Valar Atomics has completed a zero-power fueled criticality demonstration of its Ward 250 reactor at the Utah San Rafael Energy Research Center, a small but significant step toward the startup’s goal of supplying electricity to AI data centers with minimal water consumption. The test, conducted under the Department of Energy’s Reactor Pilot Program, makes the Ward 250 only the second advanced reactor to reach criticality in the United States. The achievement arrives as technology companies, including Nvidia, race to find power sources that can handle the enormous and growing energy demands of artificial intelligence without straining water supplies or running into traditional siting constraints.
Why data-center power demand is forcing a nuclear bet
AI training clusters and inference farms now require tens of megawatts of continuous, reliable electricity, and the facilities that house them consume vast quantities of water for cooling. Natural gas plants that typically serve these loads also need water for steam cycles and emissions controls. That twin pressure, electricity plus water, has pushed data-center operators to explore alternatives that can deliver steady output on a small physical footprint without the water overhead.
Valar Atomics is positioning the Ward 250 as that alternative. The reactor is designed to be compact enough to sit near a data center rather than requiring a remote generating station and long transmission lines. According to a Department of Energy announcement, the Ward 250 achieved its criticality milestone under the federal Reactor Pilot Program, a pathway that gives qualifying advanced reactor designs access to national laboratory infrastructure and streamlined federal review. The DOE described the Ward 250 as the second advanced reactor to reach this stage, following a separate design that hit criticality earlier under the same program.
The site itself is the Utah San Rafael Energy Research Center, known as USREL, a state-run facility that hosts advanced energy testing. Utah officials frame the lab as a proving ground for technologies that can meet rising power loads across the western United States. By running the Ward 250 at USREL, Valar gains access to a controlled environment with existing regulatory relationships and monitoring infrastructure, reducing the timeline and cost of early-stage testing compared to building a standalone site from scratch.
DOE records, NEPA clearance, and what the filings actually show
Two federal records anchor the public evidence for the Ward 250 project. The first is the DOE’s announcement confirming the zero-power fueled criticality demonstration. Zero-power criticality means the reactor sustained a controlled nuclear chain reaction at negligible thermal output, proving that the core geometry, fuel loading, and neutron physics work as designed. It does not mean the reactor generated usable electricity or operated at its target thermal capacity. That distinction matters because the gap between zero-power criticality and commercial power production involves years of additional testing, licensing, and engineering validation.
The second record is a DOE NEPA filing, designated CX-271015, which states that Valar proposes to construct and operate the Ward 250 research reactor at USREL. The filing received a categorical exclusion, meaning the DOE determined the project did not require a full environmental impact statement or environmental assessment. Categorical exclusions are common for research-scale reactors that fall below certain size and risk thresholds, but the designation tells readers nothing about the reactor’s eventual commercial performance, water consumption profile, or electrical output capacity.
A tension exists between these two records. The criticality announcement treats the demonstration as a completed event, while the NEPA filing describes the reactor as a proposal for construction and operation. One plausible reading is that the criticality test occurred under temporary or limited authorization, and the NEPA categorical exclusion covers a broader or longer-term phase of work at the same site. Another is that the filings reflect different stages in an evolving project description, with documentation lagging behind on-the-ground activities. Neither document resolves this ambiguity, and DOE has not published a timeline reconciling the two.
Nvidia’s name has appeared in reporting about the project, with Bloomberg coverage referencing the chipmaker’s interest in pairing advanced reactors with its data-center operations. No primary DOE or Utah state record in the public filing trail confirms a formal partnership, offtake agreement, or co-development arrangement between Nvidia and Valar Atomics. The connection, while commercially logical given Nvidia’s massive power needs for GPU clusters, lacks documentary confirmation beyond news coverage.
Unanswered questions about water, output, and commercial viability
The headline promise of a “nearly waterless” reactor powering AI workloads rests on engineering claims that the public record has not yet validated. No DOE filing, NEPA document, or Utah state publication in the available evidence provides specific water consumption figures for the Ward 250. The reactor’s cooling architecture, whether it uses air, gas, molten salt, or another medium instead of water, is not detailed in the categorical exclusion or the criticality announcement. Without those specifications, any comparison to the water intensity of a natural gas plant is speculative rather than sourced.
Electrical output presents a similar gap. The “250” in the reactor’s name suggests a target capacity, but the DOE documents do not confirm whether that figure refers to kilowatts, megawatts thermal, or megawatts electric. Zero-power criticality by definition involves negligible power production, so the successful test does not validate any specific output rating. Readers should therefore treat any precise capacity numbers in marketing materials or secondary reporting as unverified unless and until they appear in formal licensing or technical documents.
Commercial viability depends on more than physics. For data-center customers, the questions include how quickly a Ward 250 unit could be permitted at or near a campus, what security perimeter it would require, how refueling and waste handling would be managed, and whether the reactor could follow load or would run at constant output. None of these operational details appear in the DOE announcement or the NEPA categorical exclusion. The filings instead focus on the limited research scope of the Utah installation and on compliance with existing environmental review thresholds.
Cost is equally opaque. The Reactor Pilot Program is designed to lower early-stage barriers for advanced designs by providing access to federal facilities and expertise, but it does not guarantee that a particular reactor will be economical without subsidies or long-term power contracts. For AI operators accustomed to flexible cloud capacity and short hardware refresh cycles, committing to a multi-decade nuclear asset represents a very different kind of infrastructure bet.
What the milestone does – and does not – mean for AI power
Within the constraints of the public record, the Ward 250’s zero-power criticality is best understood as a proof-of-concept milestone rather than a commercial launch. It shows that Valar’s core design can sustain a controlled chain reaction under test conditions at USREL and that DOE was willing to sponsor the work under its pilot framework. It does not establish that the reactor can yet deliver “nearly waterless” electricity at the scale, cost, and reliability levels AI data centers require.
For the broader nuclear sector, the test underscores how AI-driven demand is reshaping the conversation about advanced reactors. Instead of focusing solely on decarbonizing the grid, developers are increasingly pitching compact units as dedicated power plants for single industrial loads, from chip fabs to hyperscale campuses. The Ward 250 fits that narrative, even if the details of its eventual deployment remain uncertain.
For policymakers and local communities, the case highlights the importance of transparent documentation. As more advanced reactors seek categorical exclusions and pilot designations, the gap between high-level announcements and granular technical data will matter. Investors and neighbors alike will want clarity on cooling systems, water use, siting footprints, and emergency planning zones before treating nuclear-backed data centers as a solved problem.
Valar Atomics, Nvidia, and DOE have all signaled, in different ways, that AI-era power demand may justify revisiting long-standing assumptions about where and how nuclear reactors can operate. The Ward 250’s first criticality at USREL marks an early, tightly bounded experiment within that larger shift. Until further filings and test results surface, though, the reactor remains more a promising prototype than a proven answer to the data center energy and water crunch.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
