Nuclear Firm Becomes First in US to Produce Power From Advanced Reactor

A California nuclear startup called Valar Atomics has connected an advanced reactor to an Nvidia AI chip and generated real power from it, marking the first time a next generation reactor has run computing hardware in the United States.

What Actually Happened in Utah

At its site in Utah, Valar linked its Ward 250 reactor to an Nvidia Blackwell chip during a demonstration this past Wednesday. The setup produced enough electricity to temporarily host a website, hardly a meaningful workload, but symbolically significant. The two companies also said they would work together to explore nuclear powered AI systems going forward, though no specific projects or timelines were disclosed.

Why a Trickle of Power Matters

The amount of electricity involved was tiny, barely enough to keep a basic web page running. Yet the milestone lies in the connection itself. It is the first documented case of a next generation reactor in the US supplying power directly to AI hardware. Researchers and industry watchers see this as an early proof point for an idea that has gained traction as AI data centers strain existing power grids: pairing compact nuclear plants directly with computing facilities rather than relying solely on utility connections.

How the Reactor Reached This Point

Valar's reactor reached what nuclear engineers call criticality last month, the moment a fission reaction becomes self sustaining without needing external triggers to keep going. Reaching criticality is a foundational step for any reactor, since it confirms the core design can maintain a steady nuclear reaction under controlled conditions. From there, the company moved to actually drawing usable electricity off the reactor and routing it to a chip, which is a separate and more complex engineering task than simply sustaining the reaction itself.

Where the Broader Industry Stands

Valar is one of several companies chasing next generation nuclear designs that rely on new materials and reactor configurations meant to boost safety margins and efficiency compared with conventional plants. The pitch to the AI industry is straightforward: data centers need enormous, reliable power, and small advanced reactors could theoretically be sited near or even on campus with computing infrastructure. But the sector remains young. No advanced reactor design has yet received the approvals needed to be sold commercially in the US, and Valar's demonstration, while notable, was a controlled test rather than a commercial deployment.

What Still Needs to Be Proven

A single demonstration powering a basic website does not answer the harder questions facing this technology. Scaling up from a trickle of electricity to the sustained, high volume power that AI training clusters demand is a very different engineering challenge. Regulatory approval, cost per kilowatt hour, construction timelines and long term safety data all remain open questions that no company in this space has fully resolved. The agreement between Valar and Nvidia to jointly explore nuclear powered AI systems is still described as exploratory, without committed capacity or a public roadmap.

Can Advanced Reactors Actually Keep Pace With AI Power Demand

The Utah test shows a technical pathway exists, but it does not yet show that advanced reactors can be built fast enough or cheaply enough to matter at the scale AI companies need. Whether Valar or its rivals can turn this kind of demonstration into commercial, grid ready capacity within the next several years remains genuinely uncertain.