Startup’s nuclear-inspired cooling system could make data centers more sustainable

Source Domain: news.mit.edu

The rise of artificial intelligence is driving an enormous expansion in data centers, which are projected to account for 9-17% of U.S. electricity usage by the end of the decade.
A significant portion of data center electricity is used for cooling chips that run artificial intelligence models, with current air cooling methods being inefficient.
Ferveret, founded by Reza Azizian and Matteo Bucci, is developing an innovative cooling system to make data centers more efficient by adapting a method from nuclear reactors.
Ferveret’s Adaptive Phase Cooling (APC) uses a specialized liquid to cool chips in a way that is more energy-efficient than conventional liquid cooling and entirely avoids water use.
The company’s APC system produces smaller, more frequent bubbles that faster detach and recondense, greatly enhancing heat transfer efficiency.
Ferveret’s cooling technology enhances computational power efficiency by 15% compared to state-of-the-art solutions and allows data centers to generate 35% more output “tokens” with the same power.
The technology is also modular, allowing easier deployment and maintenance, and can help build sustainable data centers in areas with limited water resources, such as solar energy-rich regions.
Ferveret is currently working with major companies like CleanSpark and FuriosaAI, and is planning to expand partnerships to scale their technology to support the growing AI industry sustainably.

Startup’s nuclear-inspired cooling system could make data centers more sustainable | MIT News