⚡ Energy

Two Nuclear Startups Went Critical in 14 Days. The Pentagon’s $477-Per-MWh Diesel Habit Just Got an Expiration Date.

Antares Nuclear and Valar Atomics achieved zero-power criticality in June 2026. An original cost analysis shows diesel generators at forward bases produce electricity at $477–$29,840 per MWh. Mass-produced microreactors target $60. The fleet-wide replacement math: $11.9 billion to power every DOD installation, against a $4 billion annual energy bill.

By Jordan Kessler · Infrastructure & Capital

Seventy-four point six gallons. That is how much diesel a military generator burns to produce one megawatt-hour of electricity at 35 percent thermal efficiency. At the fully burdened cost of delivering that fuel to a forward operating base in Afghanistan, where the Marine Energy Assessment Team pegged delivery at $6.39 per gallon, the electricity rolling out of that generator costs $477 per megawatt-hour. Fly it in by helicopter to a remote outpost and the price climbs past $7,000. At the extreme end, where the Defense Science Board has estimated convoy-protected fuel at $400 per gallon, you are paying $29,840 for one megawatt-hour of electricity to run an air conditioner and a satellite terminal.

On June 4, a company called Antares Nuclear lit up a self-sustaining fission chain reaction at Idaho National Laboratory. It was the first new reactor design to achieve criticality at INL in more than fifty years, and the fifty-third reactor ever built on the site since 1951. Fourteen days later, on June 18, Valar Atomics achieved the same milestone at the San Rafael Energy Lab in Utah with its Ward 250, a helium-cooled high-temperature gas reactor that became the first DOE-authorized reactor built and operated outside the national laboratory system.

Their target levelized cost of electricity? Somewhere between $48 and $78 per megawatt-hour.

That is not a rounding error. It is a category change.

The Race Nobody Expected to Move This Fast

In May 2025, President Trump signed Executive Order 14301, directing the Department of Energy to achieve criticality for at least three advanced reactor concepts by July 4, 2026, the nation's 250th birthday. DOE stood up its Reactor Pilot Program and selected ten companies: Aalo Atomics, Antares Nuclear, Atomic Alchemy (an Oklo subsidiary), Deep Fission, Last Energy, Natura Resources, Oklo, Radiant Industries, Terrestrial Energy, and Valar Atomics. Each company funds its own design, construction, and testing. The DOE provides the regulatory pathway and laboratory infrastructure, bypassing the Nuclear Regulatory Commission's commercial licensing process for early-stage demonstrations.

The timeline was aggressive enough that Reuters called it "extremely ambitious" in February 2026. Getting a first-of-a-kind reactor from design authorization to criticality typically takes the better part of a decade under conventional NRC oversight. Antares did it in less than twelve months. Valar followed two weeks later. The Antares Mark-0 was, per INL Director John Wagner, "the scientific and engineering validation that every subsequent step depends on." The Valar Ward 250, built on private land in rural Utah, proved the model works outside the government's own nuclear campuses, which may ultimately matter more for commercial scalability.

Separately, the Army's Janus Program has selected eight companies for the Advanced Nuclear Power for Installations initiative, aiming to deploy commercially available microreactors at military bases using a build-own-operate model inspired by NASA's Commercial Orbital Transportation Services. Nine military installations have been identified for potential deployment, including Fort Liberty (formerly Bragg), Fort Cavazos (formerly Hood), and Fort Drum. Antares has already been selected for installation at Joint Base San Antonio by 2028.

Two Designs, One Fuel, Zero Pumps

The Antares R1 commercial design outputs between 100 kilowatts and 1 megawatt of electricity, operates for six or more years between refueling, and ships in an integrated transport cradle that includes its own shielding. Its core burns TRISO (Tristructural Isotropic) fuel particles, each the size of a millet seed, containing uranium oxycarbide enriched to 19.75 percent, wrapped in layers of carbon and ceramic that make individual fuel particles essentially meltdown-proof up to 1,800 degrees Celsius. Heat moves from core to heat exchanger through sealed sodium heat pipes with no pumps, no moving parts, and no external power required. If every electrical system on the reactor fails simultaneously, the heat pipes keep working by capillary action alone.

