The Army's $2.2 billion microreactor bet: first customers for a nascent nuclear industry
"We don't expect all five of these companies are going to turn on a reactor in 2028. In fact, they definitely will not."
That is Jeff Waksman, the Army's point man on nuclear microreactors, speaking on August 26 — the same day, in the same breath, that his service announced a program whose stated goal is a fully operational commercial microreactor on a U.S. base by September 30, 2028.
Hold those two sentences next to each other and you have the whole story. The headline says 2028. The man running the program says most of the awardees won't make it.
The headline says 2028. The man running the program says most of the awardees won't make it. That is not a contradiction he's hiding. It's the design.
That is not a contradiction he's hiding. It's the design.
Last week in this space we tracked the space-nuclear push and its looming fuel problem — the fight over who gets rationed high-assay low-enriched uranium when demand outruns supply. This week that same thread lands on the ground, in concrete form: the Army just placed the first real anchor orders for terrestrial microreactors, and it did so while openly conceding it may not have the fuel.
The program is called Janus. The number is up to $2.2 billion, spread across five companies from fiscal 2027 through 2031. Antares goes to Fort Bragg, BWXT to Fort Campbell, General Atomics to Fort Hood, Radiant to Fort Benning, Westinghouse to Fort Drum.
Read as a procurement, this is unremarkable — the Pentagon buys things all the time. Read as market-making, it is the most important thing to happen to the microreactor sector, and the reason has nothing to do with reactor physics.
Start with what the $2.2B is not. It is not a grant, and it is not the bulk of the money. Waksman was explicit: "We are expecting that the majority of the money coming into these reactor programs will be private capital, not government."
The federal dollars cover design, construction, and the first year of operation. After that, the reactors are contractor-owned and operated, and the Army transitions to power-purchase agreements or other long-term offtake — with the explicit end goal of commercial sales beyond the military.
That structure has a name inside the building. Officials said repeatedly it was modeled on NASA's Commercial Orbital Transportation Services program — the arrangement that turned SpaceX from a startup into a launch provider by promising to buy rides it hadn't yet built the rockets for.
This is the thing startups could never buy for themselves. Not capital — capital they can raise. What they couldn't manufacture was a committed anchor customer willing to absorb first-of-a-kind risk, provide a regulatory path, and dangle a follow-on offtake large enough to justify building a factory.
Radiant's numbers show the mechanism working. Its Janus agreement is worth up to $750 million to deploy 15 factory-built, 1-MWe Kaleidos units — and its executives were candid about why that matters: "Operating on military installations first gives us the reliability track record to unlock commercial deployments," said commercial chief Mike Starrett. "The resulting, much larger commercial volume drives down costs, letting the government buy at scale for less."
That is the COTS logic verbatim. The military order is not the prize; it's the credential that unlocks the prize.
Now complicate it, because the same structure that de-risks the sector for investors is ruthless toward the vendors.
These are Other Transaction Authority agreements with fixed-price, milestone-based payments — and Waksman did not soften what that means. "These contractual milestones are very flexible," he said. "So if a company is not performing, if they are just disappointing, not delivering what we're expecting, we can pull future milestones. There are no termination fees."
No termination fees. The Army can walk from any of the five, at any milestone, owing nothing. Waksman warned outright that the list "could be whittled down further."
So the same flexibility that lets the Army promise "a very practical pathway to delivering at least one" reactor by 2028 is exactly what guarantees the other four aren't obligated to hit that date. The program is engineered to let most awardees slip — or fail — without breaking the headline.
And the schedules already tell you who's slipping. BWXT is targeting construction of its 20-MWe reactor at Fort Campbell in late 2028 and operation in the early 2030s. General Atomics is advancing its design "toward potential deployment." Radiant's 15 units are due by 2030. Exactly one reactor needs to be running in September 2028, and the whole architecture is built around finding that one.
Which brings us to the gate that actually matters — and it isn't reactor cleverness.
It's fuel. Nearly all these designs run on TRISO particles made from 19.75 percent HALEU. The United States barely enriches uranium at commercial scale, and the HALEU supply chain is still being stood up by the DOE and NNSA.
Waksman said the quiet part out loud. The microreactor effort is "probably going to push beyond what our supplies are as a nation."
His plan is a two-part bet: prioritize existing stockpile for "our fastest moving reactors," then hope fresh enrichment capacity — which does not yet exist at scale — comes online in time for "the later reactors in this program."
Put that next to last week's space-nuclear timeline and the tension sharpens. NASA wants nuclear power in space in 2028; the Army wants a reactor on a base in 2028; both draw from the same rationed HALEU pool. The Army has now conceded, on the record, that ground reactors are competing for enrichment the nation cannot yet supply.
Fuel is the first gate. The second is regulatory — and it's unusual. These reactors will be licensed not by the Nuclear Regulatory Commission but through an Army-run process, with the service trying to align its standards to the NRC's so vendors can carry the credential into commercial markets later.
That makes the real race clear. The winner of Janus won't be the company with the most elegant heat-pipe or the best liquid-metal loop. It will be whichever vendor first secures qualified fuel and clears the Army's regulatory path.
So here is the distinction worth holding onto. An award is a market signal, not a fueled, licensed, operating reactor. The $2.2B did something genuinely structural — it manufactured the anchor customer the industry could never buy, and it did so at minimal risk to the taxpayer, with private capital carrying the majority and no termination fees on the downside.
But a signal is not criticality. Between the contract and a running reactor sit two gates the Army does not fully control: enrichment it admits it will outrun, and a licensing process it is inventing as it goes.
Waksman himself framed the honest version. "There's a reason no one has actually built a fully operational commercial nuclear microreactor before."
The Army just made it worth trying. Whether anyone crosses the September 2028 line will come down not to who built the cleverest reactor — but to who found the fuel first.