AI has made electricity a boardroom problem, and investors are starting to treat power infrastructure as a technology category in its own right.
Fervo Energy didn't just make a strong market debut in May. It gave investors a clean, simple story to buy: AI needs more power, grids are slow, and companies that can add firm electricity at scale suddenly look less like climate bets and more like core infrastructure.
According to Barron's, the Houston geothermal company raised about $1.9 billion in its Nasdaq IPO on May 13, pricing at $27 a share before jumping 35% on its first day of trading under the ticker FRVO. The same report put Fervo's market value at roughly $8 billion, not the $10 billion-plus figure some market chatter has attached to the deal. That distinction matters. You don't need to inflate the number to see the point.
Fervo is still tiny as an operating business. Barron's noted that it reported just $138,000 in revenue last year and a $57.8 million loss. But it also has about $7.2 billion in potential contract revenue, a working 3.5 megawatt geothermal site, and a much larger Utah project under development. Investors weren't paying for what Fervo is today. They were paying for the possibility that geothermal can become dispatchable power for data centers that don't have the luxury of waiting five years for a transmission queue to clear.
The demand side is no longer theoretical. The Financial Times' tally of guidance from Alphabet, Amazon, Meta and Microsoft put their combined 2026 capital expenditure plans at about $725 billion, up sharply from last year, as the four companies keep building data centers, chips, networks and cloud capacity. The International Energy Agency has made the same pressure visible from the grid side, warning that electricity demand from data centers will more than double by 2030, while AI-specific data center demand could rise even faster.
Here's the thing: AI companies can delay a model launch, trim marketing spend, or move a workload between cloud regions. They can't run a data center without electricity. That is why the dull parts of the buildout, substations, transformers, interconnection studies, backup power and firm generation, are becoming investable in a way they weren't when software margins were doing all the work.
Nuclear is back because the clock is unforgiving
Nuclear has become the most obvious expression of that scramble. X-energy went public on Nasdaq in April and raised more than $1 billion, according to public market reports, in what was described as the largest nuclear IPO on record. Amazon had already led a $500 million investment round in X-energy in 2024 through its Climate Pledge Fund, while working with Energy Northwest on a small modular reactor project near the Columbia Generating Station in Richland, Washington.
That project is not around the corner. Energy Northwest's Cascade Advanced Energy Facility is planned around X-energy's Xe-100 reactors, with an initial 320 megawatts and possible expansion to 960 megawatts, but construction is expected near the end of the decade and the reactor remains in the Nuclear Regulatory Commission's pre-application process. If you are a hyperscaler trying to secure power for the 2030s, that timeline is acceptable. If you need capacity in 2026, it isn't.
Microsoft's deal with Constellation Energy shows the same urgency in a different form. Constellation plans to restart Unit 1 at Three Mile Island in Pennsylvania, renamed the Crane Clean Energy Center, under a 20-year power agreement with Microsoft. The plant shut in 2019, and bringing it back requires regulatory approval and heavy capital spending. Still, it tells you how far the market has moved. A reactor once seen as an uneconomic asset is now valuable because AI has changed the price of dependable power.
Don't mistake these deals for a neat climate story. They are procurement moves. Hyperscalers need carbon-free electricity where they can get it, but they need firm electricity first. Nuclear, geothermal and gas-backed power are getting attention because solar and wind alone don't solve the problem of running dense compute loads every hour of the day.
The better startup angle is the bottleneck
The strongest opportunity for startups may not be owning generation at all. Generation is expensive, slow and heavily regulated. The more immediate openings sit around the bottlenecks: grid planning, power electronics, transformer availability, cooling systems, site selection, permitting intelligence and software that helps utilities model demand before it lands on the system like a brick.
Goldman Sachs, in a report highlighted by Axios this week, estimated that about $7.6 trillion could be invested globally from 2026 to 2031 across AI infrastructure, including computing, data centers and energy capacity. That is the number founders should pay attention to. A small share of that spending flowing into the companies that shorten interconnection queues or make existing grid assets work harder can still build very large businesses.
The risk is that investors turn every power-adjacent pitch into an AI infrastructure company. They shouldn't. A startup with a dashboard for utility data is not automatically part of the AI boom. A company that can cut months from a data center connection, reduce wasted capacity in a substation, or help a utility forecast load from a real hyperscale project has something much more concrete.
Fervo's IPO made the category visible, but the category is bigger than geothermal. X-energy, Constellation, Amazon and Microsoft show you the shape of the next phase: AI capital is moving out of the server rack and into the physical economy. The winners won't be the companies with the loudest energy story. They'll be the ones sitting closest to the constraint.
Also read: Steve Eisman says investors betting on hyperscalers in the AI race are funding the wrong side of the trade • Solana now processes 95% of the world's tokenized stock trades but its token price tells a different story • The AI boom is keeping inflation alive and interest rates higher than founders want
