One of the four new nuclear reactors squeaked in at half past midnight on the Fourth of July. That was the point. A little over a year earlier, President Trump had signed an executive order directing the Department of Energy to run a pilot program that would take at least three experimental reactor designs from paper to a self-sustaining fission reaction, and it set a date: criticality by Independence Day, 2026. The department did not just meet the deadline. It beat it by one. Officials reportedly compared the feat to the Manhattan Project, which is either the most reassuring or the least reassuring analogy available, depending on which part of that history you have in mind.
The engine behind the sprint is AI. American electricity demand is projected to climb by roughly 130 percent between now and 2030, most of it to feed data centres, and the big technology firms want that power to be clean and constant rather than drawn from gas. Nuclear, alone among carbon-free sources, runs around the clock without a battery bolted to it. That is why the checkbooks are already open: Microsoft has signed a $16 billion, twenty-year deal to restart a unit at Three Mile Island; Meta has lined up as much as 6.6 gigawatts across Oklo, TerraPower, Vistra and Constellation; Google has backed Kairos Power's salt-cooled design. This month the reactor developer X-energy raised a billion dollars on the public markets. Across the sector, hyperscalers have committed to nearly 10 gigawatts of nuclear capacity that mostly does not exist yet.
The reactors that went critical this summer are small test units, not grid-connected plants. One, built by a company called Valar Atomics, is powering a single microchip that runs a website in Utah. The interesting question is how they got approved so quickly. The answer, as NPR reported, is that the pilot program was designed to route around the Nuclear Regulatory Commission, the slow, independent body that has overseen commercial reactors since the 1970s with a near-spotless safety record. The Energy Department is permitted to regulate certain experimental reactors itself, and in January it rewrote its own internal rules to do so, cutting more than 750 pages. A requirement that groundwater "must be protected" from radiological contamination became a softer instruction that "consideration must be given" to minimising it. Five hundred pages of security rules were compressed into twenty-three.
Critics were not comforted. "If you bend all the rules, you can do things quickly," said Ed Lyman of the Union of Concerned Scientists, "but we learned over the past 80 years that if you want to do things right, you have to go by the book." Katy Huff, who once ran the department's Office of Nuclear Energy, objected that the draft rules were shared with the companies being regulated but never with the public. The department counters that safety remains its top priority and that "the reduction of unnecessary regulations will increase innovation without jeopardising safety." Industry makes a subtler case: that building and testing real hardware is exactly how safety is learned, and that the data from these units will let engineers design better commercial reactors when they eventually face the NRC.
Both things may be true, which is what makes the moment genuinely hard rather than simply alarming. But there is a quieter risk the deadline glosses over. The assumption driving this whole race is that regulation was the thing holding nuclear back. It may not be. Nuclear power is expensive even at small scale, demanding specialised operators, engineers and equipment, and nobody has yet shown these venture-funded reactors can compete on cost with the batteries and gas already powering data centres today. Cutting the paperwork proves you can build a reactor fast. It does not prove the reactor was worth building.