Valar's Ward reactor takes a different approach to the same TRISO fuel, using helium gas as its coolant and graphite as its moderator in a high-temperature gas reactor design that scales to five megawatts. Where Antares optimizes for reliability and transportability at the sub-megawatt scale, Valar's architecture is designed for clustering: the company's long-term vision involves "gigasites" with thousands of reactors providing industrial heat and power for AI data centers and heavy manufacturing alongside military applications. CEO Isaiah Taylor told reporters the Ward reactor would start operating at 100 kilowatts, peak at 250 kilowatts this year, and ramp toward its five-megawatt design capacity over subsequent testing cycles.

Both designs rely on high-assay low-enriched uranium (HALEU), which is enriched to just below the 20 percent weapons-grade threshold. The fuel for Antares was fabricated by BWX Technologies in Lynchburg, Virginia, using specifications matured under the Pentagon's Project Pele, a $300 million prototype reactor program that has served as a technical proving ground for the current microreactor cohort.

The Math: Diesel Generators Are the Worst Power Plant in the DOD Portfolio

The Department of Defense is the single largest institutional energy consumer on the planet. It burns through 85 million barrels of fuel per year, uses over 30 terawatt-hours of electricity across more than 500 installations, and spends upward of $20 billion annually on energy. Of that, roughly $4 billion goes to powering installations, a figure Congress is asked to appropriate at $3.8 billion per year.

Nobody seems to have run the per-megawatt-hour arithmetic on the diesel generators that backstop these installations and serve as the sole power source in forward-deployed locations. A standard military diesel generator operates at approximately 35 percent thermal efficiency. One gallon of diesel contains roughly 38.3 kilowatt-hours of thermal energy. At 35 percent conversion, that yields 13.4 kilowatt-hours of electricity per gallon, meaning you need 74.6 gallons to produce one megawatt-hour.

The per-MWh diesel cost at a standard forward operating base, $477, is already 8 times the midpoint microreactor target of $60 per MWh from the University of Michigan's Nuclear Engineering study, which found optimized microreactors with production tax credits achieving $48 to $78 per MWh. At the tactical edge, the spread widens to 15-to-1. At air-delivery prices, it becomes grotesque. These figures do not include the cost of the generator itself, maintenance, or the lives lost protecting fuel convoys. The Army Environmental Policy Institute documented one casualty for every 24 fuel convoys in Iraq and Afghanistan, and a 2009 GAO report found that generators account for the single largest share of battlefield fuel consumption, burning 357 million gallons per year during wartime versus 26 million in peacetime.

The Fleet Replacement Calculation

Here is the question nobody in the microreactor press coverage has answered with actual numbers: what would it cost to replace diesel generation at every DOD installation with factory-built microreactors?

DOD's 500-plus installations consume 30 terawatt-hours of electricity per year. That averages to roughly 60 gigawatt-hours per installation, requiring an average continuous output of about 6.85 megawatts per site. At the Antares R1 scale of 1 megawatt per unit, that implies roughly 3,425 reactor modules across the entire portfolio. At Valar's 5-megawatt Ward scale, it drops to 685.

A 2024 study in the Proceedings of the Royal Society on the commoditization of nuclear power estimated that a mass-produced 20-megawatt microreactor could achieve capital costs of approximately $3,465 per kilowatt. Applying that estimate to 3,425 megawatts of installed capacity yields a total fleet cost of $11.9 billion. Current annual installation energy spending runs to $4 billion.

Simple payback: roughly three years. Even at the higher expert median of $5,800 per kilowatt from a Carnegie Mellon and Harvard expert elicitation study, the total rises to $19.9 billion, a five-year payback that still compares favorably to a single nuclear aircraft carrier at $13.3 billion. This calculation excludes the operational savings from eliminating fuel convoys, reducing grid vulnerability (the Navy logged over 900 power outages at its installations in fiscal year 2015 alone), and avoiding the human cost of protecting fuel supply lines.

