Nuclear-inspired cooling could boost AI data center efficiency by 35%

Startup Ferveret adapts subcooled boiling technique from reactors to cool chips with less power and zero water use.

By AI·Reporter·June 10, 2026·~5 min read

Takeaways

›Ferveret's cooling system uses nuclear-inspired 'subcooled boiling' for 35% more AI efficiency
›Water-free approach enables data centers in solar-rich, water-poor regions
›Early tests show promise, but large-scale deployment remains to be proven
›Technology could be key to sustainable AI growth if adopted by major cloud providers

The AI boom is driving massive data center expansion, with cooling alone consuming a third of their electricity. Ferveret, an MIT spinoff, aims to slash that energy use by borrowing a trick from nuclear reactors.

A cooling system that makes bubbles work harder

Ferveret's Adaptive Phase Cooling (APC) submerges servers in a specialized liquid, but with a crucial twist: it generates smaller bubbles that detach more frequently from chip surfaces. This accelerates heat transfer compared to conventional liquid cooling methods.

Key features of the APC system:

Uses a low boiling point liquid free of toxic PFAS 'forever chemicals'
Produces smaller bubbles that detach and recondense quickly
Delivered in modular, rack-mountable boxes for easy deployment
Includes control software to optimize power in real-time

In tests with UCLA, the system showed a 15% improvement in computational power efficiency over state-of-the-art liquid cooling. Ferveret claims that when combined with their power control system, data centers can achieve 35% more AI model tokens (small pieces of text or data) for the same power input.

From nuclear engineering to AI cooling

Founders Reza Azizian and Matteo Bucci adapted their expertise in nuclear reactor heat transfer to tackle data center inefficiency. The APC system is based on 'subcooled boiling,' a process used in reactors.

Azizian's experience at Microsoft and Nvidia highlighted the outdated nature of data center cooling:

"I thought, 'Holy crap, this is not how you cool facilities,'" Azizian recalls, noting air cooling can still take up 40 percent of the power going into a data center. "It was not an efficient way of doing things, but since it wasn't hurting the performance, no one cared that the cooling technology was 50 years old."

Beyond efficiency: Enabling sustainable data center locations

Ferveret's water-free approach could allow data centers to be built in regions with abundant solar energy but limited water resources, such as parts of Africa, the Middle East, and America. This expands options for sustainable AI infrastructure.

Real-world traction and next steps

Ferveret is already testing with notable players:

CleanSpark (data center developer/operator)
FuriosaAI (AI accelerator company)
Switch (major U.S. data center operator)

The startup is in talks with major cloud providers and plans to announce expanded partnerships later this year. As part of Nvidia's Inception program for startups, Ferveret is well-positioned to scale its technology rapidly.

The catch: Early days and unproven at scale

While promising, Ferveret's technology is still in early testing phases. The 35% efficiency gain claim needs validation in large-scale, real-world deployments. Additionally, the costs of retrofitting existing data centers or building new ones with this system are not yet clear.

The true test will be whether major cloud providers adopt the technology at scale, proving its reliability and cost-effectiveness in 24/7 production environments.

Why it matters: Sustainable AI growth

As AI development accelerates, data center energy consumption is becoming a critical bottleneck and environmental concern. Ferveret's approach, if successful at scale, could help the AI industry expand while minimizing its growing energy and water footprint. This aligns with both economic and sustainability goals for tech companies and regions looking to attract data center investments.