Ray Price
Valar Atomics has powered an Nvidia (NVDA) Blackwell chip using electricity generated by its Ward 250 advanced reactor, marking the first time a next-generation reactor has produced usable power for an AI workload on US soil.
What Happened at the Utah Site
The demonstration took place Wednesday at Valar's test site in Utah, where engineers connected the Ward 250 reactor directly to a Blackwell chip and used the resulting power to briefly host a website. The output was described as a trickle, nowhere near the scale needed for commercial AI data center operations, but the symbolic significance is hard to overstate: a privately developed advanced fission reactor, running independent of the traditional light water reactor fleet, delivered electrons to one of the most in demand AI accelerators on the market.
Valar and Nvidia also announced a joint agreement to explore nuclear powered AI systems going forward, though neither company detailed specific technical milestones, timelines, or capacity targets for that collaboration. The framing suggests early stage coordination rather than a committed buildout.
From Criticality to Grid Connected Chips
The Ward 250 reached criticality last month, meaning it achieved a self-sustaining fission chain reaction for the first time. That step is a prerequisite for any reactor before it can generate meaningful thermal or electrical output, and the compressed timeline between criticality and a chip powering demonstration signals an aggressive testing cadence by Valar relative to typical nuclear licensing and commissioning schedules in the US.
Advanced reactors like the Ward 250 differ from conventional pressurized or boiling water reactors in their choice of coolant, fuel form, or moderator material. Valar is among a cohort of companies pursuing designs that use alternative materials intended to improve thermal efficiency, passive safety margins, or siting flexibility compared with legacy plants. The specific technical configuration of the Ward 250, including its coolant and fuel type, was not detailed in connection with this demonstration.
Why the AI Sector Is Watching Nuclear Startups
Power availability has become a binding constraint on AI infrastructure buildout, with data center operators increasingly courting nuclear, gas, and even fusion startups to secure firm, carbon free electricity that can scale alongside chip deployments. Meta's reported plans for a cloud business selling AI computing power, and ByteDance's selection of Brazil for its largest data center outside China, both point to the same underlying dynamic: compute growth is outpacing the ability of conventional grids to deliver reliable capacity where it's needed.
Nvidia's participation in this demonstration, even at trickle scale, signals that chipmakers see direct value in de-risking future power supply chains rather than waiting passively for utilities or independent power producers to catch up. Pairing a reactor developer directly with a chip demonstration is a marketing and validation exercise as much as an engineering one, but it also creates a data point that other AI infrastructure players will reference when evaluating advanced nuclear as a viable power source.
Where the Technology Still Stands
No advanced reactor design is currently commercially available in the US, and Valar's demonstration does not change that status. The gap between a criticality milestone, a trickle power demonstration, and a licensed, revenue generating commercial reactor remains substantial, spanning regulatory approval, fuel supply chain maturity, manufacturing scale, and construction cost discipline that have challenged the broader advanced nuclear sector for years.
The companies did not disclose a timeline for scaling the Ward 250 beyond this demonstration, nor did they specify what output levels or reliability thresholds would be required before nuclear power could realistically support production AI workloads rather than a single chip in a controlled test environment.
