As artificial intelligence continues its explosive growth, the infrastructure powering it is consuming an alarming amount of energy. Data centers are projected to account for 9-17% of total U.S. electricity usage by 2030, with roughly one-third of that power devoted solely to cooling the chips that run AI models. But what if we could make this cooling process dramatically more efficient?
From Nuclear Reactors to AI Infrastructure
Enter Ferveret, a startup founded by former MIT postdoc Reza Azizian and MIT Professor Matteo Bucci, who are applying decades of nuclear reactor heat transfer research to solve AI's cooling crisis. Their innovative approach doesn't just promise incremental improvements—it could revolutionize how we think about data center sustainability.
The journey began when Azizian first walked into a data center in 2017 and was struck by the massive, energy-hungry fans filling the building. "Holy crap, this is not how you cool facilities," he recalls, noting that traditional air cooling can consume up to 40% of a data center's power. "It was not an efficient way of doing things, but since it wasn't hurting performance, no one cared that the cooling technology was 50 years old."
The Science Behind the Solution
Ferveret's breakthrough lies in their Adaptive Phase Cooling (APC) system, which adapts a nuclear reactor process called subcooled boiling. Here's how it works:
- Liquid immersion: Computer servers are submerged in a specialized liquid that absorbs heat far more efficiently than air
- Micro-bubbles: The system produces much smaller bubbles at the server surface that detach more frequently
- Accelerated heat transfer: This rapid bubble cycle dramatically speeds up heat removal from the chips
- Zero water consumption: The system uses no water and avoids toxic PFAS chemicals found in other solutions
"Liquid is a better heat transfer medium than air," explains Professor Bucci. "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."
Impressive Performance Results
The numbers speak for themselves. In collaboration with UCLA, Ferveret's testing revealed:
- 15% improvement in computational power efficiency compared to state-of-the-art liquid cooling
- 35% more AI tokens generated with the same power consumption when combined with their optimization software
- Significantly reduced infrastructure complexity with modular, rack-mounted units
Major players are taking notice. Ferveret is already testing with companies including CleanSpark, FuriosaAI, and Switch—one of the largest U.S. data center operators. They're also part of Nvidia's Inception program for startups.
Enabling AI in New Locations
Perhaps most exciting is how this technology could democratize AI infrastructure deployment. By eliminating water requirements, Ferveret's system enables data centers in regions with abundant renewable energy but limited water resources.
"The sun shines in places where you don't have much water," notes Bucci. "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 technology can help deploy data centers in regions where normally you wouldn't have the resources to do so, including Africa, the Middle East, and parts of America."
What This Means for AI Development
For AI practitioners and prompt engineers, this technology could have significant implications:
- More accessible AI: Lower operational costs could make AI services more affordable
- Expanded capacity: More efficient cooling allows for higher-performance chips and better model performance
- Sustainability: Reduced environmental impact makes AI development more sustainable
- Geographic flexibility: AI services could be deployed closer to users in previously unsuitable locations
The Road Ahead
As Ferveret scales their technology, they're addressing one of AI's most pressing challenges: how to continue growing computational capacity without straining our planet's resources. With talks underway with major cloud computing companies and expanded partnerships planned for later this year, this nuclear-inspired solution could play a crucial role in AI's sustainable future.
"The computing industry is facing a huge challenge in the form of access to power, and they have a problem with access to water in many regions," says Azizian. "The main goal for these data center operators would be to get more tokens from the power they have. We've shown we can do that."
Source: MIT News by Zach Winn