Schneider Electric has published modeling that points to a straightforward way to make AI datacenters less thirsty: run the liquid coolant hotter. According to The Register, the company's analysis suggests that liquid-cooled designs can halve water consumption compared with conventional approaches, with the temperature of the coolant playing a central role in how much water the facility ultimately needs.
The logic is that warmer coolant requires less additional cooling and evaporation, cutting the water burden of heat rejection. Schneider is a major vendor of datacenter cooling and power infrastructure, so its enthusiasm for liquid cooling is not entirely disinterested. The Register notes that the company has skin in the game, meaning the projections should be read with that commercial context in mind.
Still, the modeling adds to a growing body of evidence that water efficiency will be a key design constraint for next-generation AI facilities. If the estimates hold up in real deployments, hotter coolant loops could become a standard tool for operators trying to reconcile dense compute with environmental limits.