NextFin News - Valar Atomics has moved an advanced reactor out of the purely experimental realm and into a public, inspectable demonstration: the company says its Ward 250 reactor completed a criticality milestone in Utah, and it also showed the system powering a data-center setup tied to Nvidia’s Blackwell AI chip architecture. That is not yet the same thing as proving commercial utility-scale generation, but it is a meaningful step for a sector that has spent years promising faster, smaller and more flexible nuclear power. For AI operators, the point is obvious: the bottleneck is no longer only chips, but power.
On June 24, the U.S. Department of Energy said Valar Atomics’ Ward 250 “successfully completed a zero-power fueled criticality demonstration” at the Utah San Rafael Energy Lab in Emery County, Utah. The department said the experiment marked the first DOE-authorized reactor built outside a national laboratory and that Ward 250 was the second of several advanced reactors expected to reach criticality by a July 4 deadline set in President Trump’s May 2025 executive order. Criticality matters because it proves the reactor can sustain a controlled nuclear chain reaction, the technical threshold that must be crossed before a reactor can generate power.
A week later, on July 1, Valar said it partnered with Nvidia to develop a small data center in Utah and ran a demonstration powering Nvidia’s Blackwell chip architecture. The companies said the setup was the first time a small reactor powered a data center. Nvidia global vice president John Josephakis said the effort was meant to explore “how behind-the-meter, waterless advanced nuclear systems could support future AI factories built for the scale and reliability accelerated computing requires.”
That combination matters because it joins two of the market’s most crowded tradeoffs: AI’s hunger for electricity and the nuclear industry’s long-running claim that small reactors can be deployed quickly enough to help meet that demand. A traditional grid connection can take years. A gas turbine can be faster, but it still depends on fuel logistics and emissions constraints. Valar’s pitch is different. It says a compact advanced reactor can sit close to the load, avoid water use and deliver stable power where large grid infrastructure is slow or expensive to build.
But the distinction between a milestone and a business model remains the key issue. DOE’s language is precise: criticality is a technical achievement, not a commercial operating record. The reactor has shown it can sustain a controlled chain reaction; the company has also shown a data-center demonstration. Neither fact, by itself, proves the economics of repeatable, licensed, long-duration power delivery. Nuclear history is full of projects that passed a milestone before getting stuck in financing, regulation or construction.
The sector’s challenge is that every step after criticality gets harder. A reactor design has to clear licensing, safety-case development, fuel supply, procurement, staffing, insurance and operations. A demonstration can be orchestrated around a single site and a single event. A commercial fleet cannot. That is why investors and policymakers treat firsts in advanced nuclear as important, but not dispositive.
The policy backdrop explains why the event is drawing attention. DOE’s reactor pilot program is designed to compress timelines for advanced-reactor testing and validation. Valar’s milestone came alongside a broader federal push to prove that the United States can move faster on next-generation nuclear. If the program succeeds, it could create a template for a new class of small reactors built around industrial customers rather than around the old utility model.
That is also why Nvidia matters here. The company is one of the largest beneficiaries of the AI build-out, and it has increasingly become a bellwether for how much infrastructure the sector needs around its chips. By pairing with Valar, Nvidia is not just validating a reactor startup; it is also signaling that the AI power problem is becoming central to the next phase of compute expansion. The more data centers move toward dense, always-on workloads, the more valuable steady on-site generation becomes.
“Nine months ago, this was an empty site. Today, there's a critical reactor on it, built and operated by the Valar team,” said Isaiah Taylor, founder and chief executive officer of Valar Atomics. “We met the milestone the executive order set. This reactor was built to make power, and that's exactly where we're headed.”
Taylor’s comment is important because it shows how Valar wants the market to interpret the event: as proof of speed and execution, not just proof of concept. Yet the market will likely view the result through a harsher lens. Can the reactor be replicated? Can it be licensed at scale? Can it produce power consistently enough to matter for a customer whose business depends on uptime and predictability? Those questions are still unanswered.
Why Criticality Is A Real Milestone, But Not The Finish Line
Criticality is the point where a reactor can sustain a controlled chain reaction. That is a genuine threshold, and in advanced nuclear it is often the moment that separates a design from a functioning system. But it should not be confused with a full commercial launch. A reactor can reach criticality without yet being a meaningful contributor to the grid or to a customer’s power mix. For investors, that distinction matters because valuation often rises on the first milestone but ultimately depends on the much harder follow-through.
