The AI boom has a dirty secret that nobody wants to talk about: it's drowning in heat.
Every time you prompt ChatGPT, generate an image with Midjourney, or run a deep research query, somewhere in a nondescript building in Virginia or Iowa or Arizona, a rack of GPUs is generating enough thermal energy to fry an egg. And keeping those chips from melting currently consumes around a third of a data center's total electricity bill — not for computation, but for cooling.
Now imagine you could take that wasted energy and turn it into a competitive advantage. That's exactly what Ferveret, an MIT spin-off founded by nuclear engineers, claims it can do.
The Bubble Revolution
Ferveret's co-founders, Matteo Bucci (MIT's Esther and Harold E. Edgerton Associate Professor in Nuclear Science and Engineering) and Reza Azizian (a former MIT postdoc), didn't come from the data center world. They came from nuclear reactors — where managing extreme heat is a matter of preventing catastrophic failure, not just saving on the electricity bill.
Their technology, called Adaptive Phase Cooling (APC), borrows a principle directly from nuclear engineering called subcooled boiling. The idea is remarkably simple: submerge computer servers in a specialized liquid, then control the boiling process at the chip surface so that bubbles form smaller and detach faster than in conventional immersion cooling.
Why does bubble size matter? Physics. Smaller bubbles mean more efficient heat transfer — the heat moves from the chip into the liquid faster, requiring less energy to circulate coolant and — critically — zero water consumption. Conventional data center cooling is incredibly water-intensive; a single large facility can consume millions of gallons annually. Ferveret's system uses no water at all.
Real Results, Not Just Theory
The technology isn't just a lab curiosity. In a recent study conducted in collaboration with UCLA's Samueli Computer Science Department, Ferveret demonstrated that its APC solution delivered a 15% improvement in computational power efficiency compared with state-of-the-art liquid cooling alternatives. Some estimates suggest data centers using APC could extract up to 35% more tokens (AI outputs) with the same power budget.
The startup is already testing with major operators. CleanSpark, the data center developer and operator, is evaluating the system. So is FuriosaAI, a South Korean AI accelerator company building next-generation chips. And Switch, one of the largest data center operators in the United States, has also signed on as a testing partner.
Why This Matters Right Now
The timing couldn't be more critical. Data centers are projected to consume between 9% and 17% of total U.S. electricity by 2030, according to the Electric Power Research Institute. That's up from roughly 4% today. The AI boom — with its GPU clusters drawing 700W to 1,000W per chip — is the primary driver.
But here's the thing that most conversations miss: power isn't the only constraint. Water is. And in drought-prone regions where data centers cluster (Northern Virginia, California, Arizona), water availability is becoming a existential business risk. Ferveret's waterless approach doesn't just save money — it unlocks locations that were previously off-limits for large-scale AI infrastructure.
Beyond Cooling
There's another layer to Ferveret's approach that's worth watching. Unlike some competing immersion cooling technologies, Ferveret's liquid contains no toxic PFAS — the so-called "forever chemicals" that don't break down in the environment and are facing increasing regulatory scrutiny worldwide. As regulators in Europe and North America tighten PFAS restrictions, this could become a decisive advantage.
The startup is currently scaling its manufacturing capacity and raising a Series A round to meet what appears to be enormous demand. If their numbers hold up at production scale — and that's always the caveat with lab results — Ferveret could fundamentally reshape the economics of AI infrastructure.
Because here's the truth: the AI revolution isn't limited by algorithms anymore. It's limited by physics. And the engineers who understand heat transfer might just be the unsung heroes of the next computing era.
Photo by Sergey Omelchenko on Unsplash
