Startup’s nuclear-inspired cooling system could make data centers more sustainable

2026-06-10 · Source: MIT News - Artificial intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

Ferveret, a startup founded by MIT researchers Reza Azizian and Matteo Bucci, has developed a nuclear-inspired cooling system designed to enhance data center sustainability by reducing energy and water consumption for AI chip cooling. Announced on June 10, 2026, their Adaptive Phase Cooling (APC) solution submerges computer servers in a specialized liquid, generating smaller, more frequently detaching bubbles than other liquid cooling methods, which significantly accelerates heat transfer. This approach, adapted from nuclear reactor technology, eliminates water usage and substantially lowers electricity needs. In collaboration with UCLA, Ferveret's APC demonstrated a 15 percent improvement in computational power efficiency over existing liquid cooling, and when combined with their power control system, enables 35 percent more AI tokens from the same power. The modular, rack-mounted system is currently being tested by companies like CleanSpark, FuriosaAI, and Switch, and is part of Nvidia's Inception program.

Key takeaway

For AI Architects or MLOps Engineers evaluating data center infrastructure, Ferveret's Adaptive Phase Cooling system offers a compelling solution to reduce operational costs and environmental impact. You should consider this modular, water-free technology to achieve up to 35 percent more AI tokens from your existing power budget and enable data center expansion into regions with limited water resources. This innovation directly addresses the growing power and water demands of AI workloads.

Key insights

Adapting nuclear reactor heat transfer principles can significantly improve data center cooling efficiency and sustainability.

Principles

• Smaller, frequently detaching bubbles enhance heat transfer in immersion cooling.
• Subcooled boiling optimizes heat removal with minimal temperature differences.
• Modular cooling systems simplify deployment and maintenance.

Method

Ferveret's APC submerges servers in a low-boiling-point, PFAS-free liquid, generating small, rapidly recondensing bubbles at the chip surface to accelerate heat transfer, complemented by power optimization software.

In practice

• Improve computational power efficiency by 15% over liquid cooling.
• Increase AI model token output by 35% using existing power.
• Enable data center deployment in water-scarce regions.

Topics

• Data Center Cooling
• Liquid Immersion Cooling
• AI Infrastructure
• Energy Efficiency
• Water Conservation
• Adaptive Phase Cooling

Best for: CTO, VP of Engineering/Data, Executive, AI Architect, MLOps Engineer, Director of AI/ML

Related on AIssential

• Elemental Impact Rallies Big Tech for Data Centre Innovation — Big Tech collaborates through Elemental Impact's DCII to fund and test sustainable data center innovations, driven by AI workload demands.
• How AI data centres cut annual water consumption by 50% — Liquid cooling for AI data centers can eliminate water consumption, addressing environmental concerns for high-density compute.
• World’s first wind-powered underwater datacentre starts operating in China — Submerged, wind-powered datacenters significantly cut energy and water use through natural cooling and renewables.
• It’s not just Nscale. Here are all of Europe’s neoclouds, mapped — Europe's "neoclouds" are rapidly expanding data center capacity to meet AI-driven computing demand.
• The great American data center divide — Rural communities are increasingly resisting data center expansion due to concerns over water, energy, and environmental impact.
• NVIDIA GTC Studio with Insights from Schneider Electric — AI can both consume and optimize energy, creating a circular relationship for sustainable scaling of AI infrastructure.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT News - Artificial intelligence.