The heat is on for artificial intelligence. As high-density processors get hotter than ever before to meet rising demands, AI data center cooling risks becoming a bottleneck for innovation. 

It has become clear that traditional cooling systems are no match for the new generation of AI chips, leading the industry to scramble to come up with alternatives, including attaching liquid cold plates directly to graphics processing units, according to Jonathan Ballon (pictured, left), chief executive officer of Iceotope Technologies Ltd. While this reduces heat at the chip level, this alone does not solve the underlying problems from reaching their boiling point. 

“The industry over the last couple of years has been adopting these cold plates to focus on the hottest part of the server, which of course is the GPU,” Ballon told theCUBE. “The challenge is that the rest of the [information technology] in the data center — the power supply, the network, the storage, all of these other things — are also increasing in heat proportionately with that compute, and the air that cools the rest of that is struggling to keep up.”

Ballon and Ace Stryker (right), director of AI and ecosystem marketing at Solidigm LLC, a trademark of SK Hynix NAND Product Solutions Corp., spoke with theCUBE’s John Furrier and Jackie McGuire at the SC25 event, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed AI data center cooling in light of soaring infrastructure demands. (* Disclosure below.)

AI data center cooling at the boiling point

​​Iceotope positions itself as a precision liquid-cooling specialist for exactly this moment. The company has spent the past decade working with hyperscalers to develop systems designed around a future with no forced air inside the rack and minimal reliance on additional water, according to Ballon. 

“In the solution that we have, we don’t need any incremental water,” he explained. “Because we’re not cooling the air, we reduce the amount of power necessary by over 80%.”

The growing demand for better cooling means that data centers built in the 2027-28 timeframe will not have air built into the architecture, Ballon noted. Everything in the IT stack will need to be liquid cooled, including the storage. To make that vision real, Iceotope has been working closely with partners to ensure even the last holdout in the rack can make the leap to liquid.

“We’ve been working with Solidigm to build high-density, 100% liquid-cooled storage solutions,” Ballon said. “The partnership that we’ve built is perfect for this, and Solidigm is ahead of the pack in terms of anticipating this moment so that their portfolio will be ready for that future”

“It’s been good enough up to this point for PCIe Gen 4 and Gen 5 devices, where you’re pulling maybe 20 or 25 watts per SSD,” he said. “But as you look ahead to PCIe Gen 6, which will be adopted in big numbers going into 2026, and PCIe Gen 7 — beyond that, you might be talking about a 40-, 50- or even 60-watt power budget per SSD. It’s a very different proposition, and all of a sudden air cooling is no longer sufficient.”

Here’s the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of the SC25 event:

(* Disclosure: Solidigm sponsored this segment of theCUBE. Neither Solidigm nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE