\n\n
Key developments: Recent research shows that the vast majority of water used by AI data centers is not the on‑site cooling water but the water consumed at power plants to generate the electricity they need. Meta disclosed that indirect use made up about 96% of its 21 billion gallons of water in 2024, a 120% rise since 2021. University of Michigan estimates add another 14 billion gallons for three planned Michigan centers, while U.S.‑wide indirect use may be 12 times the 17.5 billion gallons of direct use, according to Lawrence Berkeley National Laboratory. Estimates range from 211 billion gallons to as high as 4 trillion gallons annually.
Why it matters: The indirect water demand links AI growth to water stress in regions already facing drought, such as Arizona and Indonesia. In Arizona, new AI facilities could add 5 billion gallons of onsite water demand and orders‑of‑magnitude more offsite, threatening the Colorado River supply that feeds Phoenix. In Indonesia’s Batam island, residents have protested severe household water shortages they attribute to data‑center expansion. Without accounting for this hidden consumption, policymakers and companies risk underestimating AI’s environmental impact and overlooking critical water‑security challenges.
What to watch next: • Standardized reporting protocols for indirect water use, urged by experts like Brandon Oyer (AWS). • Adoption of low‑water cooling technologies (air‑chilled, closed‑loop, immersion) and their trade‑offs in electricity demand. • Regulatory actions, such as California’s new data‑center water‑estimate bills and Arizona’s water‑allocation cuts. • Corporate disclosures beyond onsite use—Meta, Amazon, and others may expand reporting as stakeholder pressure grows. [[1, 2]]