Hyperscalers are burning through electricity faster than grids can supply it, and investors are racing to back the companies that build the pipes, plants, and hardware that will close the gap.
Fervo Energy didn't just go public in May. It cracked open a category. The geothermal startup raised $1.89 billion in its IPO on May 13, pricing at $27 a share before popping 33% on its first day of trading, and landed a valuation north of $10 billion. That made it the largest clean energy IPO on record, and the fuel behind the pop was straightforward: Fervo has 658 megawatts of binding power-purchase agreements in place, a backlog potentially worth $7.2 billion, and a signed framework agreement with Google Energy covering roughly 3 gigawatts of buildout through March 2028. When the biggest hyperscaler on the planet is willing to commit at that scale, institutional money follows.
The timing is not a coincidence. Hyperscaler capital expenditure is on track to exceed $725 billion in 2026, according to Bloomberg, and power availability has quietly become the binding constraint on where and how fast that money lands. The IEA reported that electricity demand from data centers surged 17% in 2025, with AI-focused facilities growing even faster. By 2030, the agency projects that data center power use will double overall and triple for AI-specific infrastructure. Goldman Sachs has put a number on what closing that gap requires: a 160% increase in data center power by 2030, with utility re-rating opportunities worth roughly $5 trillion. That is not a rounding error. That is a structural reordering of the energy sector.
If Fervo's IPO signaled the opening of the category, nuclear is where the next bets are accumulating. X-Energy raised approximately $1.1 billion in its IPO and began trading on Nasdaq under the ticker XE in 2026. Terrestrial Energy, listed as IMSR, announced a collaboration with Riot Platform for a co-located nuclear and data center project targeting up to 4 gigawatts of capacity for hyperscale AI workloads. Constellation Energy, which operates the largest nuclear fleet in the United States across 21 reactors and 12 sites, is already signed to restart Three Mile Island to supply Microsoft's data centers, with first power expected in 2027.
The pipeline of conditional offtake agreements between data center operators and small modular reactor projects has grown from 25 gigawatts at the end of 2024 to 45 gigawatts today, per IEA data. Amazon alone has committed a $500 million investment in X-Energy for next-generation SMR development, alongside separate partnerships with Energy Northwest and Dominion Energy. Don't mistake this for altruism. These are procurement contracts dressed as climate commitments. Hyperscalers need baseload power that doesn't flinch when clouds block the sun, and nuclear is the only carbon-free source that can deliver it at the scale required.
The land arbitrage is already visible. Within a 20-mile radius of operating or restarting nuclear plants, data center land values have climbed materially, according to analysis from Build.inc, because proximity to the source eliminates transmission loss and cuts interconnection queue times. Power certainty is now the primary differentiator in data center site selection. That makes nuclear-adjacent sites a scarce commodity, and scarcity attracts capital.
Where VC is quietly moving
Venture capital has historically sat out infrastructure at this scale. The capital intensity is too high, the timelines too long, and the exit multiples too compressed relative to software. That calculus is shifting. VC investment in energy hit $9.6 billion in 2025, its highest level in more than a decade, and the smart money is flowing toward the companies solving specific chokepoints in the supply chain, not the commodity generation itself.
Heron Power, which builds hardware designed to move electricity from renewable sources into the grid and data centers, closed $140 million in funding backed by Andreessen Horowitz and Breakthrough Energy Ventures. ThinkLabs AI, which runs AI-native simulations of the electric grid for utility planning and operations, raised a $28 million Series A from Energy Impact Partners, with Nvidia's venture arm NVentures and Edison International's parent company both participating. ThinkLabs doubled its utility customer accounts in the first quarter of 2026. These are not moonshots. They are picks-and-shovels plays targeting the most immediate friction in the buildout, and they are attracting the kind of investors who know the difference between a trend and a bottleneck.
The broader framework matters here. As of early 2026, roughly 190 gigawatts of hyperscale data center capacity has been announced across 777 projects globally, with about 148 gigawatts in planning, 21 gigawatts under construction, and 12 gigawatts already live. None of those projects can go operational without a power source. That supply gap is a business opportunity, and founders are starting to treat it that way.
Frankly, the more interesting question is not which company wins. It is which category proves most durable. Nuclear restarts carry real execution risk and multi-year timelines. Geothermal, as Fervo demonstrated, can move faster but requires specific geology. Grid hardware and transmission software are capital-light by comparison and sit on the critical path for every other solution. McKinsey estimates global AI-powered data center infrastructure capex will reach $7 trillion by 2030. A fraction of that sum finding its way into the companies that solve the interconnection queue, the substation delay, or the grid stability problem would create a generation of infrastructure companies worth more than anyone is currently pricing in. The energy buildout is not the story behind the AI story. It is a story in its own right.
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