2028 Is an Executive Order, Not a Product Date

Criticality in Idaho, then 470 million dollars

On 4 June a reactor the size of a delivery van reached criticality at Idaho National Laboratory. Antares called it Mark-0: sodium heat-pipe cooled, fuelled with high-assay low-enriched uranium in TRISO compacts, run as a zero-power test. It was the first privately developed non-light-water reactor to go critical in the United States in more than four decades, the first novel design to do so at that laboratory in over fifty years, and the 53rd reactor built on the site since 1951.

Seven weeks later, on 27 July, the company announced a 470 million dollar Series C. Paradigm and Caffeinated Capital co-led it, with Point72 Ventures, Shine Capital and Industrious Ventures alongside. The structure is worth reading closely: 370 million dollars of equity and 100 million of debt. The stated use is the step from a demonstrated reactor to fielded power systems, with a Mark-1 producing electricity in 2027 and first deployments to military installations in 2028.

Every account of this has led on the money. The money is the least interesting part. What matters is which door the reactor walked through, because that door decides whether the 2028 date means anything to anyone outside the United States Air Force.

The permission did not come from the nuclear regulator

Antares was authorised by the Department of Energy. It is the first company in that department's Reactor Pilot Program to receive agency authorisation and complete a fuelled criticality test, and the test ran on a national-laboratory site. The Nuclear Regulatory Commission, which licenses civilian power reactors in the United States, did not issue that permission and was not the body being satisfied.

This is the whole story, and it is missing from every headline. Advanced nuclear has never been held back by physics or by capital. It has been held back by the licensing and siting path, which is measured in years, is adversarial by design, and belongs to a civilian regulator answerable to the public that lives nearby. Antares did not shorten that path. It stepped off it, onto federal research land, under a departmental pilot, for customers who own their own fences.

That is a legitimate route and a real engineering achievement. It is simply not the same achievement as demonstrating that a reactor can be built next to somebody's business, which is the thing a vendor will imply when the milestone reaches a slide deck.

An executive order is setting the schedule

The 2028 date has a source, and it is not a product roadmap. An executive order requires a reactor to be operating at a defence department installation by 30 September 2028. The Air Force runs a programme, Advanced Nuclear Power for Installations, to make that happen. Antares is building toward a deadline that already exists in law.

Read the debt in that light. One hundred million dollars of borrowing inside a 470 million dollar round is the genuinely new signal here, because it means a lender is now willing to underwrite an advanced reactor developer at all. But debt is underwritten against a contracted counterparty, and this counterparty is a government operating under a statutory clock. That is the strongest customer covenant available anywhere. It is not a template that transfers to a merchant project selling power to whoever will buy it.

Jordan Bramble, the company's chief executive and co-founder, put the relationship plainly: the military has been a partner every step of the way. He is describing an advantage, not a coincidence.

One megawatt is the number to plan against

Now the deflating part, and it is the number most likely to be left off the slide. Antares builds microreactors in the range of 100 kilowatts to 1 megawatt. That is the design class, not a first-unit limitation that scale will fix on a schedule anyone has published.

Put that against the loads that actually worry European operators. A mid-sized manufacturing site draws several megawatts. A modest colocation hall draws more. A single row of modern AI servers can approach the top of the Antares range on its own. This class of machine is built for a forward base, a remote radar site, a research station, a place where the alternative is diesel flown in by helicopter. It is a superb answer to that problem and it is not an answer to yours.

Which means that even in the most favourable case, where the technology works exactly as promised and arrives in Europe quickly, it does not remove your dependence on a grid connection. It was never designed to.

What to ask the next vendor who cites 2028

This milestone will be in front of you within months, in a procurement deck or an energy strategy paper, offered as evidence that behind-the-meter nuclear has arrived. Three questions settle it quickly. Which regulator issued the authorisation, and does an equivalent route exist in this country. What is the electrical output in megawatts, and how does that compare with our peak demand. Who is the contracted offtaker on the reference project, and are we able to be that kind of customer.

In Europe the answers are unfavourable and worth knowing early. There is no EU or UK equivalent of a departmental pilot that substitutes for civilian nuclear licensing, and defence land does not carry an exemption from it. France and the United Kingdom each maintain a separate defence nuclear safety authority, but those bodies oversee naval propulsion and weapons programmes, not powering an industrial estate. A reactor on a European site meets the ordinary civilian regime whoever owns the ground.

So the useful work this quarter is unglamorous and entirely within your control. Get your grid connection queue position in writing, with the indicative energisation date and the conditions attached to it. That number is knowable this week, it moves your load in this decade, and unlike a reactor timeline it belongs to a process you can actually influence.

Read next: The Grid, Not the Chips, Now Caps Britain's AI Buildout  |  The Money Buys Test Rigs, Not Permission