Valar Atomics and Nvidia staged a live demonstration in Emery County, Utah on July 1, powering an Nvidia Blackwell AI chip directly from Valar’s Ward250 microreactor on stage, a small moment with a large implication: the AI industry’s power problem has gotten serious enough that a nuclear startup founded in 2023 is now a credible infrastructure partner for the world’s most valuable chipmaker. The Ward250, a helium-cooled reactor, went critical on June 18 and is currently generating 100 kilowatts, a modest output that the two companies are already using as the foundation for a much larger pitch: a 30-megawatt pilot data center in the Utah desert, cooled with what they describe as near-zero local water use, pairing the reactor’s helium cooling loop with Nvidia’s own water-efficient chip cooling technology.
Valar’s founder, a descendant of a Manhattan Project-era engineer by his own account, set out with the explicit goal of making nuclear reactors smaller, safer, and buildable on something closer to a factory line than a decades-long mega-construction project. The Utah facility, which broke ground in September 2025, is the company’s first attempt to prove that model works in practice rather than on a slide deck, and the Nvidia partnership gives it an anchor customer with essentially unlimited appetite for firm, reliable power.
Valar is a small player in a very large trend. As of May 2026, every major cloud hyperscaler has signed at least one nuclear power deal specifically to feed AI data center capacity, spanning thirteen announced projects committing more than 9.8 gigawatts of nuclear generation. Microsoft locked in 835 megawatts through a $16 billion, 20-year power purchase agreement tied to restarting Three Mile Island’s Unit 1 reactor. Google committed to 500 megawatts from Kairos Power’s next-generation reactors. Amazon put $700 million into X-energy to secure up to twelve of its Xe-100 small modular reactors. Meta has gone furthest of all, lining up as much as 6.6 gigawatts across four separate providers: TerraPower’s Natrium reactor, Oklo’s Aurora design, and conventional nuclear capacity from Vistra and Constellation. Across eighteen tracked facilities, the industry has now committed to roughly 31.2 gigawatts of nuclear-linked data center capacity.
The reason every one of these companies is suddenly comfortable signing multi-decade nuclear power contracts, an asset class the tech industry avoided almost entirely for the previous two decades, is straightforward: AI data centers consume electricity at a scale the existing grid was never built to absorb on the timelines hyperscalers now operate on. Combined electricity consumption from data centers, AI workloads, and cryptocurrency mining could approach 1,050 terawatt-hours in 2026, a figure that, treated as a single country, would rank as the world’s fifth-largest electricity consumer, ahead of most G20 economies.
Solar and wind can’t reliably deliver the constant, always-on baseload power a data center running AI workloads around the clock actually needs, and new natural gas plants face years-long turbine backlogs and permitting delays of their own. Nuclear, despite its own long construction timelines for conventional reactors, offers something increasingly attractive to companies with balance sheets large enough to fund it directly: predictable, contracted, carbon-free baseload capacity that doesn’t depend on weather or a congested regional grid, provided you’re willing to either restart an old plant or bet on a new generation of smaller, faster-to-build reactor designs like Valar’s.
The International Atomic Energy Agency has been tracking this shift closely enough to publish its own analysis on the trend, noting that data centers, artificial intelligence, and even cryptocurrency mining operations are increasingly looking to advanced nuclear technology specifically to meet power needs that conventional grid connections and renewable buildouts can’t satisfy on the timelines the industry now operates on. That a United Nations nuclear watchdog is treating commercial AI data center demand as a meaningful driver of next-generation reactor deployment, alongside the traditional drivers of climate policy and energy security, is itself a sign of how fast this reshuffling has happened. As recently as 2023, nuclear power and AI data centers were treated as almost entirely separate policy conversations.
The Philippines has its own version of this exact power constraint waiting on the other side of its AI ambitions. The reported $10 billion AI infrastructure hub taking shape near Clark Freeport Zone is, at its core, a bet that the Philippines can host serious data center and compute capacity, and every one of the hyperscaler deals described above is evidence that hosting that kind of capacity increasingly requires a firm, dedicated power source that the existing national grid can’t simply absorb as an afterthought.
The Philippines already has an unusual asset in this conversation: the mothballed Bataan Nuclear Power Plant, built in the 1980s and never fueled, alongside renewed interest from the Department of Energy in small modular reactor technology as a potential path back into nuclear power. Whether Clark’s AI hub ends up genuinely viable at the scale being pitched, rather than remaining a headline investment figure with no reliable power behind it, may hinge less on the dollar amount pledged and more on whether the Philippines can credibly solve the same firm-power problem that just pushed Microsoft, Google, Amazon, and Meta into the nuclear industry directly.
Share this article