The U.S. achieved an unprecedented milestone in June 2026 when three distinct advanced nuclear microreactors reached criticality within a single month, marking the first time any country has accomplished this feat. These aren't the massive reactors of decades past. One founder drove the core of his machine to Idaho Falls in the back of a Ford F-150. Now, the AI infrastructure industry is paying close attention, seeing in these compact reactors a potential solution to its most pressing problem: finding enough power to run data centers without waiting years for grid connections.

What Exactly Happened in June 2026?

On June 4, Antares Nuclear's Mark-0 became the first to reach criticality at Idaho National Laboratory, followed by Valar Atomics' Ward 250 on June 22 at the San Rafael Energy Lab in Utah, and Deployable Energy's Unity on June 30, also at Idaho National Laboratory. Each represents a different engineering approach to the same problem: how to build a nuclear reactor small and portable enough to sit next to a data center.

The three reactors differ significantly in their design philosophy. Antares' Mark-0 uses a sodium heat-pipe cooling system with specialized HALEU TRISO fuel. Valar's Ward 250, roughly the size of a minivan, is a helium-cooled high-temperature gas reactor. Deployable's Unity took a pragmatic route, using standard enriched uranium and commercially available materials instead of exotic components, making it easier to source parts and replicate.

These machines are genuinely portable. Deployable Energy, founded less than a year ago by Bobby Gallagher, a former Australian Army soldier who built offshore oil rigs, designed its 1-megawatt Unity reactor to fit inside a shipping container. To prove the point, Gallagher drove the reactor core to Idaho Falls in a pickup truck. One megawatt sounds small, but it's enough to power roughly 800 homes or one modest server room.

Why Does This Matter for AI Companies?

The connection between microreactors and AI infrastructure is straightforward: data centers are consuming electricity at an unprecedented rate. The International Energy Agency estimates data centers consumed about 415 terawatt-hours of electricity in 2024 and projects that figure will more than double to around 945 terawatt-hours by 2030, slightly more than Japan's entire annual electricity consumption today.

Hyperscalers have already begun securing nuclear power directly. Microsoft locked up 835 megawatts from Three Mile Island under a 20-year power purchase agreement, and Amazon secured 1.92 gigawatts from Talen's Susquehanna plant. But those deals involve restarting or extending the life of existing reactors, not building new ones. New nuclear construction traditionally takes a decade or more. Microreactors offer a different path: if you can build one, you can build fifty and stack them next to a compute campus without waiting for transmission line approvals.

The market signal is already visible. Deployable Energy stated in its letter to the Nuclear Regulatory Commission that it has achieved commercial traction with over $10 billion in letters of intent, ranging from data centers to remote island communities. Valar Atomics made the connection even more explicit: on July 1, it generated 100 kilowatts from its reactor, enough electricity to power an Nvidia Blackwell processor, timed precisely with the announcement of a new deal with Nvidia to build a 30-megawatt nuclear data center in Utah.

How to Understand the Path From Demonstration to Commercial Deployment

• Criticality vs. Commercial Operation: Reaching criticality means the reactor core sustains a controlled chain reaction, proving the physics works. It does not mean the reactor is producing meaningful electricity or that the business model is viable. All three June reactors achieved zero-power criticality, a physics demonstration rather than a full operational test.
• Regulatory Approval Timeline: The three reactors were built under the Department of Energy's Reactor Pilot Program and Nuclear Energy Launch Pad, fast-track initiatives designed for demonstration. Commercial deployment requires approval from the Nuclear Regulatory Commission, a different and more rigorous process. Deployable Energy's founder stated the company is likely to apply for a commercial license later in 2026, with NRC review expected to take six to twelve months, and commercial reactors targeted for 2028.
• Manufacturing and Scaling: Building one reactor is a proof of concept. Building fifty requires established supply chains, manufacturing partnerships, and regulatory precedent. Deployable Energy's choice to use standard uranium and commercially available materials rather than exotic fuel types may accelerate this process by reducing sourcing complexity.

"BWXT spent a decade working to design mPower. Our job is to complete its development then design and deploy the first optimised, vertically integrated SMR power plant," said Benjamin Kellie, CEO of Applied Atomics.

Benjamin Kellie, CEO at Applied Atomics

The Broader Nuclear Landscape Reshaping Around AI Power Demand

The microreactor milestone is part of a larger realignment in nuclear energy. Applied Atomics, a nuclear company headquartered in Anchorage, Alaska, recently secured an exclusive license for BWX Technologies' mPower small modular reactor design, positioning the 195-megawatt reactor directly at the intersection of nuclear energy and AI infrastructure. The mPower design is factory-built, transportable by conventional freight, and uses standard low-enriched uranium fuel with a refueling cycle of at least two years.

U.S. electricity demand is forecast to grow at its fastest rate in a generation, with data center construction alone expected to require more than 300 gigawatts of new capacity by 2035. That demand is reshaping the entire SMR competitive landscape. NuScale, once considered the frontrunner for U.S. SMR commercialization, cancelled its flagship Utah project in late 2023 after costs escalated. X-energy, Kairos Power, and TerraPower are each advancing designs at different stages of Nuclear Regulatory Commission review, backed by Department of Energy grants and private capital.

The geopolitical dimension adds urgency. Several allied nations, including Canada and Poland, are actively evaluating SMR imports as part of energy sovereignty strategies. Applied Atomics' license covers deployments beyond North America, leaving open the prospect of export agreements that would bring it into competition with Russian ROSATOM and Chinese CNNC designs currently courting the same markets.

What Experts Say About the Timeline and Realistic Expectations

Industry observers are cautious about overstating what the June criticality achievements mean. Emily Tucker, vice president on the energy team at advisory firm Capstone, cautioned that the criticality milestones matter as technology demonstrations but do not signal that commercialization is near. Alison Hahn, former head of advanced reactors at the Department of Energy and now at the Nuclear Energy Institute, noted the pilot program is explicitly for demonstration reactors, not commercial deployment.

The contrast with what is happening at Darlington in Ontario illustrates the difference between demonstration and grid-scale deployment. Ontario Power Generation is building a full grid asset with a basemat weighing close to 953 metric tons, measuring 37 meters across and requiring one of the world's largest crawler cranes to lower it into place. That is the traditional path to nuclear power. Microreactors represent a fundamentally different bet: smaller, modular, and deployable without years of site preparation.

The commercial clock will start running once the Nuclear Regulatory Commission begins its formal review process. For capital allocators watching the AI infrastructure build-out, the strategic calculus is shifting: reliable baseload power is becoming a bottleneck asset. SMR licenses, once niche clean-energy plays, are now proxy bets on where the next generation of compute clusters will be sited. The June 2026 microreactor milestone was a physics demonstration. The real test will be whether these designs can move from the lab to the data center floor within the next two years.

" }