As artificial intelligence workloads surge, the infrastructure supporting them faces a mounting energy crisis. Data centers, the backbone of modern AI, are projected to consume between 9 and 17 percent of total U.S. electricity by the end of the decade, with roughly a third of that power dedicated solely to cooling the very chips that run the models. MIT spinout Ferveret has emerged with a novel solution that promises to dramatically shift this calculation, demonstrating a 35 percent boost in AI efficiency by rethinking thermal management from the ground up.
What Is Ferveret’s Adaptive Phase Cooling System?
Ferveret’s technology, called Adaptive Phase Cooling (APC), adapts principles originally developed for nuclear reactor heat transfer to the data center floor. Instead of relying on energy-hungry fans or wasteful water systems, APC submerges servers in a specialized liquid engineered to absorb heat far more efficiently than air. The key innovation lies in the bubble dynamics at the chip’s surface: the system produces significantly smaller bubbles that detach more frequently and recondense quickly in the surrounding liquid. This accelerated bubble-rewetting cycle dramatically improves heat transfer, allowing chips to run at higher performance without thermal throttling.
The company was founded by Reza Azizian, a former MIT postdoc in nuclear engineering and an alumnus of Nvidia, and Matteo Bucci, MIT’s Esther and Harold E. Edgerton Associate Professor in Nuclear Science and Engineering. Their combined expertise in heat transfer physics has produced a system that not only cools more effectively but does so with zero water consumption and no toxic PFAS “forever chemicals” used in some competing approaches.
How Does Ferveret’s Cooling Boost AI Performance by 35%?
In a recent study conducted with the Samueli Computer Science Department at UCLA, Ferveret found that its APC solution alone delivered a 15 percent improvement in computational power efficiency compared to the best existing liquid cooling methods. The full 35 percent gain emerges when APC is paired with Ferveret’s proprietary power control software, which monitors sensors inside each cooling box and dynamically adjusts server power levels to optimize operating conditions in real time. By minimizing the energy wasted on cooling and power delivery inefficiencies, data centers can extract significantly more useful work—measured in tokens, the small units of text or data AI models process—from every watt consumed.
“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,” Azizian stated. “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.”
Why Liquid Cooling, and Why the Bubble Size Matters
Liquid is inherently better at transferring heat than air—a fact anyone who has plunged a hand into room-temperature water intuitively understands. When that liquid reaches its boiling point, the phase change from liquid to vapor absorbs a tremendous amount of thermal energy, making it the most efficient thermodynamic method for removing heat from a chip. However, boiling introduces complexity: operators must capture and reliquefy the vapor while controlling pressure, temperature, and fluid inventory.
Ferveret’s solution leverages a phenomenon called subcooled boiling, a process adapted from nuclear reactor engineering. The company’s proprietary liquid has a low boiling point, and the physical conditions at the chip surface cause the formation of very small vapor bubbles. These tiny bubbles detach rapidly and recondense in the cooler liquid above, accelerating the heat transfer cycle without the need for complex vapor management systems. The result is a compact, modular system—each server fits into its own cooling box—that sidesteps the large immersion tanks common in other approaches.
Real-World Testing and Industry Adoption
Ferveret is not waiting for theoretical validation. The company is already testing its APC systems with CleanSpark, a data center developer and operator, as well as FuriosaAI, an AI accelerator company, and Switch, one of the largest data center operators in the United States. The company is also part of Nvidia’s Inception program for startups and is in active discussions with major cloud computing providers, indicating strong commercial traction.
The modular design is a deliberate choice. “Most immersion cooling solutions are large tanks that people submerge the servers in,” Azizian explained. “We have a smaller, modular rack-mounted solution that makes it adaptable to the current infrastructure, so it’s easier for people to deploy our technology.” This approach lowers the barrier to entry for data centers that want to transition from air cooling without a complete infrastructure overhaul.
Beyond Efficiency: Sustainability and Geographic Flexibility
The zero-water-consumption aspect of Ferveret’s system carries significant implications for where data centers can be built. Many regions with abundant solar or wind energy—ideal for powering AI compute—lack the water resources needed for traditional evaporative cooling. Bucci highlighted this advantage: “The sun shines in places where you don’t have much water, so the advantage of us being water-free is we allow you to build data centers where you have solar energy but nothing to cool the data center down.” This opens the door for data center deployment in arid regions across Africa, the Middle East, and parts of the American Southwest, unlocking renewable energy sources that were previously impractical for high-density compute.
What This Means for AI Infrastructure
Ferveret’s breakthrough arrives at a critical juncture. The explosive growth of AI models demands ever more compute capacity, but the industry is increasingly constrained by power availability, water scarcity, and environmental regulations. Solutions that can squeeze more usable compute—more tokens per watt—from existing infrastructure offer an immediate, practical path forward without waiting for breakthroughs in chip architecture or grid expansion.
The 35 percent efficiency gain reported by Ferveret is not a theoretical ceiling but a measured result in collaboration with an academic partner. When combined with the company’s broader roadmap for scaling and partnerships, it signals a meaningful shift in how data centers can approach the thermal challenge. For professionals evaluating cooling options for new or existing facilities, the key criteria to assess now include bubble dynamics, system modularity, power control integration, and water consumption—capabilities that Ferveret’s APC system directly addresses.
For developers and operators watching this space, the immediate takeaway is clear: the next wave of AI efficiency gains may come not from the chip itself, but from how we keep it from melting. Ferveret’s technology provides a compelling, testable option for any organization looking to extract more performance from their AI hardware while meeting sustainability targets. The company plans to announce expanded partnerships later this year, and those announcements will be worth monitoring closely as the industry races to power the future of artificial intelligence.
