The distinction Hoang draws is between liquid cooling (which is mandatory for the high-density racks used in generative AI workloads) and water consumption, which is not. Modern AI infrastructure requires racks running at 400 kilowatts or more, a thermal load that air cooling cannot handle. But liquid cooling, Hoang argues, can be implemented in a closed-loop configuration that circulates a sealed volume of cooling fluid without drawing from or discharging into external water supplies.
Schneider Electric presented two theoretical case studies comparing air-cooled and liquid-cooled facilities in Dallas, Texas and Paris, France. In Dallas, transitioning from air cooling to liquid cooling reduced projected annual water consumption from 382,000 cubic metres to 197,000 — a 48% reduction. In Paris, the drop was from 108,000 to 51,000 cubic metres, a 53% decrease. In both scenarios, a fully closed-loop system would eliminate external water draw entirely.
The company's closed-loop systems use a factory-sealed cooling fluid that Schneider Electric tests at its Motivair facilities. The fluid is designed to last the operational life of the data center without replenishment, with no evaporation or discharge. Rich Whitmore, CEO of Motivair by Schneider Electric, noted that liquid cooling has existed since the 1980s but is no longer optional: "People don't have a choice — if you want advanced AI systems going in, you have to cool them."
The remarks respond directly to investigative reports published over the past year documenting the water footprints of large-scale AI facilities and the strain they place on local water supplies. Schneider Electric's position is that those footprints reflect design choices made by operators, not unavoidable technical constraints — and that the industry has the tools to build differently.
This analysis is based on reporting from Sustainability Magazine.
Image courtesy of İsmail Enes Ayhan.
This article was generated with AI assistance and reviewed for accuracy and quality.
