Valar Atomics to partner with NVIDIA on Utah data center aimed at reducing water use for AI

Valar Atomics, a nuclear power startup, said it is partnering with NVIDIA to develop a small data center in Utah designed to demonstrate how computing facilities needed for artificial intelligence can conserve water, Reuters reported on July 1.

The project is tied to Valar’s nuclear site, Reuters said, where the company is working with a microreactor, a small-scale nuclear power system intended to produce electricity for industrial and data applications. The companies said the initiative is meant to address a practical constraint for data centers serving AI workloads: water consumption for cooling.

Reuters also reported that the partnership includes a demonstration in which Valar Atomics powered NVIDIA hardware. The reporting said Valar ran a demonstration powering NVIDIA’s Blackwell, the company’s latest AI chip architecture for data centers, highlighting how the compute supply chain could be paired with a non-water-intensive energy approach.

NVIDIA’s involvement goes beyond the engineering demonstration. Reuters cited an NVIDIA executive, John Josephakis, a global vice president, saying the company is exploring how “behind-the-meter, waterless advanced nuclear systems” could support future AI factories that require both scale and reliability.

For NVIDIA, the tie-up is consistent with its broader effort to move from selling chips to enabling complete AI systems. In this case, the focus is less on GPUs alone and more on the energy and infrastructure around them, an increasingly prominent topic as demand for AI compute expands and utilities and cooling resources become limiting factors for data center buildouts.

The initiative also reflects how nuclear developers are looking for new pathways into the power-hungry technology sector. Instead of positioning reactors only for electricity generation, Valar is framing the microreactor concept as an enabler for data centers that may struggle with cooling water availability, according to Reuters’ account of the companies’ claims.

Still, key commercial details were not disclosed in the publicly reported coverage. Reuters did not spell out the projected capacity of the Utah data center, the expected construction or demonstration timeline, the sitespecific energy delivery terms, or whether the companies plan to scale the concept beyond the pilot. It is also unclear from the report how “waterless” performance will be measured in practice, especially under sustained AI computing loads.

What to watch next is whether the partnership matures into a defined deployment plan, including clear metrics for power availability, cooling approach, and water savings, and whether NVIDIA expands such collaborations with other energy providers if the Utah concept performs as intended.