Overview
- The UNU‑INWEH report released in early June projects AI‑focused data centres could use about 9.3 trillion litres of water by 2030, a volume equal to the basic annual domestic needs of roughly 1.3 billion people, and about 945 TWh of electricity per year.
- Researchers say everyday AI use, or inference, now drives roughly 80–90% of AI energy consumption so continued use of deployed models is the main factor behind the projected rise in power and water demand.
- The study warns that judging sustainability by carbon alone is misleading because some low‑carbon options like bioenergy can sharply increase water and land footprints, and greater efficiency can raise total consumption through higher use.
- Local responses are accelerating as droughts and strained grids intersect with data‑centre buildouts, with utilities and municipalities reviewing permits, debating who pays for upgrades, and considering moratoria in places such as Tennessee and Texas.
- The report notes compute capacity is concentrated in a few countries, which concentrates environmental and governance pressures, and calls for integrated planning of energy, water and land use to protect communities and guide regulation.