Data centers are hunting for 24/7 power, and geothermal just got interesting
The AI boom’s done something very old-fashioned to the tech industry: it’s made electricity the new bottleneck. Model training, inference, storage, cooling, networking and the sheer sprawl of modern data centers all pull from the same grid. Utilities have noticed. So have hyperscalers, who are now shopping for power the way they once shopped for office space, except the lease terms are longer and the appetite’s much less polite.
That shift has sent operators looking beyond the usual menu of generation. Solar is cheap when the sun shows up. Quick aside. Wind is useful when the air cooperates. Both can be part of the answer, but neither gives a data center what it wants most, which is round-the-clock output with fewer surprises. AI workloads don’t clock out because clouds rolled in over a wind farm. The chips keep chewing through electricity at night, on holidays and during those awkward weeks when the weather refuses to fit the spreadsheet.
AI doesn’t wait for sunrise, so power planners are getting very interested in anything that can keep the lights on without checking the forecast.
That’s where geothermal has started to look a lot less niche. The pitch is simple enough to fit on a whiteboard. You can generate electricity continuously instead of relying on sun or wind patterns, if you can tap heat from underground. For a data center operator trying to lock down long-term supply, that kind of predictability’s attractive. It isn’t glamorous. And it won’t get a lot of likes on social media. Show up every hour of the day, which is more than can be said for some cleaner generation options, it does, however.
For years, geothermal power lived a fairly small life in energy planning. It worked best in places with the right geology, which meant a limited number of hot spots and not much room for scale. That made it useful, but not exactly a mainstream conversation for big computing loads. The recent scramble around AI electricity demand has changed the mood. When the grid gets tight and everyone is hunting for firm power, “limited” starts to look like “interesting if somebody can crack the geography problem.”
That somebody, at least for now, is Fervo Energy. The company has been trying to make geothermal less dependent on lucky location and more like something utilities and data center developers could actually plan around. In plain English, Fervo is working on a version of geothermal that can be built more widely than the old model, which was often tied to naturally perfect underground conditions. That matters because the data center crowd doesn’t want a power source that works only if the rocks happen to cooperate.
The timing could hardly be more convenient, if you’re the sort of person who likes your timing neat and your capital stack slightly monstrous. Google has widened its backing of Fervo, and the company has also closed a fresh funding round. Put those two things together and geothermal starts to look less like a science-fair curiosity and more like an asset class. Not a cheap one, either, and not a simple one. But a real one, with the sort of financing and corporate attention that usually comes before steel gets bought, permits get filed, and executives begin saying things about “capacity” with unusually serious faces.
This is the part that makes the geothermal story different from the usual clean-tech optimism parade. A lot of climate tech lives on promises that are still waiting for a buyer. Fervo, by contrast, now has a bigger named customer in Google and more money in the bank, which gives the market something sturdier to react to. Quick aside. Utilities can model around it. Data center developers can ask whether it fits into procurement plans. Power and politics folks can argue about whether the grid should make room for it. Even the digital culture crowd, which usually treats electricity as background noise until the Wi-Fi dies, has reason to care when AI keeps bending demand upward.
There’s also a practical charm to geothermal that tends to get lost in cleaner-than-thou marketing copy. It doesn’t need a battery the way intermittent renewables often do if a buyer wants steady supply. It doesn’t need a giant weather hedge. It just needs the right subsurface conditions and a company willing to spend real money finding them. That’s a tougher sell than putting panels on a roof, but it may suit the kind of customers now entering the market. Data centers aren’t buying a virtue signal. They want firm megawatts and fewer headaches.
And yes, the whole thing has a slightly comic edge. The same industry that spent years moving fast and breaking things is now staring at the utility map and realizing that rocks may be the grown-up answer. No one gets to call that glamorous. They do get to call it useful.
For the moment, the conversation has shifted from “could geothermal matter?” to “can it scale fast enough for buyers that need power every hour, not just on a sunny Tuesday?” That question leads straight to the projects now coming online and the first real proof that this can work outside a slide deck.
