Saturday, July 25, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeData CentersReport
Data Centers · Report

Valar Atomics partnered with NVIDIA to develop a first-of-a-kind SMR-powered (small modular reactor) AI data center in Utah.

Nuclear frontiers: NVIDIA's direct datacenter buildout validates grid-independent power model; SMRs now viable for compute clustering.
Trade pressSlicast · July 7, 2026 · US · Source: Google News
importance 74

Valar Atomics, the nuclear startup founded in 2023, has partnered with Nvidia to develop a pilot data center in Emery County, Utah. The facility will be powered by Valar's Ward 250 helium-cooled microreactor and is designed to generate approximately 30 megawatts of output—sufficient to run significant AI computing workloads independently of the local grid. The partnership pairs Valar's nuclear technology with Nvidia's latest Blackwell AI chip architecture. During a demonstration event in July 2026, Nvidia's Blackwell chips were successfully powered by the reactor.

Valar achieved a critical milestone on March 31, 2026, when the Ward 250 microreactor reached zero-power fueled criticality—sustained a controlled nuclear chain reaction for the first time. It became the first U.S. Department of Energy-authorized microreactor to accomplish this milestone outside a national laboratory.

The Utah facility operates as a behind-the-meter power solution, meaning the data center generates its own electricity on-site rather than relying on transmission lines and utility companies. The system employs closed-loop cooling, resulting in near-zero water usage—a material advantage in a water-scarce state.

Valar raised $450 million at a $2 billion valuation in 2026. The company's vision extends beyond this single project to a "gigasite model"—grid-independent data center campuses that can be deployed wherever compute is needed, regardless of existing power infrastructure.

The AI industry faces an acute electricity constraint. Training large language models and operating GPU clusters at scale requires power densities that traditional data centers were not built to handle. Grid interconnection queues have ballooned, and in many regions, new data center projects face multi-year delays simply to connect to existing infrastructure. This bottleneck has made energy sourcing the primary constraint on AI infrastructure expansion.

Nuclear power addresses this constraint effectively. Unlike solar and wind, nuclear plants generate baseload power continuously, regardless of weather. Unlike natural gas, they produce zero carbon emissions during operation. And unlike large conventional nuclear plants, which require a decade or more to build, microreactors like the Ward 250 are designed for factory manufacturing and rapid deployment.

If Valar's model scales successfully, it could reshape the economics of proof-of-work mining by eliminating two major cost variables: volatile electricity prices and regulatory risk from local utilities.

The market is pricing nuclear's role in AI infrastructure accordingly. Valar's $2 billion valuation—achieved just three years after founding—reflects aggressive investor conviction. Competitors including Oklo, NuScale, and Kairos Power are pursuing their own small modular reactor designs, but Valar's first-mover advantage with DOE authorization outside a national laboratory provides a meaningful head start.

Regulatory uncertainty remains the primary risk. While the DOE has authorized this pilot, scaling to a network of gigasites will require navigating NRC licensing, state-level permitting, and public sentiment around nuclear energy. These hurdles, though surmountable, could delay timelines and increase deployment costs.

Read the original
Valar Atomics partnered with NVIDIA to develop… · Slicast