The nuclear comeback playing out across the country right now is mostly a size contest. Startups want to shrink a reactor down to a shipping container, drop a full one a mile underground in a borehole, or stamp them out on a factory line like delivery vans.<\/p>\n
The pitch underneath all of it is the same: smaller, cheaper, and quicker to build than the giant domed plants everyone pictures. But the smallest serious reactor in America is not for sale, and the people building it have no interest in your money.<\/p>\n
It belongs to the U.S. government, it is roughly the size of a parked sedan, and the entire reason it exists is to let everyone else practice first.<\/p>\n
The machine is called MARVEL, which is a fairly heroic name for an 85-kilowatt reactor. It is going up at Idaho National Laboratory<\/a>, the federal nuclear lab out in the high desert near Idaho Falls, and according to the Department of Energy<\/a> it will be the first new test reactor the lab has finished in more than four decades.<\/p>\n It is cooled by liquid metal, runs without a single pump, and stands about 15 feet tall on a footprint you could cover with one parking space. It will also, once it finally switches on, make roughly enough electricity to run a handful of server racks.<\/p>\n That mismatch, years of work and national headlines for the output of a beefy backup generator, is the whole story, and it makes a lot more sense once you understand what MARVEL is actually for.<\/p>\n MARVEL stands for Microreactor Applications Research, Validation and Evaluation, and the design is genuinely strange the moment you look under the hood. It is cooled by a sodium-potassium eutectic, a liquid metal that stays runny at room temperature, which puts it in the same family as the much bigger sodium-cooled reactor Bill Gates’ TerraPower just started building in Wyoming<\/a>.<\/p>\n The fuel is uranium-zirconium-hydride, a high-assay version of the TRIGA fuel that sits inside university research reactors all over the country. The reactor runs hot, somewhere between 500 and 550 degrees Celsius, and it moves all that heat by natural circulation alone. No pumps.<\/p>\n The hot metal rises through the core on its own, dumps its heat into a set of exchangers, and four small Stirling engines turn the rest into electricity, up to about 20 kilowatts of it.<\/p>\n The numbers are deliberately tiny. A grid reactor is measured in hundreds of thousands of kilowatts of electricity; MARVEL makes about 20, and the whole thing is sized to be hauled around and experimented on, not to anchor a grid.<\/p>\n There is one detail worth savoring: the fuel itself is being built by TRIGA International, a joint venture between France’s Framatome and General Atomics, at a plant in southeastern France, under a contract for 37 fuel elements. So America’s headline government microreactor runs on French-made fuel. Nobody planned the irony, but there it is.<\/p>\n Almost every reactor in the news lately is big, or at least biggish. Canada just lowered a 953-tonne slab of steel and concrete into a shaft<\/a> to start the Western world’s first grid-scale small modular reactor, and even that “small” machine fills two soccer fields and powers around 300,000 homes.<\/p>\n MARVEL sits at the far opposite end of that scale. But the gap in size is not really the point. The point is the category.<\/p>\n MARVEL is not a power reactor at all. It is a test reactor, a tool for running experiments rather than a plant for selling electricity, and INL’s own project page states it plainly: “the MARVEL design is not intended to be a commercial product.” Nobody is going to light up a town with 20 kilowatts.<\/p>\n That distinction is the thing that explains the otherwise baffling economics of spending years and serious public money on a reactor that produces almost nothing.<\/p>\n So why pour all this into a reactor that makes barely any power? Because the federal government wants to hand private companies a real, licensed, operating microreactor to run their experiments on, so they do not each have to build one from scratch just to learn the basics.<\/p>\n Abdalla Abou-Jaoude, the MARVEL lead at INL, put it to the Deseret News<\/a> as “the people’s reactor”: taxpayer-funded, not built for profit, sitting there for industry to learn from.<\/p>\n And industry is already lining up. INL has named its first five end users<\/a>, and the list reads like a map of where small nuclear is headed. Amazon Web Services, plus a separate team of data-center operator DCX USA and Arizona State University, want to test running a data center directly off a microreactor.<\/p>\n That is the same islanded-power idea behind China’s truck-mounted reactor aimed at AI data centers<\/a> and even the reactor America wants on the Moon by 2030<\/a>: power that just sits there and works where the grid does not reach. GE Vernova wants to use MARVEL to prove out running a reactor remotely, with almost nobody on site.<\/p>\n And an oil-and-gas group of ConocoPhillips, NOV and Shepherd Power wants to tap MARVEL’s heat to desalinate the dirty water that comes up alongside oil.<\/p>\n Building something this small and this new was a lot harder than the modest output suggests, and to their credit the INL team has been candid about it.<\/p>\n At the American Nuclear Society’s annual conference on June 1, engineers walked through everything that went sideways<\/a>. MARVEL was supposed to power itself with four off-the-shelf Stirling engines bolted straight onto the core, drop-in and plug-and-play. They were not.<\/p>\n Tweaks to the cooling system brought flow-meter failures, coolant leaks and helium bubbling, and vibration testing got bad enough that engineers worried the building might shake apart. The team pulled the engines back out, returned them to their original packaging with the rubber bumpers still on, and added heat exchangers between the engines and the reactor.<\/p>\n Then there was the control system. MARVEL steers its reaction with rotating drums instead of the vertical control rods nearly every American reactor uses, a technique nobody had actually built in about four decades, so the team had to relearn it from the ground up.<\/p>\n Nearly every part in the drum stack, the bearings, springs, motors and switches, had no nuclear-grade off-the-shelf version, which meant qualifying each one by hand.<\/p>\n John Jackson, who runs the DOE’s microreactor program, summed up the whole experience at the conference: “You change one lever and 97 other levers change in unpredictable ways.” None of that shows up on a spec sheet, which is precisely why INL suspects the binder of hard-won mistakes might end up as valuable as the reactor itself.<\/p>\n MARVEL has slipped before. It was once supposed to be running by the end of 2024. What moved it again was policy: a May 2025 executive order on nuclear energy, plus a 2026 spending bill that funneled billions into advanced reactors, which Abou-Jaoude told the Deseret News pulled the schedule forward by about a year.<\/p>\n The official DOE page still lists dry initial criticality in 2027 and full power in 2028. But at the June conference, the team laid out a tighter plan: reactor assembly starting in August, fuel arriving in October, and the first “dry criticality” test, a near-zero-power run to check the reactor’s physics before any coolant goes in, targeted for late 2026 or early 2027.<\/p>\nA liquid-metal reactor that fits in a parking space<\/h2>\n
953 tonnes on one end of the scale, a sedan on the other<\/h2>\n
The point was never the electricity. It is the practice.<\/h2>\n
Shrinking a reactor turned out to be the hard part<\/h2>\n
Assembly starts in August<\/h2>\n