A small nuclear reactor has reached criticality, and the company behind it plans to use it to power AI computing.
According to Tech Times, a nuclear microreactor has cleared criticality — the point at which a reactor sustains its own nuclear chain reaction — and Crusoe intends to test it as a power source for an "AI factory" data center.
A report carried on MSN adds detail: Crusoe and Aalo Atomics have partnered to run what would be the world's first GPU workloads on a sodium-cooled microreactor at Idaho National Laboratory, with the test slated for 2027.
A few plain-language translations are worth making. A microreactor is exactly what it sounds like — a nuclear reactor small enough to be built in a factory and delivered to a site, rather than constructed over years like a conventional power plant. "Sodium-cooled" refers to using liquid sodium instead of water to carry heat away from the reactor core. GPUs are the specialized chips that do the heavy lifting for training and running AI models, and they are enormously power-hungry — which is why the industry keeps circling back to nuclear.
Idaho National Laboratory is a federal research site, so this is a demonstration rather than a commercial deployment. Criticality is a real milestone, but it is a step on the path, not the finish line — the actual pairing of reactor and GPUs is still years out, per the 2027 date cited.
It matters because AI's electricity appetite is now outrunning the grid in many places, and a working reactor-plus-GPU demonstration would show whether purpose-built nuclear power can be plugged directly into data centers instead of competing with everyone else for existing supply.