The Department of Energy's Reactor Pilot Program gave a cohort of companies a real deadline — July 4, 2026 — to get a nuclear microreactor to criticality. Four made it. Antares' Mark-0 went critical in early June, the first privately developed non-light-water reactor to do so in roughly forty years. Valar Atomics' Ward 250 followed. Deployable Energy's Unity reached criticality July 1. Aalo Atomics' test reactor made it at 00:20 local time on July 4 — hours before the deadline closed.[1] Four different companies, four different reactor designs, one government deadline, and real, physical machines that now provably work — not funding announcements.
What makes this a pattern rather than a coincidence is what was happening in parallel, on the corporate side, completely independent of the DOE program. Microsoft signed a 20-year power purchase agreement with Constellation Energy in September 2024 to restart Three Mile Island's undamaged reactor as the Crane Clean Energy Center — a $1.6 billion project backed by a $1 billion DOE loan that closed in November 2025. Originally targeted for a 2028 restart, it's now tracking for 2027, ahead of schedule, with fuel delivery expected by the end of 2026.[2] Google followed a month later, signing a Master Plant Development Agreement with Kairos Power in October 2024 for a 500-megawatt fleet of fluoride-salt-cooled reactors by 2035 — the first corporate agreement of its kind for an entire fleet of small reactors rather than a single plant.[3]
Amazon and Meta each did the same thing, with different partners. Amazon backed X-energy with roughly $500 million in October 2024, followed by another $700 million in November 2025 and an IPO raising over $1 billion in 2026 — funding the Cascade Advanced Energy Facility in Richland, Washington, up to twelve Xe-100 reactor units built with Energy Northwest, the same public-power utility that already operates the Pacific Northwest's only commercial nuclear plant.[4] Meta signed with TerraPower in January 2026 for Natrium reactor units to power its Prometheus AI data center in New Albany, Ohio, and separately partnered with Oklo Inc. to build a 1.2-gigawatt nuclear campus on a former Cold War uranium-enrichment site in Piketon, Ohio.[5] Oklo's other deals go further than any single hyperscaler: a 12-gigawatt agreement with data-center developer Switch, announced December 2024 — a company already on the record in Oluwadi's own Nevada coverage — plus non-binding agreements with Equinix and Diamondback Energy, and a 20-megawatt, 20-year PPA with Wyoming Hyperscale signed in May 2024.[6]
None of these four companies needed to coordinate with each other to arrive at the same answer. Microsoft, Google, Amazon, and Meta compete on almost everything — model quality, chip supply, cloud market share — and yet on the specific question of where their power is going to come from a decade from now, they converged on the identical answer within about eighteen months of each other: build or restart nuclear capacity, dedicated to one customer, priced for decades in advance. That's not an industry trend chasing a headline. It's four independent capital-allocation decisions, made by companies with every incentive to find a cheaper answer if one existed, that landed on the same one anyway.
The DOE Reactor Pilot Program is the proof layer underneath those corporate bets, and it's worth being precise about what it actually proved. Aalo's July 4 milestone was its Aalo-X Critical Test Reactor — a smaller precursor unit built to validate the physics, not the 50-megawatt "Aalo Pod" design the company plans to sell commercially, which pairs with a data-center partnership already signed with Crusoe Energy for a 2027 demonstration.[7] That distinction matters: none of the four Pilot Program reactors is yet the commercial-scale machine a data center will actually run on. What the July 2026 deadline proved is narrower and still real — that the underlying reactor physics behind this generation of microreactor designs works in practice, on a government-enforced schedule, not just in a simulation or a pitch deck. The hyperscalers' multi-billion-dollar bets were placed before that proof existed. The proof arrived anyway, roughly on the timeline the bets assumed it would.
Four competitors, the same constraint, the same answer, arrived at separately. This site has already covered two pieces of this same buildout — the single site in Ohio where a Cold War uranium-enrichment plant is becoming a Meta-Oklo nuclear campus, and the state, Virginia, now rewriting the rules the whole industry has to build under. This is the piece connecting them: not one company's deal or one state's pushback, but four of the most competitive companies in the world independently concluding that the actual constraint on AI isn't chips or capital. It's electrons — and none of them trust the grid to deliver enough, on time, without owning a reactor themselves.