Liquid cooling becomes standard as racks exceed 100 kW

What the development involves

In direct liquid cooling, coolant is delivered to cold plates mounted directly on the processors. A coolant distribution unit (CDU) separates the server loop from the facility's cooling system. Schneider Electric has built its own portfolio following the acquisition of Motivair (Schneider Electric, 2025), and Google shares its CDU design through the Open Compute Project (Google Cloud).

The technology is already in use at scale. AirTrunk reports more than 20 MW of liquid-cooled capacity at its JHB1 facility in Malaysia (AirTrunk, October 2024).

What benefits are documented

A life cycle assessment from Microsoft, published in Nature, shows that cold plates and immersion cooling, compared with air cooling, deliver 15–20% lower energy demand, 15–21% lower greenhouse gas emissions and 31–52% lower blue water consumption. The system boundary covers the entire chain from chip to building and grid (Microsoft Research).

HPE reports up to 90% lower electricity use for cooling itself (HPE). This is a vendor claim and applies only to cooling, not to the facility's total electricity use.

What it could mean in Sweden

For new facilities, the question is no longer whether there should be liquid cooling, but what share of the load it should carry and how the system is built out in phases. Piping, space for coolant distribution units and temperature levels are decided early and are difficult to change later.

What needs to be investigated

How the remaining air-cooled load is handled. How leakage risk and water quality are managed. What the interface between building and IT should look like while standards are not yet fully aligned, and how the choice of IT and supplier is left open to the operator.