MIT-Backed Startup Brings Nuclear Cooling Tech to AI Data Centers

⚡ AI Executive Summary

Ferveret, founded by MIT researchers, has developed an advanced liquid immersion cooling system inspired by nuclear reactor heat-transfer principles to dramatically reduce data center energy consumption. The company's Adaptive Phase Cooling technology produces smaller, faster-detaching bubbles that enhance heat removal efficiency without water consumption—a critical advantage as data centers are projected to consume up to 17 percent of U.S. electricity by decade's end, with cooling alone accounting for roughly one-third of that load. The technology addresses a structural inefficiency in modern compute infrastructure: traditional air cooling wastes enormous amounts of power, while existing liquid systems remain operationally complex. Ferveret's modular, rack-mounted approach offers utilities and hyperscalers a path to higher computational output per watt while enabling geographic flexibility for renewable-rich but water-scarce regions. As AI model training and inference demands continue escalating, solutions that decouple cooling from water availability become strategically important for grid planners managing peak demand and for operators seeking compliance with carbon targets. The technology's demonstrated 15–35 percent efficiency gains suggest meaningful reductions in aggregate data center demand growth—a crucial variable for long-term load forecasting.