A startup founded by MIT researchers is adapting cooling technology from nuclear reactors to make AI data centers dramatically more efficient, cutting energy waste by 35% while using zero water. The innovation comes as artificial intelligence's explosive growth is straining power grids worldwide, forcing engineers to rethink how they manage the intense heat generated by the chips that train and run AI models.

Why Is Data Center Cooling Such a Big Problem?

Data centers are projected to account for 9 to 17 percent of total electricity usage in the United States by the end of the decade. Today, roughly a third of all electricity flowing into data centers goes toward cooling, not computing. That's a staggering waste of energy at a time when AI companies are racing to build more powerful systems. Traditional air cooling, which relies on massive fans, can consume up to 40 percent of a data center's total power draw.

The problem intensifies because modern AI chips pack more and more computing components into smaller spaces, generating extreme heat. Data center operators have increasingly turned to liquid cooling as a solution, but most existing systems are inefficient, complex, and often rely on toxic chemicals.

How Does Nuclear-Inspired Cooling Work?

Ferveret, founded by Reza Azizian, a former MIT postdoctoral researcher in nuclear engineering, and Matteo Bucci, MIT's Esther and Harold E. Edgerton Associate Professor in the Department of Nuclear Science and Engineering, adapted a process used in nuclear reactors called subcooled boiling. The system submerges computer servers in a specialized liquid that absorbs heat far more efficiently than air.

What sets Ferveret's approach apart is the physics of bubble behavior. The company's Adaptive Phase Cooling (APC) solution produces much smaller bubbles at the server surface, which detach more frequently and quickly recondense in the surrounding liquid. This accelerates the heat transfer cycle, moving thermal energy away from chips much faster than conventional liquid cooling systems.

"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," said Reza Azizian, co-founder of Ferveret.

Reza Azizian, Co-founder, Ferveret

The system uses a liquid with a low boiling point and contains no toxic PFAS (per- and polyfluoroalkyl substances) chemicals, often called "forever chemicals," that other immersion cooling approaches rely on. Ferveret delivers its APC system in small, modular boxes that house individual servers, making deployment easier and maintenance simpler than large tank-based systems.

What Are the Real-World Performance Gains?

In a recent study conducted in collaboration with the Samueli Computer Science Department at the University of California at Los Angeles, Ferveret's APC solution delivered a 15 percent improvement in computational power efficiency compared to state-of-the-art liquid cooling solutions. When combined with Ferveret's power control software that optimizes operating conditions in real time, the company reports that data centers can extract 35 percent more tokens (small pieces of text or data used in AI models) from their AI systems using the same amount of electricity.

The startup is already testing its solutions with major industry players. 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, are all evaluating Ferveret's technology.

How to Evaluate Advanced Cooling Solutions for Data Centers

• Energy Efficiency Metrics: Look for systems that improve computational power efficiency by at least 15 percent compared to existing liquid cooling approaches, and verify claims through independent testing or peer-reviewed studies.
• Water Consumption: Prioritize cooling solutions that eliminate water use entirely, since water scarcity is becoming a critical constraint for data center expansion in many regions.
• Chemical Safety: Ensure cooling liquids do not contain PFAS or other persistent toxic chemicals that accumulate in the environment and pose long-term health risks.
• Deployment Flexibility: Choose modular, rack-mounted systems that integrate with existing infrastructure rather than requiring complete facility redesigns or large tank installations.
• Real-Time Optimization: Select solutions that include software controls to adjust power distribution dynamically, maximizing computational output per watt consumed.

Where Did This Technology Come From?

Azizian's journey to founding Ferveret began in 2013 when he was a postdoctoral researcher at MIT working on heat transfer in nuclear reactors alongside Bucci, then a research scientist. The two collaborated on optimizing how nuclear facilities move thermal energy, a field where scientists have spent decades refining techniques because heat transfer directly determines how much energy can be extracted from a reactor core.

Azizian later moved into industry, working on cooling systems for Microsoft's HoloLens augmented reality headset and then joining Nvidia, the company that produces the graphics processing units (GPUs) essential for training modern AI models. In 2017, when Azizian walked into his first data center, he was struck by the inefficiency of the massive, noisy fans filling the building. "I thought, 'Holy crap, this is not how you cool facilities,'" he recalled, noting that air cooling still consumes up to 40 percent of power in many data centers despite being 50 years old technology.

Azizian reconnected with Bucci and proposed applying their nuclear reactor expertise to data center cooling. The two founded Ferveret in 2021.

Why Does This Matter Beyond Data Centers?

The broader context is urgent. As AI integration accelerates across industries, data center power demand is growing faster than electricity supply in many regions. Energy companies are pursuing hybrid approaches to meet this demand, combining nuclear power with natural gas to create flexible, large-scale power plants dedicated to AI infrastructure. However, these hybrid facilities still rely partly on fossil fuels, creating tension between meeting immediate energy needs and achieving long-term decarbonization goals.

Ferveret's technology addresses a critical piece of this puzzle: if data centers can extract significantly more computational value from each watt of electricity, the total power demand grows more slowly, reducing pressure on grids and the need for new fossil fuel capacity. The 35 percent efficiency gain translates directly to fewer new power plants needed to support AI expansion.

"Heat transfer determines how much energy you can extract from the reactor core, which translates directly to revenue," explained Matteo Bucci, co-founder of Ferveret and MIT professor.

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

The technology also eliminates water consumption, addressing another sustainability concern. Data centers consume enormous quantities of water for cooling, straining local water supplies in drought-prone regions. Ferveret's zero-water approach removes this constraint entirely, allowing data centers to expand in water-scarce areas.

As AI continues reshaping global energy markets, innovations like Ferveret's nuclear-inspired cooling represent a practical path forward: not replacing existing infrastructure overnight, but making it dramatically more efficient so that the energy we already generate goes further. For a technology sector racing to balance explosive growth with environmental responsibility, that efficiency gain may prove as valuable as any new power plant.