The AI Buildout Needs Reactors.
The Reactors Need a Fuel One U.S. Company Is Already Shipping.

Research Desk · August 31, 2026

TRISO fuel research at a U.S. national laboratory. Image via Wikimedia Commons. 

Investors have watched the artificial-intelligence story from a strange angle this year. The chips got the headlines. Then the power bills did.

Data centers are now drawing electricity on the scale of a mid-sized country, and the companies building them have run out of patience waiting in utility interconnection queues. So they went shopping for their own reactors.

That part has been on the front page for a year. Here is the part that has not: those reactors do not run on ordinary uranium fuel, and until very recently no independent American company was producing the fuel they do run on at commercial scale.

Last week that changed, and the company that changed it filed its first set of public numbers.

The Cartridge, Not the Printer

Standard Nuclear (NYSE: STDN) does not build reactors. It makes the fuel that goes inside them, a format called TRISO.

A TRISO particle is a speck of uranium roughly the size of a poppy seed, wrapped in three ceramic and carbon shells. Those shells are the containment. The fuel is engineered to hold its own fission products even if everything around it fails, which is precisely why the new microreactor designs can be small, factory-built, and sited near a data center.

Consider the economics the way you would a printer and its ink. The reactor is the printer: bought once, financed over decades, photographed for the press release. The fuel is the cartridge, bought again and again for as long as the machine runs.

Cartridge businesses are usually the better business. This one comes with an unusual moat, because the ink is genuinely hard to make.

The uranium these designs need is high-assay low-enriched, and industry figures put annual worldwide output of it near 900 kilograms, against tens of millions of kilograms of conventional reactor-grade material. It also costs several times more per kilogram.

America spent decades letting its enrichment capacity atrophy while leaning on Russian supply, a dependency Washington has since legislated its way out of. The Department of Energy has been rationing its own stockpile to a short list of reactor developers to keep their demonstration projects moving.

So the constraint in advanced nuclear is not reactor designs. There are dozens of those, well funded and racing each other. The constraint is the ceramic-coated seed that goes inside them, and the company on the other side of that shortage is the one getting paid to solve it.

The Order Book Grew Six-Fold in Five Months

The August 26 report was the first hard evidence that this is a manufacturer rather than a laboratory. Revenue came in at $4.7 million against $0.6 million a year earlier, with $3.1 million of it from actual fuel deliveries.

More telling than the growth rate was the margin. Gross profit was $3.2 million, a 67 percent gross margin, against a gross loss in the same quarter a year ago. Early-stage manufacturers rarely print that on their first commercial shipments.

During the quarter the company delivered 50 kilograms of uranium as TRISO fuel, then shipped the balance shortly after quarter end to complete a full reactor core for Radiant Industries and its Kaleidos microreactor. Management describes it as the first complete core of commercially produced TRISO fuel supplied by an independent American manufacturer.

The order book moved faster than the revenue, which is the number to watch in a business like this. Funded backlog, meaning contracts a customer has actually paid to advance, went from $8.2 million at the end of March to $61.9 million at the end of June, then to $119.3 million after an August fuel supply agreement with Antares Nuclear.

Total contract backlog followed the same curve: $91.3 million in March, $241.5 million in June, $576.9 million after Antares. Behind that sits a qualified pipeline management put at roughly $696.3 million, none of which is counted in backlog yet.

The balance sheet is the part income-minded readers will appreciate. The July listing brought in about $137.7 million net, leaving roughly $239.9 million of pro forma cash against no debt. The plants are already substantially paid for.

The Capacity Is Built and Waiting

Two production plants, SN-TN in Oak Ridge, Tennessee, and SN-ID in Idaho, are substantially complete. Each is designed to start at up to one metric ton of annual TRISO output and scale toward 2.5 metric tons, for combined capacity near five metric tons.

Set that against a global HALEU supply measured in hundreds of kilograms and the scale of what has been built becomes clear.

The Department of Energy has approved the preliminary safety analysis for both sites, and module commissioning is underway. Management is targeting authorization to operate in the fourth quarter of this year, which is the gate that turns funded backlog into recognized revenue.

Beyond that, a joint venture with Framatome in Richland, Washington, cleared its regulatory review in June and is slated to begin production in 2027. Board-level policy experience was added in August with the appointment of a former chief counsel for nuclear policy at the Department of Energy.

Risk Factors

A story this clean deserves an honest ledger on the other side. Here is what has to go right, and what happens if it does not.

The Bottom Line

This is a picks-and-shovels position on the AI power buildout, one step further back in the chain than the utilities and the reactor developers everyone else is bidding on.

The appeal is that the choke point has been identified, the capacity to relieve it has been built and paid for, and the order book grew six-fold in five months while the revenue was still small. That sequence, orders first and revenue after, is what an early industrial ramp actually looks like from the outside.

What it is not is a core retirement holding. There is no dividend, no earnings yet, and a valuation that has borrowed several years of the future. Position size is the entire discipline here.

For readers who want the nuclear side of the AI story, a small speculative sleeve is the sensible expression, sized so a delayed authorization is an annoyance rather than a problem. Patient money can also wait for that authorization to be granted and pay up afterward. Certainty costs something, and it is often worth the price.