Pentagon selects three microreactor companies for Air Force bases as military nuclear programme advances toward 2030

Summary: The Pentagon has narrowed its Advanced Nuclear Power for Installations (ANPI) programme from eight companies to three, advancing microreactor deployment at Buckley Space Force Base (Colorado) and Malmstrom Air Force Base (Montana) by 2030. The original eight vendors included BWXT, Oklo, X-energy, Kairos Power, Radiant, General Atomics, Westinghouse, and Antares. The commercially owned reactor model, backed by Executive Order 14299 and $125 million in Congressional funding, addresses military grid vulnerability while serving as a proving ground for reactors that could also power AI data centres.

The Pentagon has narrowed the field for its programme to install microreactors at US Air Force bases, selecting three companies from an original pool of eight to advance toward deployment, Bloomberg reported on Tuesday. The down-selection is the most concrete step yet in the Advanced Nuclear Power for Installations programme, known as ANPI, a joint effort between the Defense Innovation Unit, the Air Force, and the Army that aims to make military bases energy-independent by replacing their reliance on a civilian power grid that is increasingly vulnerable to cyberattacks, extreme weather, and the cascading demands of AI-driven energy consumption.

The programme began in April 2025, when the DIU selected eight companies to develop microreactor proposals: Antares Nuclear Energy, BWXT Advanced Technologies, General Atomics Electromagnetic Systems, Kairos Power, Oklo, Radiant Industries, Westinghouse Electric Company, and X-energy. Each was tasked with designing commercially owned and operated reactors that could be built on military land, licensed through the Nuclear Regulatory Commission, and maintained by the vendor throughout their operational life. The military would buy the electricity without owning the reactor, a model designed to accelerate deployment by sidestepping the decades-long procurement cycles that have historically paralysed defence infrastructure projects.

Why Air Force bases need their own power plants

The Department of Defense consumes more than 30 terawatt-hours of electricity annually across more than 500 installations, making it the single largest energy consumer in the US government. The overwhelming majority of that power comes from the civilian grid. That dependence is now treated as a strategic vulnerability. Cyberattacks on US energy infrastructure have increased by roughly 70% in recent years. The grid itself is under growing strain from data centre construction, with the International Energy Agency projecting that data centre electricity consumption will exceed 1,000 terawatt-hours globally by the end of 2026. Military bases that host missile fields, space surveillance operations, and nuclear command infrastructure cannot afford to compete with AI training clusters for grid capacity.

Two Air Force installations have been selected as the first deployment sites. Buckley Space Force Base in Aurora, Colorado, hosts the Aerospace Data Facility, one of the Department of Defense’s primary satellite ground stations. Malmstrom Air Force Base in Great Falls, Montana, oversees 150 Minuteman III intercontinental ballistic missiles spread across 13,800 square miles of Montana prairie. Both bases require uninterrupted power for operations that are, by definition, existential. Nancy Balkus, the Deputy Assistant Secretary of the Air Force for operational energy, has said that energy security at these installations is not an efficiency question but a readiness question. The target is operational microreactors at both sites by 2030.

The technology

Microreactors are nuclear fission reactors that typically produce between one and 20 megawatts of electrical power, small enough to fit on a few truck trailers and large enough to power a military base or a small data centre. They use advanced fuel forms, most commonly TRISO (tristructural isotropic) particles encased in ceramic and graphite shells that can withstand extreme temperatures without melting down. Several of the ANPI candidates use high-assay low-enriched uranium, or HALEU, which is enriched to between 5% and 20% uranium-235, higher than conventional reactor fuel but well below weapons grade.

The designs vary significantly. BWXT’s Project Pele, developed separately for the Army, is a 1.5-megawatt transportable reactor that completed initial testing at Idaho National Laboratory and uses TRISO fuel with a gas-cooled design. In February 2026, the Pentagon airlifted a five-megawatt microreactor prototype from California to Utah, the first military nuclear airlift, demonstrating the transportability that makes these systems attractive for expeditionary and remote base operations. Oklo, whose chairman is OpenAI chief executive Sam Altman, designs a compact fast reactor called the Aurora that uses metallic fuel and targets both military and commercial applications. X-energy, which went public with Amazon’s backing, is developing the Xe-100, an 80-megawatt high-temperature gas-cooled reactor that uses TRISO-X fuel pebbles. Kairos Power is building a fluoride salt-cooled reactor. Radiant Industries, founded by former SpaceX engineers, is developing a portable one-megawatt reactor designed for rapid deployment.

Only NuScale Power has received full design certification from the Nuclear Regulatory Commission for a small modular reactor, but NuScale’s design is a 77-megawatt light-water reactor, far larger than what ANPI requires. The ANPI programme’s commercially owned model means that vendors will need to secure their own NRC licences for reactors sited on military land, a regulatory path that has not been tested at this scale. The Atomic Energy Act provides a military exemption for reactors operated by the armed forces, but the ANPI model explicitly uses commercial operators, which means NRC jurisdiction applies.

The policy architecture

The programme sits within a broader policy push that has acquired unusual bipartisan momentum. Executive Order 14299 explicitly links nuclear power to AI infrastructure at military installations, directing federal agencies to accelerate the siting and permitting of advanced reactors. The ADVANCE Act, signed into law with an 82-to-14 Senate vote, streamlines NRC licensing for advanced reactor designs. Congress has appropriated $125 million for military microreactor development. The Army’s separate Project Janus programme is evaluating nine additional bases for microreactor deployment.

The convergence of military energy security and commercial AI infrastructure is not coincidental. The Department of Energy has identified 16 federal sites, many adjacent to existing nuclear facilities, as candidates for data centre construction. Nuclear-powered AI data centres are attracting dedicated venture capital, with Valar Atomics raising $450 million at a $2 billion valuation to build small modular reactors purpose-built for AI workloads. The same microreactors that power a missile field in Montana could, in a commercially licensed configuration, power an AI training cluster in Texas. The ANPI programme is a military procurement initiative, but it is also a proving ground for the reactors that the technology industry hopes will solve its energy problem.

What stands in the way

The 2030 deployment target is ambitious by nuclear standards. No advanced microreactor design has completed NRC licensing. HALEU fuel supply remains constrained, with Centrus Energy as the only domestic commercial producer and Russia historically the dominant global supplier, a dependency that sanctions have complicated. Community opposition to nuclear facilities, even small ones on existing military bases, has slowed previous projects. The cost economics of microreactors at the one-to-20-megawatt scale remain unproven in commercial operation, though the commercially owned model shifts that financial risk from the Department of Defense to the vendors.

The nuclear waste question also persists. Microreactors produce far less spent fuel than conventional power plants, but the United States still lacks a permanent repository for any nuclear waste. Advanced fuel forms like TRISO are more proliferation-resistant and easier to store than conventional spent fuel rods, but “easier” is relative in an industry where waste management has been a political impossibility for four decades.

The broader debate over nuclear power and AI has tended to focus on fusion, the technology that is always 20 years away, or on gigawatt-scale conventional plants that take a decade to build. Microreactors occupy a different niche: small enough to be manufactured in a factory rather than constructed on site, simple enough to operate with minimal staffing, and modular enough to scale by adding units rather than building larger. The military is betting that this niche is real. The down-selection from eight companies to three means the Pentagon has now seen enough proposals to decide which designs are credible and which are not. The three that remain have roughly four years to prove that a nuclear reactor can be as reliable, and as unremarkable, as the diesel generators that military bases currently keep for backup power. If they succeed, the implications extend well beyond the fence line of an Air Force base.