A startup born from MIT nuclear engineering research is bringing reactor cooling technology to data centers, potentially solving one of AI's biggest infrastructure headaches: the enormous energy and water needed to keep chips cool. Ferveret's new cooling system could deliver 35% more computing output from the same amount of power, according to recent testing with major data center operators.

Today, roughly one-third of all data center electricity goes toward cooling, a figure that becomes more critical as artificial intelligence workloads intensify. Data centers are projected to account for 9 to 17 percent of total U.S. electricity usage by the end of the decade, making efficiency gains increasingly urgent.

How Does Nuclear Cooling Technology Work in Data Centers?

Ferveret's approach adapts a process called subcooled boiling, which has been refined over decades in nuclear reactors. The system submerges computer servers in a specialized liquid that absorbs heat far more efficiently than traditional air cooling from fans. What sets Ferveret apart from other liquid cooling solutions is the physics of bubble behavior: the company's Adaptive Phase Cooling (APC) system produces much smaller bubbles at the server surface that detach more frequently, accelerating the heat transfer cycle.

"Liquid is a better heat transfer medium than air. That's why when you stick your hand into room temperature water it still feels cold. When liquid is boiling, it becomes even better at removing heat because the phase change requires a lot of energy, which is the energy you remove from the chip," explained Matteo Bucci, MIT's Esther and Harold E. Edgerton Associate Professor in the Department of Nuclear Science and Engineering.

Matteo Bucci, Esther and Harold E. Edgerton Associate Professor, MIT Department of Nuclear Science and Engineering

The company uses a liquid with a low boiling point and avoids toxic PFAS "forever chemicals" that other immersion cooling approaches rely on. Unlike traditional immersion cooling systems that require large tanks, Ferveret delivers its solution in modular, rack-mounted boxes that house individual servers, making deployment easier and maintenance simpler.

What Are the Real-World Performance Gains?

In collaboration with UCLA's Samueli Computer Science Department, Ferveret found its APC solution delivered a 15% improvement in computational power efficiency compared to state-of-the-art liquid cooling systems. When combined with the company's power control software that optimizes operating conditions in real time, data centers can extract 35% more tokens (the small pieces of text or data that AI models process) from the same amount of electrical power.

Tokens are the fundamental unit of AI output, so this efficiency gain translates directly to more AI inference and training capacity without requiring additional power infrastructure. For hyperscalers and data center operators facing grid constraints, this represents a significant competitive advantage.

How to Evaluate Advanced Cooling Solutions for Data Centers

• Energy Efficiency Metrics: Look for systems that reduce cooling power consumption as a percentage of total facility power, ideally demonstrating measurable improvements over incumbent technologies through independent testing.
• Water Consumption: Prioritize solutions that minimize or eliminate water use, since water scarcity and regulatory restrictions are becoming critical constraints in many data center regions.
• Deployment Flexibility: Assess whether cooling systems can integrate with existing infrastructure through modular designs rather than requiring complete facility redesigns or large tank installations.
• Operational Complexity: Evaluate whether systems require specialized pressure, temperature, and fluid inventory controls that add maintenance burden or operational risk.
• Scalability and Reliability: Confirm that solutions have been tested with major operators and can handle the power densities required by modern AI accelerators and GPUs.

Ferveret is already testing its technology with major industry players. The company has partnerships with CleanSpark, a data center developer and operator; FuriosaAI, an AI accelerator company; and Switch, one of the largest data center operators in the United States.

The startup was founded in 2021 by Reza Azizian, a former MIT postdoc in nuclear engineering, and Matteo Bucci, an MIT professor. Azizian's journey to data center cooling began when he worked on Microsoft's HoloLens augmented reality headset and later joined Nvidia, the dominant GPU manufacturer. When he walked into his first data center in 2017, he was struck by the inefficiency of the cooling approach.

"I thought, 'Holy crap, this is not how you cool facilities.' 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," Azizian recalled.

Reza Azizian, Founder, Ferveret

Azizian and Bucci began collaborating on applying their nuclear reactor heat transfer expertise to data centers. In nuclear reactors, optimizing heat transfer directly impacts revenue by determining how much energy can be extracted from the reactor core. That same principle applies to data centers, where cooling efficiency directly affects how much computing capacity can be delivered per watt of power.

Why Does This Matter Now?

The timing is critical. Chip manufacturers have packed increasingly more components onto processors as AI demand has exploded, creating unprecedented heat density. This has forced data center operators to move beyond traditional air cooling toward liquid cooling, often through immersion techniques that submerge chips entirely. However, boiling liquid adds complexity because operators must capture and reliquefy bubbles while managing pressure, temperature, and fluid levels.

Ferveret's system simplifies this complexity while delivering superior performance. The modular design means data centers can deploy the technology incrementally rather than undertaking massive infrastructure overhauls. The software layer that optimizes power allocation in real time adds another efficiency layer that most competitors don't offer.

"Our goal is to make data centers as sustainable as possible and help them use every single watt of power to generate tokens, which are the most useful outputs. Our system enables the operation of more powerful chips, it helps data centers waste a lot less energy, and it accomplishes all that with zero water consumption," said Reza Azizian.

Reza Azizian, Founder, Ferveret

As AI infrastructure continues to expand globally, cooling efficiency will become a primary differentiator for data center competitiveness. Ferveret's approach suggests that solutions borrowed from mature industries like nuclear power can unlock significant gains in emerging infrastructure challenges. The startup's partnerships with major operators indicate the industry is ready to move beyond incremental improvements toward fundamentally different cooling architectures.