\n\n
Key developments: Data‑center operators are rolling out water‑saving cooling technologies such as closed‑loop liquid and immersion cooling, which dramatically reduce the water needed on‑site. However, these systems increase electricity use, pushing the bulk of water consumption upstream to thermal power plants that need water for steam generation. Recent studies show indirect water use dwarfs direct use – up to 12 times higher – bringing total U.S. AI data‑center water footprints to between 211 billion and 4 trillion gallons annually, with Meta reporting 96 % of its 21 billion‑gallon usage as indirect. Global estimates now put AI‑related water use at 765 billion liters per year, and the International Energy Agency projects data‑center water demand could hit 1.2 trillion liters by 2030.
Why it matters: The shift from onsite to offsite water consumption obscures the true environmental impact of AI expansion, complicating water‑resource planning in drought‑prone regions like Arizona and the Southwest. Indirect usage ties AI growth to fossil‑fuel‑intensive power generation, amplifying both water stress and greenhouse‑gas emissions. Lack of standardized reporting, as highlighted by Meta’s disclosure and calls from industry leaders for a unified metric, hampers transparency and policy response.
What to watch next: • Adoption rates of ultra‑efficient cooling (e.g., immersion, micro‑fluidic) and their net electricity impact.
• Regulatory moves requiring indirect water‑use reporting, such as California’s new data‑center water‑estimate bills.
• Shifts toward renewable‑powered data centers, which could lower the water‑intensity of electricity supply.
• Forecasts for AI‑driven water demand by 2030, especially in high‑growth regions, to gauge whether supply‑side interventions (water‑rights allocations, drought‑risk assessments) keep pace with AI‑induced demand. [[1, 2, 3]]