The clamour for new energy generation has become loud enough that it is beginning to drown out everything else in its wake.
Energy generation was always critical to progress; that is tautological. But in the days when long-term increases in demand were predictable and steady, it did not carry with it the sense of urgency we now see. It is because AI changed everything, aided by de-globalisation and more recently, the fossil-fuel dependency brought into bright relief in the Strait of Hormuz.
Everyone and everything from science to risk capital to industry is now chasing every last joule, at almost any cost. Oil, coal, solar, wind, tidal, hydro, battery. And nuclear. Especially nuclear. If nuclear energy were cheap, safe and abundant, then wonderful things for our planet would follow, so it is said.
Which brings us to July 4th, 2026. US Independence Day.
To understand why an American president would set a nuclear deadline by the calendar of national mythology, you have to see the race beneath the race. The United States is trying to win the contest for artificial intelligence, and it has belatedly grasped that this is essentially a contest for electricity. Frontier models are trained and served in hyperscale data centres, and those centres are voracious in a way the grid has not previously had to reckon with.
After fifteen years of flat demand, American electricity load is rising again. While figures vary somewhat, most research has data-centre demand rising by about 100% by 2030, consuming around 10% of power, up from about 4% today (these are both global and US projections from the International Energy Agency, Berkeley National Laboratory and others). There is no sign of slowdown.
The Trump administration has been unusually explicit about this causal chain. Its July 2025 AI strategy was titled, with no false modesty, “Winning the Race: America’s AI Action Plan”, and the White House framed the surrounding flurry of orders as a way to clear “energy bottlenecks to AI supremacy”.
Which is the context for Executive Order 14301, signed on 23 May 2025 as one of four nuclear orders that day. “Reforming Nuclear Reactor Testing at the Department of Energy” did something radical. Rather than route new reactors through the Nuclear Regulatory Commission (the body that has torpidly regulated American reactors for fifty years and is sometimes blamed for the country’s nuclear power plant paralysis), it invoked a dusty provision of the Nuclear Regulatory Commission exempting reactors built “under contract with and for the account of” the department.
Regulatory loophole
A regulatory loophole, so to speak, avoiding a costly political battle.
By August 2025, the administration had named eleven projects across ten companies, each bearing every cent of the cost. The stated goal was for at least three reactors to reach “criticality” (the moment a reactor sustains its own fission chain reaction) by Independence Day 2026.
While this seems like an astonishingly short schedule (it usually takes years, if not decades, to get approvals the old way), it is important to realise that “criticality” does not mean the generation of electricity, it just means regulation was lightened to enable companies to quickly get to the point where a chain reaction is sustainable (splitting the atom and releasing enough neutrons to split another, and then another and so on). That is a long way from electrons being fed into a grid.
In any event, given that Independence Day, July 4, is upon us, how is it going? Pretty well, actually.
On 4 June 2026, a month early, one of the awardees, Antares Nuclear, took its Mark-0 microreactor critical at Idaho National Laboratory. Two weeks later, Valar Atomics achieved zero-power criticality with its Ward 250 reactor in Utah. Two down. As I write, the field scrambles for the third – likely Aalo Atomics, which has the most finished hardware, or Oklo’s isotope subsidiary, Atomic Alchemy.
All of these reactors sport novel designs (here’s the partial and jumbled jargony mouthful, far too wonky to unravel – sodium-cooled, gas-cooled, helium-cooled, thorium-fuelled, ceramic-uranium-coated, liquid-metalled, borehole-sunk and other marvels). More important is that they are all small, inexpensive, scalable, iterable, (mostly) truck-transportable, innovative and largely untested technologies designed for an accelerating world, all prodded by the insatiable demands of AI. The clearing of the regulatory pathway and its weary paperwork and processes, at least to the point of criticality, looks to have been a winning strategy.
Triumphalism should pause
And here the triumphalism should pause, because the race beneath the race is one America may already be losing. As I mentioned – criticality is ignition, not electricity. Antares, which looks to be on the fastest track, does not expect to produce power until 2027 and field deployment not expected until 2028. Meanwhile, China is way beyond criticality – it is already connecting plants to the grid. Its Linglong One, a 125-megawatt small modular reactor on Hainan Island, is expected to begin commercial operation this year, which would make it the world’s first land-based commercial SMR delivering grid power. It was also the first SMR anywhere to pass an IAEA safety review, a full decade ago.
Nor is that the half of it. China’s HTR-PM at Shidaowan – a twin-reactor, helium-cooled, pebble-bed plant – is already in commercial operation, which is to say China has a grid-connected version of precisely the high-temperature architecture some of the US players are pitching as cutting-edge. In a Gobi Desert facility, Chinese scientists have run the world’s only operational molten-salt reactor, refuelling it while it stays online. China has roughly 29 reactors under construction, and while they are not all small, inexpensive and modular, it is close to half the global total, and China is marketing its reactors for export under its “Belt and Road” global marketing (and influence) programme. That is the kind of export sale that creates decades of dependency for fuel and maintenance.
It would be wrong, then, to assume America is winning the race to switch reactors on. What America has done is nurture the first of a different kind of reactor – privately built, designed to be stamped out at speed rather than poured once in concrete over a decade. That is the whole American wager – that a swarm of fast-iterating startups will ultimately out-scale a state programme building large plants one at a time.
It is a real bet, and not an obviously foolish one – it is roughly the bet that worked for rockets like those manufactured by SpaceX.
Commercial scramble for power
There is a final irony. For decades, nuclear power was sold to a sceptical public as the answer to global warming. Greens fought over it, governments dithered, and almost nothing got built.
What has finally broken the logjam is not the climate at all, but the simple commercial scramble for power. The reactors that all these years of environmental argument could not summon are now being conjured, in eighteen months, by the demands of a chatbot. That their carbon-free output might also happen to help the planet is, in the pitch decks of the new nuclear founders, very much an afterthought.
No matter. AI is the big game now. And it turns out that the big question is not how intelligent it is or will be or when, but how to feed the players a steady stream of joules.
[Image: Valar Atomics]
The views of the writer are not necessarily the views of the Daily Friend or the IRR.
If you like what you have just read, support the Daily Friend