Google이 AI 데이터센터 확장을 지원하기 위해 $10.2 billion을 투자하여 396 MW의 지열 발전 용량을 확보했다.

On September 1, 2026, Google signed a 396-megawatt power purchase agreement (PPA) with Fervo Energy, the largest enhanced geothermal systems (EGS) deal ever signed. The agreement includes an option to expand toward nearly 1 gigawatt by June 2030—a single contract representing roughly 10% of America's entire installed geothermal power fleet.

Read this announcement as proof that geothermal has arrived, and you're half right. Read it as proof that a scarce clean firm resource has suddenly become abundant, and you're wrong. Fervo's own release calls the power "a foundational building block for a potential data center in Utah," while carefully hedging that the data center itself remains "subject to engineering feasibility, state and local approvals, and commercial conditions." The power purchase is firm. The data center is not. That distinction matters because it is the first clue to how much of this story is confirmed and how much still rests on the durability of a cost curve tested in only one county.

This is Google's fourth disclosed geothermal commitment in three years: a 3.5 MW pilot in 2023, a 115 MW deal in Nevada in 2024, up to 150 MW from Ormat in February 2026, and now this. Exercise every option and the company's total disclosed geothermal book climbs toward 1.26 GW. It is also Fervo's validation moment: the company priced the largest clean energy initial public offering ever in May, and this PPA sent its stock up 23% in a single session. The strategic question worth asking isn't whether enhanced geothermal works—Fervo's drilling data settles that. It's who gets access to the next gigawatt of a resource geographically trapped in exactly the wrong place to solve America's power problem.

America's installed geothermal power fleet is small and concentrated: roughly 3.97 GW of nameplate capacity across 99 plants, with California and Nevada alone accounting for more than 90% of it. Set against that base, Google's 396 MW contract represents nearly a tenth of the entire existing US fleet. The U.S. Department of Energy's commercial-liftoff scenario for EGS targets 2–5 GW deployed across four to six states by 2030, an aggressive case of 8–10 GW, and a longer-dated projection of 90 GW of total US geothermal capacity by 2050—a figure combining power generation with non-power uses like direct-use heating and not directly comparable to the 3.97 GW power-only baseline. Even the conservative reading of that trajectory implies an industry that must grow by an order of magnitude within a decade.

The demand side explains why anyone would underwrite that bet. Wood Mackenzie projects US data center capacity growing from roughly 24 GW to 110 GW between 2026 and 2030—68% of all US load growth in that window—and tracks more than $2 billion raised by next-generation geothermal companies since 2019, naming Google and Meta as the sector's most active hyperscaler partners. Every gigawatt-scale clean firm resource, whether gas, nuclear, or geothermal, is being bid for against that backdrop. It's the structural reason a technology this early and capital-intensive can command a $10 billion public-market valuation before booking its first full quarter of profitable revenue.

Meta is running a visibly different playbook. Rather than concentrating with one scale leader the way Google has with Fervo, Meta has split roughly 300 MW across two earlier-stage providers: 150 MW with Sage Geosystems in Texas (the first next-generation geothermal deal east of the Rockies) and 150 MW with XGS Energy in New Mexico, a zero-operating-water design. Both companies' flagship projects sit materially earlier in de-risking than Cape Station. Sage's commercial-scale site wasn't even finalized when its Meta PPA was signed, and XGS's pilot well was roughly 100 meters deep against Cape Station's 15,000-plus-foot commercial wells. Diversification is a rational hedge against single-geology risk, but it also means Meta is paying for optionality on unproven sites, while Google is paying a premium to concentrate with the only company that has actually drilled at commercial scale.

Google's geothermal build-out fits a template already documented across three other procurement moves this year: the company arrives as anchor tenant, underwrites a nascent technology's learning curve with a long-dated offtake commitment, and absorbs the early, expensive supply of a resource that is scarce today but improving on a visible cost curve. That's the shape of Google's carbon capture and storage offtake with Broadwing, its vertical-integration play in the Intersect acquisition, and its bring-your-own-capacity virtual power plant structure. It's now the shape of Fervo's IPO too. In each case, Google is paying a premium to be customer of record while a technology proves it can scale, not paying market rate for a mature commodity. The tradeoff is higher price today for priority access tomorrow.

Utah's large-load statute, the likely legal mechanism behind this deal, gives Rocky Mountain Power 90 days to determine whether it can serve a prospective Utah data center load on the existing grid. If it can't, the statute explicitly authorizes the customer to pursue independent generation, including "closed private generation systems," as an alternative to standard grid interconnection. Neither Fervo's nor Google's announcement specifies which path this deal takes. That ambiguity matters: a grid-delivered structure puts geothermal in direct competition with gas and nuclear as a wholesale resource, while a private-wire or behind-the-meter structure would put it in the same on-site generation category that hyperscalers are increasingly using to skip interconnection queues altogether.

Fervo priced its IPO at $27.00 a share on Nasdaq under the ticker FRVO in May, the largest clean energy IPO on record, reaching a roughly $10.2 billion valuation after a 33% first-day pop. That valuation was built almost entirely on backlog and hyperscaler validation rather than current profitability; Fervo had not booked a full quarter of revenue at the time.

What underwrote it instead was Cape Station's advantage stack: geologic data from a decade of federal research at Utah FORGE, existing transmission, an adjacent operating hydrothermal plant at Blundell, and federal approval to build to 2 GW. No other prospective US EGS site currently combines all four. That combination, not the drilling technology itself, is the article's real subject and the reason Google's PPA is better read as evidence of what a hyperscaler will pay to be first in line, not evidence that the line has gotten any shorter. Understanding exactly how that advantage stack was built and how hard it would be to replicate anywhere else is the difference between treating this deal as a template and treating it as an outlier.

Geothermal's proven, conventional resource base sits almost entirely in California and Nevada, with Utah's much smaller installed base notable mainly for hosting Cape Station itself. The fastest-growing data center load is concentrated in PJM's footprint, Texas, and the Southeast—almost none of which overlaps with that resource. The one real exception: Meta's Sage Geosystems EGS project in Texas, still early-stage, is the first evidence that enhanced geothermal can travel east of the Rockies.