AI and Water: Why Data Centers Matter for Local Supplies

Microsoft’s expansion of AI infrastructure is reshaping its water needs, and the numbers are sizable. The company used about 7.9 billion liters in 2020 and about 10.4 billion liters in 2024, and internal forecasts once suggested its annual need could climb to 28 billion liters by 2030; engineers later revised that estimate to roughly 18 billion liters as new cooling designs and better data were incorporated. Such revisions, which reflect both design changes and updated information from leased facilities, show how quickly projections can shift when engineering choices are made.

In some regions the projected demand is especially visible. Projections showed large draws near Phoenix, in parts of India and near Jakarta, places that already face water stress; Microsoft has reduced some site estimates — for example, Phoenix estimates fell when higher operating temperatures were adopted, and Jakarta’s figure was also lowered after updated local data. These place-specific changes underline that a corporate forecast is not the same as a final, locally managed water plan.

To limit freshwater use, Microsoft is rolling out closed‑loop and chip‑level cooling systems that recycle water and avoid evaporation, and it says that ‘Next‑generation datacenters consume zero water for cooling.’ Pilot installations for the zero‑evaporation design are planned in Phoenix and Mount Pleasant, Wisconsin, and the company reports a sustained improvement in Water Usage Effectiveness across its fleet. While such systems can raise energy demand slightly, they can substantially cut the fresh water withdrawn each year.

Even so, companies, communities, and engineers must balance rapid AI growth with local water realities, and transparency about site data will help public understanding and planning.