On June 4, at Idaho National Laboratory, a graphite-and-sodium machine called Mark-0 sustained a fission chain reaction for the first time. It is the 53rd reactor built at the INL site since 1951, and by raw thermal output it is among the least impressive — a zero-power criticality test produces watts, not megawatts. But Energy Secretary Chris Wright was not exaggerating when he framed it as a four-decade first: no privately developed non-light-water reactor had reached criticality in the United States since the early 1980s. Antares Nuclear, a company founded in 2023, did it in roughly eleven months from the executive order that created its regulatory pathway.

That last sentence is the one that should hold a physicist’s attention and an executive’s. The neutronics of a small HALEU-TRISO core are well understood; criticality itself was never in scientific doubt. What was in doubt — what has been in doubt since the Clinch River Breeder died in 1983 — is whether the American system could authorize, fuel, and operate a novel reactor concept on a timeline measured in months rather than decades. Mark-0 is the first datapoint saying yes.

What Zero-Power Criticality Actually Demonstrates

It is worth being precise, because the press coverage mostly is not. A zero-power fueled criticality test brings the core to k_eff = 1 — the condition where each fission generation produces exactly enough neutrons to sustain the next — at a power level low enough that the fuel barely warms. No meaningful heat is extracted. No electricity is generated.

What the test validates is the neutronic model of the core: the predicted critical configuration, control element worth, excess reactivity, and the reactivity feedback coefficients near ambient temperature. If your Monte Carlo transport predictions said the core would go critical with a given fuel load and control position, and it does, your cross-section data, your geometry model, and your manufacturing tolerances are all confirmed in one shot. For a first-of-a-kind core fueled with high-assay low-enriched uranium (HALEU, enriched to just under 20% U-235) in TRISO compacts inside a prismatic graphite block, that is a real result — it anchors every subsequent licensing calculation.

What it does not validate: heat transport at power, materials behavior under combined irradiation and temperature, the power conversion system, or load-following dynamics. Antares’ own framing was appropriately sober — CEO Jordan Bramble called it a learning platform that “will continue to generate learning in a safe, rapid setting.” The company’s stated sequence is criticality in 2026, electricity production from a demonstration unit in 2027, and deployed units in 2028. The first commitment landed on schedule, a month ahead of the program deadline.

The Machine: Heat Pipes Instead of Pumps

The Mark-0 and its commercial successor, the R1, belong to the heat-pipe reactor family — the same architecture NASA and Los Alamos matured through the Kilopower/KRUSTY program, which in 2018 became the first new US reactor concept of any kind to reach criticality in decades. The design logic is radical simplification: instead of a pressurized primary coolant loop with pumps, valves, and the failure modes they carry, heat leaves the core through an array of sealed sodium heat pipes.

Each pipe is a self-contained two-phase machine. Sodium evaporates at the hot end embedded in the core (sodium heat pipes operate efficiently in roughly the 600–1,100 °C window), the vapor streams to the cold end, condenses against the power conversion heat exchanger, and the liquid returns by capillary action through a wick. No moving parts, no pumping power, no single-point loss-of-coolant accident — a failed pipe is an incremental capacity loss, not an event, because the array is redundant by design. The R1 closes the cycle with a recuperated nitrogen Brayton loop operating below 300 psi — modest pressures, industrial turbomachinery, nothing exotic on the secondary side.

The fuel carries the safety case. TRISO particles — uranium oxycarbide kernels wrapped in buffer carbon, inner pyrolytic carbon, silicon carbide, and outer pyrolytic carbon — retain fission products to beyond 1,600 °C, well above anything the core can reach in an unmitigated transient at this power density. Each particle is its own pressure vessel. The fuel for Mark-0 was fabricated by BWXT in Virginia, which matters more than it sounds: an actual domestic TRISO production line delivering fuel to an actual reactor is supply-chain validation the entire advanced-reactor sector benefits from. The R1 is specified at 100 kWe to 1 MWe with six-plus years between refuelings, off-grid capable.

The Real Innovation Is the Authorization Pathway

Executive Order 14301, signed in May 2025, directed DOE to reform reactor testing and set a target: at least three advanced reactors critical under DOE authorization by July 4, 2026. The resulting Reactor Pilot Program selected eleven projects — Aalo Atomics, Antares, Atomic Alchemy, Deep Fission, Last Energy, Natura Resources, Oklo, Radiant, Terrestrial Energy, and Valar Atomics among them — and ran them through DOE’s own authorization process rather than NRC licensing, the same legal authority under which national laboratories have always operated test reactors.

