The rise of artificial intelligence is driving a massive expansion of data centers, projected to account for 9% to 17% of total electricity usage in the U.S. by the end of the decade. Currently, about one-third of the electricity used by data centers is dedicated to cooling the chips that run AI models. Startup Ferveret aims to enhance this efficiency. Founded by Reza Azizian, a former MIT postdoc in nuclear engineering, and Matteo Bucci, an associate professor at MIT, the company adapts cooling methods from nuclear reactors, using no water and significantly less electricity. Their cooling system submerges servers in a specialized liquid that absorbs heat far more effectively than air.
What sets Ferveret's Adaptive Phase Cooling (APC) apart from other liquid cooling systems is the production of smaller bubbles at the server's surface, which detach more frequently, accelerating heat transfer. Ferveret is already testing its solutions with companies like CleanSpark, FuriosaAI, and Switch. A recent study in collaboration with UCLA found that Ferveret's APC solution improved computational power efficiency by 15% compared to state-of-the-art liquid cooling solutions. By combining these savings with their power control system, the company claims data centers can achieve 35% more tokens from their AI models with the same power.
“Our goal is to make data centers as sustainable as possible and help them use every single watt of power to generate tokens,” Azizian states.
Azizian and Bucci first met at MIT in 2013 while researching heat transfer in nuclear reactors. After Azizian transitioned to industry, focusing on chip cooling, he was struck by the inefficiency of traditional air cooling in data centers. “I thought, ‘Holy crap, this is not how you cool facilities,’” he recalls.
Ferveret was founded in 2021 and has since adapted nuclear reactor heat transfer principles to data centers. They utilize immersion cooling techniques, which are more effective than air cooling, with their system producing smaller bubbles that enhance heat transfer efficiency. Their modular APC system is easier to deploy and maintain than traditional large immersion cooling tanks, and they also provide software that optimizes power distribution in real-time.
In addition to improving efficiency, Ferveret’s technology promotes sustainability by making it feasible to operate data centers in remote areas with abundant renewable energy but limited water resources.
Ferveret is currently in discussions with large cloud computing companies and is part of Nvidia's Inception program for startups. The founders aim to scale their technology to support the AI industry without straining the planet’s resources further.
Blogger's Review: Ferveret's innovative cooling technology not only enhances the energy efficiency of data centers but also provides sustainable solutions. By leveraging heat transfer principles from nuclear reactors, Ferveret's system is poised to advance AI technologies, especially in resource-scarce regions. Its modular design and real-time power adjustment capabilities show a profound understanding of modern data center needs and challenges.