The Strongest Case Against

Nuclear energy has earned its credibility problem the hard way. NuScale Power's Idaho small modular reactor project escalated from $5.3 billion to $9.3 billion before being cancelled outright. Vogtle Units 3 and 4, the only new nuclear construction completed in the United States in decades, came in at $35 billion, double the original estimate, and seven years behind schedule. Octopus Energy's Centre for Net Zero found in 2025 that renewables plus a small amount of natural gas cost 43 percent less than SMRs for a 120-megawatt facility. Every prior "nuclear renaissance" has failed to deliver on cost projections. The entire microreactor business case rests on mass production achieving manufacturing learning curves that have never once been demonstrated for nuclear reactors anywhere in the world.

The HALEU fuel supply chain presents a more immediate constraint. The United States has precisely one domestic source of HALEU production, Centrus Energy's pilot facility in Piketon, Ohio, which began producing small quantities in November 2023. Before that, the only available HALEU came from downblending Russian highly enriched uranium. Urenco USA received NRC authorization for LEU+ production (5 to 10 percent enrichment, below HALEU's 20 percent threshold) at its New Mexico facility in late 2025, with commercial output expected mid-2026, but for full HALEU supply, the scaling challenge remains enormous. Ten companies trying to fuel reactor demonstrations simultaneously strains a supply chain that barely exists.

And zero-power criticality, the milestone Antares and Valar actually achieved, is not electricity. Wagner, the INL director, was careful to clarify: "This is not electricity generation. It is not full-power operation." The Mark-0 reactor has no power conversion or heat removal systems. It proves the physics work. Antares CEO Jordan Bramble has committed to electricity production in 2027 and power to the warfighter in 2028. Those are the milestones where cost projections either hold or collapse.

Limitations of This Analysis

The LCOE estimates for mass-produced microreactors ($48–$78/MWh) are theoretical projections for nth-of-a-kind production runs. First-of-a-kind units will cost three to five times more, and whether factory production achieves nuclear learning curves comparable to solar panels or lithium-ion batteries remains unproven. Our fleet replacement calculation of $11.9 billion assumes every installation needs standalone nuclear generation, which overstates the case: many CONUS installations sit on reliable commercial grids and might need microreactors only for backup resilience rather than primary power. The fully burdened cost of fuel estimates vary widely across sources and theaters, and the DOD has never published an official methodology for calculating them, making all FBCF-based comparisons inherently approximate. We do not account for spent fuel disposal costs, which remain undefined for TRISO/HALEU waste, nor for the insurance, security, and decommissioning costs specific to distributed nuclear installations.

What You Can Do

If you work in defense procurement or installation management: The Janus Program is accepting proposals through the ANPI solicitation. Nine sites are in play. The economics strongly favor any installation where grid reliability is below 99.5 percent or where backup generation consumes more than 50,000 gallons of diesel annually. Run the $477-per-MWh number against your actual fuel logs.

If you work in energy policy: The HALEU supply chain is the binding constraint, not reactor design. Centrus's Piketon pilot and Urenco's LEU+ facility cannot supply ten reactor programs simultaneously. Every month of HALEU production delay is a month of microreactor deployment delay. The DOE's fuel pilot program selections (Oklo, Terrestrial Energy, TRISO-X, Valar Atomics) are a start. They are not sufficient.

If you invest in nuclear equities: BWX Technologies (NYSE: BWXT) fabricated the fuel for both the Antares and Project Pele reactors. Oklo (NYSE: OKLO) holds three Reactor Pilot Program selections and targets an Aurora powerhouse at $70 million for 15 megawatts. Watch whether Antares hits its 2027 electricity production date. That is the real validation event, not the criticality tests.

The Bottom Line

The Pentagon does not have an energy problem. It has a delivery problem that turns $2.19-per-gallon diesel into $477-per-megawatt-hour electricity by the time it reaches the people who need it. Two startups, working under an executive order that bypassed the traditional decade-long regulatory apparatus, just proved their reactor physics work. If mass production achieves even half of its projected cost reductions, $11.9 billion in microreactors replaces a $4 billion annual fuel habit and eliminates the logistics chain that has killed more American service members than many of the enemies those bases were built to fight. None of this is theoretical anymore. What remains uncertain is whether the manufacturing and fuel supply chains can scale before the next conflict makes the answer urgent.