Energy

Three privately built microreactors reached criticality inside one month

Three privately built microreactors reached criticality under the DOE pilot program before July 4, and one powered an Nvidia chip. What the milestone changes for compute buyers, and what it does not yet prove.

EnergyBy Servola Tech Desk2026-07-054 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Three privately built microreactors reached criticality inside one month

Key takeaways

Three privately funded advanced microreactors reached criticality under the US Department of Energy Reactor Pilot Program before the July 4 deadline, the first time any country has done so with three distinct designs in a single month.
Deployable Energy took its Unity reactor from project kickoff to criticality in roughly 150 days on a single-digit-million-dollar budget, an order-of-magnitude compression of nuclear build timelines.
Valar Atomics wired its Ward 250 reactor to an Nvidia DGX Spark on July 1, the first US instance of an advanced reactor supplying a live AI compute load.
Criticality is a physics milestone, not a commercial power plant. The demonstration ran near 100 kilowatts of thermal output; the advertised 30 MW data center in Utah is still a plan with no schedule or binding order.
Helium cooling in these gas reactors targets near-zero water use, a direct answer to the water constraint now blocking data center siting in dry regions.

What actually happened

Between early June and July 1, three privately funded advanced reactors reached initial criticality under the US Department of Energy Reactor Pilot Program. Antares Nuclear brought its Mark-0 unit critical at Idaho National Laboratory, Valar Atomics followed with its Ward 250 reactor at the San Rafael Energy Lab in Utah, and Deployable Energy achieved criticality with its Unity unit at Idaho late on June 30. The DOE announced the third milestone on July 1 and stated that the United States had become the first country to reach criticality in three distinct advanced microreactor designs within a single month.

The program traces to a May 2025 executive order that set a July 4, 2026 deadline for at least three test reactors to go critical using DOE authorization rather than the standard Nuclear Regulatory Commission licensing path. All three units met it. Valar Atomics is also the first DOE-authorized reactor built and operated outside the national laboratory system, which is the part that matters commercially: the authorization route it proved is the one a private operator would use to site a unit on an industrial or compute campus.

The Nvidia demonstration, and what it does not prove

On July 1 Valar connected its Ward 250 reactor to an Nvidia DGX Spark, a desktop-class AI machine built on the Grace Blackwell superchip, plus a web server running a demonstration page. This is the first US case of an advanced reactor supplying a live AI compute load, and it is a genuine engineering marker. It is also very small. The reactor ran at roughly 37 percent of its intended output, near 100 kilowatts of thermal power, and the DGX Spark it fed draws about 240 watts. The advertised 30 MW nuclear-powered data center in Utah remains a plan: no construction schedule, no investment figure, and no binding hardware order have been published.

The honest read is that physics has been demonstrated and economics have not. A reactor that splits atoms in a shielded test bay is a long way from a plant that delivers firm megawatts through a grid interconnection, a fuel supply chain, and a decade of operating cost. Treat the criticality milestone as evidence that the timeline is compressing, not as evidence that dispatchable nuclear power is available to buy today.

Why the timeline compression is the real signal

The number worth carrying into a planning meeting is Deployable Energy's. It took its Unity reactor from project kickoff at Idaho National Laboratory to criticality in about 150 days on a single-digit-million-dollar budget. Conventional advanced-reactor programs measure the same distance in years and hundreds of millions. The compression comes from the DOE authorization path, factory-style fabrication, and reactor cores small enough to move on a truck. Valar's Ward 250 is a 5 MW helium-cooled, TRISO-fueled gas reactor roughly the size of a minivan; Deployable's Unity is a one-megawatt unit that fits in a shipping container.

For an operator, that changes the shape of the bet. The constraint on new compute and new industrial load is increasingly firm power on a schedule, not capital. A reactor class that can be authorized and built in months, sited without water, and co-located with the load is the first credible answer to that constraint that is not another multi-year grid interconnection queue. None of these units is selling commercial power yet. But the group that would sell it just proved the build cycle is measured in months.

What an owner should do now

Do not repower a facility around a reactor that has run for a day. Do treat firm-power procurement as a live strategic question rather than a settled one. If your growth plan depends on new compute or new industrial load in a grid-constrained or water-constrained region, the set of credible options widened this month, and the vendors in it now have a demonstrated authorization route and a build cycle measured in months rather than years.

The practical move is to open the conversation early and cheaply: ask the microreactor developers what a co-located unit would require in permitting, fuel supply, and cost per firm megawatt, and get those answers on paper before you commit a multi-year grid application. The companies that win the next round of compute and industrial buildout will be the ones that treated firm power as a design input, not an afterthought.

Frequently asked questions

Does criticality mean these reactors are producing usable power now?

No. Criticality means a reactor has achieved a self-sustaining nuclear chain reaction, which is a physics milestone. The Valar demonstration ran near 100 kilowatts of thermal output and powered a desktop-class Nvidia machine. Commercial delivery of firm megawatts through a grid interconnection, with a fuel supply chain and a decade of operating cost, is a separate and much larger step that none of these units has taken.

Why does helium cooling matter for data centers?

These are gas-cooled reactors that use helium rather than water, targeting near-zero water use in the reactor loop. Water availability is now a hard constraint on data center siting in dry regions. A firm-power source that does not draw cooling water removes one of the siting blockers that has been forcing compute projects into fewer viable locations.

How is this different from the nuclear deals Meta, Amazon, and Google have signed?

The hyperscaler deals are power purchase and financing agreements tied to large conventional plants or to reactors that are years from delivery. This milestone is a physics and authorization proof for small factory-built units on a compressed build cycle. The two trends point the same direction, but this one lowers the barrier to a private operator siting a reactor next to its own load rather than contracting for grid power.

Firm power on a schedule is the constraint that now governs new compute and new industrial load. The set of credible answers to it grew this month, and the operators who treat that as a planning input rather than a headline will be the ones siting capacity when their competitors are still waiting in a queue.

NuclearSMRValar AtomicsNvidiaData CentersDOE

Aramco Just Bet 800 Million on Cheaper AI

Together AI raised 800 million dollars at an 8.3 billion valuation, led by Aramco Ventures, to run open-weight models cheaply. What the inference-layer bet means for owners weighing AI costs.

4 min read

Decisions

When Your Supplier Funds Your Own Demand

Nvidia now lends and buys back the GPUs it sells. A calm owner-facing lesson on separating real demand from vendor-subsidized demand before you extrapolate.

2 min read

Digital Sovereignty

The EU Council Revived the Rule That Scans Private Messages

The EU Council revived the lapsed rule that lets providers scan private messages with AI. For owners, confidential communication is now a governance choice.

2 min read14 views

← All articles