A 50-Year-Old Reactor Gets a Second Life: Powering AI with Waste Heat

This summer, a research reactor at the University of Utah will do something it’s never done in five decades: generate electricity for a small AI data center. The TRIGA reactor, originally built by General Atomics for neutron experiments, will produce 2 to 3 kilowatts by capturing its own thermal output through a novel generator from Elemental Nuclear. That’s enough to run a high-performance GPU node handling live AI workloads.

Ted Goodell, the reactor’s manager, calls it a first. “No university reactor has produced electricity before,” he told Gizmodo. “This shows that small, safe reactors could live at data centers, not just in labs.”

Elemental Nuclear’s founder, Mike Luther, sees a bigger principle at work. “Nuclear fission can power the computational systems driving artificial intelligence,” he said in a university announcement.

This is a proof of concept. A full AI data center needs hundreds of megawatts. But the logic is compelling: microreactors—portable, factory-built fission plants—could sit next to server racks, eliminating transmission losses and providing emissions-free baseload power. AI’s electricity demand is straining U.S. grids. Goldman Sachs estimates Big Tech will need 85 to 90 gigawatts of new nuclear by decade’s end. Data center power draw could triple by 2030.

Tech giants are already placing big bets. Microsoft revived Three Mile Island’s Unit 1 for 835 megawatts by 2028. Meta locked in 6.6 gigawatts across projects. Google partnered with Kairos Power for 500 megawatts of molten salt reactors. Amazon committed to over 5 gigawatts from X-energy’s high-temperature gas units. Startups like Last Energy, Aalo Atomics, and Oklo are piling in, with Oklo now public and targeting 14 gigawatts.

China is moving faster, building 26 reactors—including 12 small modular units for 960 megawatts—dedicated to AI training.

Back in Utah, the TRIGA test uses a compact Brayton cycle turbine to convert waste heat into power. The fuel is safe: uranium-zirconium hydride. But commercialization faces hurdles: fuel certification, regulatory approvals, grid connections. Westinghouse’s Vogtle plants ran seven years late and $18 billion over budget. NuScale canceled its first U.S. project on cost.

Still, the pressure is real. AI waits for no one. If a 50-year-old reactor can keep a GPU humming this summer, it’s a small spark for an industry chasing terawatt-hours.

Source: Webpronews