Ferveret, a startup founded by two MIT nuclear engineering researchers, is using a different approach to cool AI servers: machines are submerged in a low‑boiling liquid whose thermal management techniques draw on reactor physics. A central claim of the system is that it requires no water — the cooling is completely water‑free.
Studies and tests
A University of California, Los Angeles (UCLA) study found that Ferveret’s solution achieved about 15 percent better energy efficiency compared with state‑of‑the‑art liquid cooling. The same study reported that Ferveret’s control software delivered up to 35 percent more tokens per watt, meaning more model work per unit of energy.
Ferveret is currently running tests with data center operators, including Switch, one of the largest data center operators in the United States.
Implications for siting data centers
Historically, data center locations have been influenced by water availability because server cooling typically requires substantial water. Removing the need for water as a cooling medium eliminates that geographic constraint.
Practically, this means that sun‑rich but water‑poor regions — for example parts of the Middle East, Africa, and the American Southwest — become more viable sites for high‑performance AI data centers. As a result, the principal constraint on where to build shifts from water access to the availability and price of electrical power.
Conclusion
While much of the industry competition focuses on watts and energy efficiency, the less conspicuous strategic prize is land and site flexibility: whoever decouples cooling from water can build where energy is cheapest and where competitors are unlikely to follow. Ferveret’s water‑free, reactor‑physics‑inspired approach therefore promises both efficiency gains and a potential reshaping of the geographic distribution of AI infrastructure.