Fervo’s Utah plant is the proof point everyone wanted
That scramble for steady electricity’s where Fervo Energy’s Utah project stops sounding theoretical and starts sounding like infrastructure. The company says the plant’s now producing carbon-free power at utility scale, which is a different category entirely from a pilot site or a promising drill test. A plant can sit on a slide deck for years. A plant that’s sending electrons onto the grid’s harder to wave away.
The timing matters because utilities and data centers are both hunting for the same thing: firm power that doesn’t disappear when the sun goes down or the wind drops off. The Department of Energy has also been flagging the broader jump in electricity demand from server farms and other new loads, which has pushed more attention toward resources that can run all day without much drama. That demand growth is now part of the planning problem, whether utilities like the spreadsheet or not.
What Fervo is doing in Utah’s usually described as improved geothermal, a phrase that sounds a bit lab-coated until you strip it down. Conventional geothermal depends on naturally lucky geology. You need hot rock, water and enough permeability for fluids to move through underground cracks without much help. That works in a limited number of places, mostly out West, where the right conditions happen to line up. If the geology is wrong, the concept usually stops there.
Improved geothermal tries to loosen that constraint. Instead of waiting for a naturally perfect reservoir, engineers drill down, steer the well sideways and use hydraulic fracturing to open flow paths in hot rock. Water is pumped underground. It absorbs heat, and then it comes back up another well as hot fluid that can drive power generation equipment. In plain English, the company is building an underground circulation system where nature didn’t leave one lying around.
The trick is not finding a magical hot spring. It’s getting water to move through hot rock that already exists.
That sounds simple, and of course it isn’t. Directional drilling at depth is fussy work. The rock can behave differently from one section to the next, fractures don’t always open the way engineers expect, and a project can run into the sort of underground surprises that make budgets twitchy. Still, the logic is clean enough that even people outside the geothermal crowd can follow it. If you can make hot dry rock act more like a natural reservoir, you can use geothermal energy in places that never had a traditional geothermal plant before.
That geographic part is the real hook. Usual geothermal in the United States has been clustered in a relatively small set of western strongholds, where heat and permeability show up together often enough to make projects pencil out. Fervo’s Utah plant is aimed at a broader map. The idea isn’t to keep geothermal locked inside a few lucky basins in Nevada and California. It’s to see whether the same basic method can work wherever the rock is hot enough and the subsurface can be engineered well enough.
That is why the Utah project has drawn attention far beyond the clean-tech crowd. It is one thing to say enhanced geothermal might help serve data centers someday. It is another thing to point to a plant in Utah that is actually producing electricity now. The commercial logic starts to look less like optimism and more like procurement. A utility-scale site gives buyers, regulators, and investors a concrete object to inspect instead of a promise to squint at.
The Google connection has also helped push the story out of the niche-energy aisle. Google’s 396-megawatt power purchase agreement with Fervo Energy gave the Utah build a large corporate customer and a lot more visibility. That doesn’t solve every problem, and it certainly doesn’t mean geothermal suddenly becomes easy. But it does tell you a major buyer is willing to treat enhanced geothermal as something more than a science project. In this corner of power markets, that counts for a lot.
For years, geothermal was the sort of energy source people liked in principle and then immediately filed under “limited geography.” The Utah plant complicates that habit. If the wells keep performing, Fervo can point to a working model that isn’t tied to one special patch of ground. The circle of possible sites widens, if the model can be repeated in other hot-rock regions. That’s the part utilities care about, because a technology that only works in a few ZIP codes is interesting. A technology that can be placed where load’s growing is a different conversation entirely.
So the Utah facility isn’t just a ribbon-cutting moment for Fervo Energy. It’s a field test for whether improved geothermal can move from a very clever idea into something that utilities might actually slot into long-term planning. The rock is hot, the wells are drilled, and the power is flowing. What happens next’s whether the rest of the industry treats that as a one-off, or as the first practical template for putting geothermal in places that have spent decades relying on everything else.
The real test is whether geothermal can scale fast enough for the AI boom
The Utah plant was the proof of concept. Now comes the part where the spreadsheets get mean.