In the current nuclear cycle, that follow-through has to happen under unusual time pressure. The administration’s July 4 target gave the field a deadline that is more symbolic than commercial, but symbols matter when the sector is trying to prove momentum. Valar now has a visible result it can point to, and the company’s backers can argue that the accelerated process is working. Skeptics, however, will note that a deadline can produce headlines before it produces a scalable business.
The economics are still the central unknown. Advanced reactors promise smaller footprints, modular construction and closer proximity to loads. In theory, that should lower some of the transmission and siting friction that burden conventional projects. In practice, the cost of building, licensing and operating a novel reactor can overwhelm those advantages if the design is not reproducible. One-off engineering success is not the same as an industrial production model.
That is where Valar’s demonstration becomes interesting. The company is no longer asking the market to imagine what a small reactor might do. It is showing a real system tied to a real computing use case. That does not remove the questions around capital intensity, but it does narrow the gap between concept and deployment. For a sector that often lives on promises, a narrow gap is progress.
It is also a signal to competitors. If Valar can keep moving quickly, other advanced-reactor developers will face pressure to show equally concrete validation. The competition is no longer just for permits or public interest. It is for credibility in a race where the first company to demonstrate a practical customer use case may gain an advantage in capital raising, hiring and policy attention.
What The Nvidia Demonstration Changes
The Nvidia link gives the story a sharper commercial edge. AI data centers are among the few demand centers in the economy that can plausibly absorb large amounts of new generation and justify on-site power experiments. Their power needs are enormous, their timelines are compressed and their operators are used to spending heavily on infrastructure when the payoff is scale. That makes them a natural proving ground for advanced nuclear claims.
At the same time, the data-center angle exposes the limits of the demo. The question is not whether a reactor can be adjacent to a chip load for a short test. The question is whether it can support reliable, continuous operations across months and years. Behind-the-meter power is attractive only if it is dependable, maintainable and cheaper or cleaner than the alternatives over time. That bar is high.
Nvidia global vice president John Josephakis framed the experiment as a way to explore “how behind-the-meter, waterless advanced nuclear systems could support future AI factories built for the scale and reliability accelerated computing requires.” The phrasing reflects the two biggest selling points of the concept: reduced water dependence and close-to-load reliability. It also makes clear that the effort is exploratory. Nvidia is testing a possibility, not declaring a standard.
That nuance matters because the AI industry is already under pressure from public concern about power use and water use. New data centers are drawing scrutiny from communities, utilities and regulators. If advanced nuclear can genuinely reduce water consumption while providing firm power, it could become a credible part of the solution set. If it cannot, the industry will keep leaning on grid upgrades, gas backup, renewables and increasingly elaborate cooling systems.
Valar says its reactor is helium-cooled, a design choice that it argues avoids water use. That aligns with a broader push among advanced-reactor developers to tailor systems for industrial applications where water access is limited or politically sensitive. The technical promise is compelling. The operational proof is still early.
What Investors And Policymakers Should Watch Next
The next stage will reveal whether this is a one-time breakthrough or the start of an actual deployment path. The obvious checkpoints are sustained operation, licensing progress, replication of the design and evidence that the data-center model can be repeated without extraordinary handholding. Those are the points where technical success becomes commercial relevance.
Policymakers will also be watching whether the DOE pilot program produces more than a string of headlines. If multiple advanced reactors can reach criticality and then transition toward power production, the government can argue that the United States is finally compressing a cycle that has historically moved too slowly. If the projects stall after the early milestones, the program will be remembered as a proof-of-concept exercise rather than a turnaround.
The broader market implication is straightforward: advanced nuclear is getting closer to the AI infrastructure conversation, but it has not yet crossed the line from demonstration into dependable supply. Valar’s progress is real, and it matters. What remains unproven is whether real can also become repeatable.
The most important takeaway is not that advanced nuclear has solved the power problem. It is that the power problem is now serious enough to make advanced nuclear look plausible. In a market where the gap between possibility and deployment has usually been measured in years, that shift alone is notable.
Explore more exclusive insights at nextfin.ai.
