Liquid cooling set to dominate AI data centres

We stand at an inflection point for AI data centres. And liquid cooling is set to dominate the future of such data centres.

While average rack densities are rising, it remains below 8kW per rack, says Uptime. But some cutting-edge AI deployments are now crossing 100kW per rack.

Greater heat capacity

As a result, AI data centres are increasingly looking to liquid cooling, which has 4x the heat capacity of air for the same mass. And liquid mass a lot more.

Liquid cooling can be divided into 3 main categories:

• Rear door coolers (RDC) - Chiller water to each rack.
• Direct to chip (DTC) - Chilled water to CPUs and/or GPUs.
• Immersion cooling - Immersion in a liquid (not water).

Mix and match

Liquid cooling isn't necessarily suitable for all data centres though. It faces challenges related to:

• Reliability.
• Maintenance
• Liquid quality.
• Floor loading (immersion).

Some insights from JLL:

• A hybrid approach is typically employed today.
• RDC and DTC are favoured for retrofitting DCs.

Indeed, the xAI Colossus supercomputer uses a hybrid DTC and RDC approach.

Air still has role to play

Finally, while immersion cooling is expected to play an increasing role, broad implementation is still "a few years" away, says JLL.

This means that air-cooling will continue to have an important role to play for some time yet.

Indeed, my opinion is that air cooling might never go away: The only question is the proportion of air-cooling and liquid cooling that will be deployed.

Check out the JLL 2025 Global Data Center Outlook here.