And for geothermal to matter to data centers, it has to do more than produce clean power in one favorable location. It has to show up at a price buyers can live with, on a timetable they can trust and in places that aren’t blessed with unusually friendly geology. That’s the hard part, and it’s why Fervo Energy’s latest financing round and its deeper relationship with Google matter so much. The company said its public offering brought in a little over $2 billion, a pile of money that can go into more drilling, more sites, and a lot of expensive trial and error. With a capital base that large, Fervo isn’t just talking about one plant in Utah. It’s trying to turn improved geothermal into an systems business.
That distinction matters for the AI data center electricity crunch. Hyperscalers can buy wind and solar, sure, but those sources don’t always match the round-the-clock load profile of a giant computing campus. Geothermal does, at least in theory. The pitch becomes simple: clean power that doesn’t clock out when the sun does, if the wells can be built reliably and the power can be delivered where the servers sit. Data center operators love boring, and they just prefer boring with voltage.
The whole wager is whether geothermal can stop being a lucky exception and start behaving like a repeatable product.
Google’s expanded deal with Fervo says a lot in a few words. A company that size doesn’t sign up for novelty’s sake. It wants firm supply, predictable output and a path to scale. When a buyer with Google’s balance sheet backs a geothermal developer, it sends a practical signal to the market: this isn’t merely a science project wearing a hard hat. The deal gives Fervo more than revenue. It gives the company a named customer whose demand helps justify the next round of wells, transmission planning and site development. In power markets, that sort of commitment can do more work than a hundred conference panels.
Still, the question hanging over the sector isn’t whether geothermal can work in Utah or Nevada. Those places have long made sense for geothermal because the underground conditions are unusually favorable. The real test is whether Fervo’s version of the model can survive in less obvious places, where the rock may be hotter than expected, cooler than hoped, or simply a headache to drill through. If the economics only pencil out in a narrow slice of the West, geothermal remains a regional tool. Then it becomes something data centers can actually plan around, if the method can travel.
That portability depends on a handful of stubborn problems. Drilling risk sits near the top of the list. Even with horizontal drilling and hydraulic fracturing, wells can miss target zones, underperform, or cost more than planned. A few bad wells might be fine for a pilot. They are a headache when you’re trying to sign contracts large enough to feed an AI cluster that runs all day and all night. The geology can also shift from site to site in ways that don’t show up neatly in a pitch deck (and that’s no small thing).
Permitting is another slow grind. Geothermal projects still need land access, environmental approvals, water plans, and local buy-in, all before the plant starts sending electrons anywhere useful. That process can drag. Transmission adds another layer. A geothermal site may sit on excellent rock and still be too far from a grid connection or a major load center to make financial sense. The electrons have to travel somewhere, and wires are expensive, politically fraught and often harder to build than the generation itself.
Then there’s execution, which sounds dull until a project misses a deadline and the cost curve bites back. Geothermal developers need to repeat the same basic project many times with enough consistency that financiers stop treating each plant as a coin toss. That means drilling crews, subsurface models, reservoir management and construction schedules have to line up with unusual discipline. One plant’s news. Ten plants with similar performance would start to look like a business model.
For data center buyers, that repeatability may matter more than the romance of the technology. Operators shopping for clean power don’t need poetry. They need contracts, delivery dates, and enough confidence that a plant coming online in 2029 will still perform in 2039. Geothermal can offer that sort of long-term stability, but only if the developers can get past the early errors that usually come with a new industrial play. A one-off success gets applause, and a series of them gets financing.
The money Fervo has raised gives it a better shot at proving that sequence. Google’s broader support gives the company a customer with real demand. Neither solves the geology, the permitting mess, or the transmission bottlenecks. They do, however, buy time and scale, which is often how infrastructure starts to look real. If the next round of projects comes in on budget and on schedule, geothermal could move from “interesting clean power option” to something that data center planners build into their base case. If not, it stays where plenty of promising energy ideas end up: admired, discussed, and left on the sidelines while the servers keep getting hungrier.