This is the part worth understanding mechanically. DOE authorization is not regulatory exemption — Antares had to clear a Nuclear Safety Design Agreement and the full INL safety review apparatus. What it removes is the multi-year, adversarial, precedent-bound NRC licensing sequence for a test reactor whose purpose is to generate the very data that licensing eventually needs. The sequencing inversion is the innovation: build the small dangerous-only-on-paper machine first, measure it, then walk into NRC licensing for commercial units with experimental data instead of simulation alone. Mark-0’s criticality data now feeds the design and licensing basis for every R1 that follows.

Two NSDAs have been approved so far — Antares and Oklo — and Aalo, Valar, and Oklo have construction underway. Whether the program seats three critical reactors by July 4 remains open; that it seated one by June 4 silences the structural objection that it could not be done at all.

Who Is Paying: The Kilowatt-Class Business Case

Antares closed a $96 million Series B in December 2025 — $71 million in equity led by Shine Capital plus $25 million in debt earmarked for equipment, factory build-out, and uranium procurement — bringing total backing past $140 million. The near-term customer is not a utility. It is the Department of Defense.

The Army’s Janus program, established as a program of record in October 2025, targets reactors from kilowatt class up to 20 MWe for installation energy. The Air Force’s ANPI effort selected Radiant, Antares, and Westinghouse for on-base microreactor deployments targeting 2028. The economics that make kilowatt-scale nuclear rational are logistics economics: DOD studies have placed the fully burdened cost of fuel delivered to forward and remote locations at $10 to $50 per gallon once convoy protection and airlift are priced in. Against that benchmark, a factory-built reactor that runs six years without refueling does not need to beat grid power on $/MWh — it needs to beat the diesel convoy, and the diesel convoy is the most expensive electricity in the world that nobody calls electricity.

Factory production is the other half of the model. A 100 kWe–1 MWe unit is a manufactured product, not a construction project — the cost curve, if it materializes, comes from serial production learning rates, not from economies of scale per unit. That is the same wager Oklo, Radiant, and Aalo are making at their respective size points, and it is the inverse of the gigawatt-plant logic that has governed nuclear economics since the 1970s.

The Landscape: A Race With Multiple Finish Lines

Honest comparison requires noting what Mark-0’s competitors are doing at larger scale. Oklo’s Aurora — covered here two weeks ago — targets 15–75 MWe with a liquid-metal fast spectrum and holds the second approved NSDA; its INL build is in progress. Radiant’s Kaleidos (~1.2 MWe, helium-cooled) is queued for testing in INL’s DOME facility. Aalo is pursuing a 10 MWe-class plant aimed squarely at data-center campuses. Westinghouse’s eVinci brings heat-pipe architecture with an incumbent’s supply chain. At the kilowatt-to-single-megawatt point Antares occupies, the honest reading is that Antares is now ahead on demonstrated neutronics — the only one of the cohort with a critical core — while others are ahead on unit power or on NRC engagement.

One distinction matters for anyone mapping this to AI infrastructure: kilowatt-class machines do not power hyperscale data centers — a single GPU hall draws tens of megawatts. The relevance to the AI buildout is indirect but real: every TRISO production run, every DOE authorization precedent, every demonstrated fast pathway de-risks the 10–100 MWe class machines that will sit behind compute campuses, and the defense market provides the early production volume that moves the factory down its learning curve before commercial customers have to bet on it.

The Physics That Matters: Why Small Cores Are Honest Cores

There is a deeper engineering logic to starting at zero power and small scale. Reactor safety is fundamentally about the ratio of stored energy to passive heat-rejection capability. A kilowatt-class heat-pipe core with TRISO fuel has decay heat measured in watts after shutdown and a fuel form stable hundreds of degrees beyond any reachable temperature — the safety case approaches self-evidence, which is precisely why a fast authorization was defensible. As the sector scales from Mark-0’s watts to Aurora’s tens of megawatts to whatever powers the 2030s grid, each step up in stored energy must be matched by demonstrated — not simulated — passive response. The Reactor Pilot Program’s real product is not any single reactor. It is an institutional muscle, atrophied since 1983 and now demonstrably rebuilt: the ability to go from concept to critical core inside a year, generating real data at small scale before committing capital at large scale.

Eleven months from signature to sustained chain reaction. The hard part of the American nuclear renaissance was never the neutrons.

---

Timothy Porritt is founder of Porritt Inc., building AI-powered tools for process safety, engineering compliance, and industrial operations. Based in Salt Lake City, Utah.