Valar Atomics Reactor Powers Nvidia AI Chip in US First
Valar Atomics successfully powered an Nvidia AI chip with its Ward 250 nuclear reactor in Utah, marking a US first. This demonstration, combined with a joint feasibility study for a 30-megawatt water-free AI computing facility, highlights the potential for off-grid, sustainable power for data centers. Other tech giants like Google, Microsoft, Meta, and Amazon are also exploring nuclear power solutions for their energy-intensive AI operations, signaling a significant shift in the data center industry's approach to energy supply.
A California nuclear startup, Valar Atomics, achieved a US first by successfully powering an Nvidia Blackwell AI chip with its Ward 250 test reactor. The demonstration took place on July 1, 2026, at the San Rafael Energy Research Center in Emery County, Utah, where the reactor generated 100 kilowatts of thermal energy to run a desktop machine and host a website. This event symbolizes a significant step towards off-grid, water-free power solutions for the rapidly growing AI industry.
Following the demonstration, Valar Atomics and Nvidia announced a joint feasibility study for a 30-megawatt, entirely water-free AI computing facility. This proposed facility would bypass public electricity grids and rely on Valar's helium-cooled, TRISO-fueled reactors, paired with Nvidia's new DSX data center design featuring closed-loop liquid cooling. This partnership suggests a future where major tech companies like Nvidia could operate AI factories independently of conventional utility infrastructure and large water demands, addressing key constraints in AI expansion.
While the current demonstration involved modest power output, the rapid progress from criticality to usable electricity in 13 days highlights Valar's advanced high-temperature gas-cooled reactor technology. The move by tech giants into nuclear power is underscored by other partnerships, including Google with Kairos Power, Microsoft with Three Mile Island, and Meta and Amazon pursuing their own nuclear arrangements. Despite the promise, industry observers caution that commercial viability, stringent Nuclear Regulatory Commission (NRC) licensing, and scaling from proof-of-concept to hundreds of megawatts remain significant hurdles for advanced nuclear energy to fully meet the immense demands of frontier AI clusters.