Schneider says hotter coolant can make AI datacenters less thirsty
Modeling suggests liquid-cooled designs can halve water consumption, though the company has skin in the game
Schneider Electric's white paper reveals that the cooling architecture of a datacenter significantly impacts its water and energy consumption. The firm compares four models of a 100 MW datacenter, from conventional air cooling to optimized liquid cooling systems. The analysis shows that shifting from air to liquid cooling generally enhances energy efficiency and can lower water usage in areas with water scarcity.
As Schneider is investing in liquid cooling technology, the findings align with the company's interests. Their model indicates that air-cooled facilities in Dallas, Texas, consume the most onsite water. Liquid cooling at 45°C (113°F) reduces water use by at least 50%. The results emphasize the importance of location, operating temperature, and cooling architecture in determining a facility's water requirements, even as AI workloads grow.
Water consumption is primarily influenced by the equipment that dissipates heat outside the data center, with cooling towers consuming 5 to 20 times more water than dry coolers. Schneider suggests that liquid cooling sets at 32°C (90°F) is typical for early AI datacenters. By increasing the supply temperature to 45°C (113°F), the company claims to achieve substantial water savings due to the extended range of conditions where ambient air can reject heat without mechanical chilling.
The report advises early evaluation of water consumption during the design phase and consideration of alternative water sources to lessen the demand on local supplies. The debate over datacenters' water usage is particularly heated in the US, where consumption has increased by threefold in the last decade, and local opposition to new projects is rising.
In the UK, the government has been criticized for overlooking water demand while promoting AI server farm construction. Schneider emphasizes that evaluating water and energy efficiency concurrently is crucial, as the optimal design depends on the relative importance of power usage effectiveness (PUE) and water consumption, as well as local energy and water costs.
Written by urgent.news from The Register Science's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
