CAMBRIDGE, MASSACHUSETTS — Ferveret, a startup, is testing its liquid cooling technology with data center operators CleanSpark and Switch, and with artificial intelligence accelerator developer FuriosaAI. The company's cooling solution is adapted from a nuclear reactor process called subcooled boiling and aims to improve energy efficiency in artificial intelligence computing.
Ferveret was founded in 2021 by Reza Azizian and Matteo Bucci. Azizian is a former postdoctoral researcher in nuclear engineering at the Massachusetts Institute of Technology (MIT), and Bucci is the Esther and Harold E. Edgerton Associate Professor in the Department of Nuclear Science and Engineering at MIT. The two began collaborating on nuclear reactor heat transfer research in 2013. Bucci became an assistant professor at MIT in 2016. Azizian previously worked on the Microsoft HoloLens and at Nvidia, which produces graphical processing units used in AI models.
Ferveret's cooling system submerges computer servers in a specialized liquid. This adaptive phase cooling system produces smaller surface bubbles than conventional liquid cooling systems. These bubbles detach frequently to accelerate heat transfer and quickly recondense in the surrounding liquid. The cooling liquid has a low boiling point and does not contain perfluoroalkyl or polyfluoroalkyl substances.
A study conducted by Ferveret with the Samueli Computer Science Department at the University of California, Los Angeles, reported that the cooling solution improved computational power efficiency by 15 percent compared to existing liquid cooling technology. Ferveret states that its power control system optimizes operating conditions, allowing data centers to generate 35 percent more artificial intelligence tokens using the same amount of power. Tokens consist of small pieces of text or data. Around one-third of data center electricity is used to cool chips running AI models, and air cooling can account for up to 40 percent of total data center power consumption.
Azizian, co-founder of Ferveret, said, "Our goal is to make data centers as sustainable as possible and help them use every single watt of power to generate tokens, which are the most useful outputs." He added, "Our system enables the operation of more powerful chips, it helps data centers waste a lot less energy, and it accomplishes all that with zero water consumption."
No independent assessment was available for this report.