Ferveret study says waterless cooling lifts AI efficiency

Ferveret said a benchmark study with UCLA's Computer Science Department found its waterless cooling system improved compute efficiency by 15% over direct-to-chip liquid cooling, using Nvidia H200 graphics processors as the reference platform.

The San Jose-based company said the study recorded zero water consumption and a facility-level power usage effectiveness score of 1.03 for its Adaptive Phase Cooling Solution. The benchmark measured server-level computational efficiency in TFLOPs per kilowatt of server power, a metric focused on graphics processor performance rather than facility efficiency.

Cooling has become a central issue for data centre operators as artificial intelligence workloads increase heat output and electricity demand. Ferveret is pitching its system as an alternative to conventional water-based cooling as operators face tighter limits on power and water use.

Its approach is based on subcooled boiling, a technique drawn from nuclear reactor systems. In this process, small bubbles form and detach repeatedly before recondensing in the surrounding liquid, refreshing fluid at the chip surface and improving heat transfer.

According to Ferveret, those thermal conditions allow chips to run at lower temperatures and sustain higher power levels. That, in turn, could reduce the time needed to train machine learning models by allowing hardware to operate at higher clock speeds for longer.

Chief Executive Officer Reza Azizian linked the findings to broader strains on data centre infrastructure.

"AI and neo-cloud are redefining the scale, density and intensity of modern computing, and legacy cooling simply cannot keep pace. It is critical that we improve the environmental footprint of data centers as this new era unfolds," said Reza Azizian, Chief Executive Officer, Ferveret. "Our Adaptive Phase Cooling solution eliminates the need for water while dramatically improving computational efficiency, allowing customers to extract more compute from the same power envelope. It delivers the scalability, sustainability, and flexibility that today's AI infrastructure demands."

Study findings

The work was carried out with UCLA's Intelligent Connectivity Labouratory, or ICON Lab. Omid Abari, an associate professor in UCLA's Computer Science Department, said the testing indicated gains beyond lower energy use.

"Ferveret isn't just a cooling solution-it's a performance multiplier that redefines expectations for modern computing infrastructure," said Omid Abari, Associate Professor, UCLA Computer Science Department. "Our recent study shows that Ferveret cooling reduces the time required to train machine learning algorithms by enabling hardware to operate at higher sustained clock speeds. In other words, Ferveret not only provides a more efficient thermal solution but also delivers better performance, resulting in shorter training times."

The claims come as data centre operators and investors look for ways to support more intensive computing without placing proportionate demands on local utilities. Ferveret said eliminating water dependency could open up sites previously considered unattractive because of limited water access or tighter environmental constraints.

The company cited US Department of Energy figures showing that data centres in the United States consume up to 4.5% of total electricity production, with that share forecast to rise to 12% by 2028. It also pointed to estimates that US data centres could consume more than 700 billion gallons of water a year under current cooling approaches.

Investor backing

Ferveret said it has attracted backing from investors including TO VC, Aramco Venture, Cerberus, Y Combinator, Baruch Future Ventures, Verso Capital, Acclimate Ventures, Cathexis Ventures, Valkyrie, E14 and Climate Capital. It did not disclose the size of the investment.

Charles Goodwin, a partner at TO VC, said the firm viewed Ferveret's background in nuclear thermal management as a distinguishing factor in the crowded data centre infrastructure market.

"The data center industry urgently needs breakthrough technologies that can meet the accelerating performance demands of AI infrastructure. Co-founders Reza Azizian and Matteo Bucci spent years solving thermal management challenges in the nuclear power industry and are now applying those techniques to address the fundamental limitations of data center cooling. Their vision to deliver sustainable, high-density cooling that unlocks more compute per watt, without consuming a single drop of water, is what truly differentiates Ferveret in a market where every megawatt matters."

Founded in 2021 by MIT alumni, Ferveret develops cooling systems for AI data centres from its base in San Jose, California. The company said its design operates near ambient pressure and is intended to reduce infrastructure complexity compared with conventional liquid cooling systems.