced nuclear looks promising around the turn of the decade, while Google's commercial fusion deal points further into the 2030s.

Geothermal currently occupies a useful gap between those options. It can run around the clock, requires no fuel deliveries, produces very little direct carbon pollution and may be deployable in hundreds-of-megawatts blocks before Google's next generation of AI data centers needs them.

Google is scaling the bet carefully rather than betting its whole power strategy on geothermal. The technology has to prove that Cape Station can deliver the same drilling and reservoir performance across hundreds of megawatts, and Google continues buying vastly more solar, wind and storage capacity.

But the direction is now clear. Google started with a tiny Fervo demonstration, moved into utility-scale procurement and is now lining up enough geothermal capacity to support an entire hyperscale data center. The company is buying so much geothermal because dependable power has become valuable enough that a once-niche energy technology suddenly fits one of Google’s most urgent infrastructure problems.