How liquid cooling cuts data-center water use to near zero: a simple explainer

As artificial intelligence drives a building boom in data centers, an uncomfortable side effect has drawn growing scrutiny: the enormous amount of water these facilities consume to stay cool. A cooling design highlighted in a widely shared technical post describes a way to run the systems warm, at around 45 degrees Celsius, and in doing so cut water use to near zero.

To understand the breakthrough, it helps to know why data centers use water at all. The chips inside generate heat, and that heat has to go somewhere. Many facilities rely on evaporative cooling, which works much like sweating: water is evaporated to carry heat away. It is effective, but it consumes large volumes of water that are lost to the air.

The newer approach is liquid cooling, in which coolant is piped directly to the chips rather than blowing cold air across them. Liquid carries heat far more efficiently than air, which is increasingly necessary as AI accelerators pack more power into each rack than traditional servers ever did.

The key insight in the design is the operating temperature. Conventional cooling tries to keep coolant cold, which often requires energy-hungry chillers or water-wasting evaporation. By engineering the system to run warm, at roughly 45 degrees Celsius, the facility can reject heat to the outside air without needing to evaporate water to get there.

That warm-running strategy is what slashes water consumption. When the coolant is already warm relative to the outdoors, the heat can be shed through dry coolers, essentially large radiators, rather than evaporative towers. The water that would otherwise be boiled off into the atmosphere simply is not needed.

The environmental stakes are considerable. Data centers have faced criticism in regions where their water draw competes with agriculture and households, particularly in hot, dry climates that are otherwise attractive for cheap land and power. A design that removes most of the water demand directly addresses one of the industry's sharpest local conflicts.

There are trade-offs to weigh. Running hardware warmer requires careful engineering to ensure chips stay within safe limits, and the components must be designed to tolerate higher coolant temperatures without losing reliability or performance. The approach works best when the whole system, from chip to cooling loop, is built around it.

The timing matters because AI workloads are intensifying the heat problem. Modern accelerators concentrate so much computation in a small space that air cooling alone struggles to keep up, which is pushing the industry toward liquid cooling regardless. Designing those liquid systems to also save water turns a necessity into an environmental gain.

Water is only part of the data center's footprint, of course. These facilities still draw significant electricity, and the source of that power shapes their overall climate impact. Reducing water use does not erase the energy question, but it removes one of the most locally contentious pressures.

The broader lesson from the explainer is that infrastructure choices made now will shape AI's resource footprint for years. By rethinking something as basic as the temperature a system runs at, engineers can change how much water an entire industry consumes, an example of how careful design can blunt the environmental cost of the computing boom.