Immersion cooling evolves in response to IT power density

While it is tempting to think of generative AI as a disruptive force, it is, in many ways, an accelerant propelling the data center industry forward — speeding up change, but along a path that began some years ago. Direct liquid cooling (DLC) is a prime example: server silicon evolution had been headed toward DLC long before generative AI entered the scene — but without the demands of AI-specific IT, the change would surely have been much slower.

Anecdotally, Nvidia’s next generation of high-performance AI hardware (expected to reach the market in volume in 2025) will predominantly use liquid cooling to maximize performance. In response, some large data center operators are speeding up their planning and design for scaled DLC rollouts by a few years, Uptime Intelligence learned in private discussions with industry participants. Many operators expect to see significant adoption of DLC in their facilities soon, with most projecting that at least 10% of their racks will have liquid cooling within 5 years, according to Uptime Intelligence’s 2024 Data Center Cooling Systems Survey.