\n\n
Key developments: Researchers and companies are highlighting that while newer cooling technologies dramatically cut onsite water use, the surge in electricity demand pushes water consumption upstream to power plants. Meta’s 2024 report shows 96% of its 21 billion gallons of water use was indirect, a 120% rise since 2021. University of Michigan work adds another 14 billion gallons for three planned Michigan AI centers. National estimates range from 211 billion to 4 trillion gallons of indirect water use, with Lawrence Berkeley lab noting indirect use is about 12 times direct consumption.
Why it matters: Indirect water use ties AI’s growth to the water‑intensive electricity sector, threatening water‑scarce regions like Arizona, where new AI facilities could demand 5 billion gallons onsite and far larger volumes indirectly. This hidden footprint strains watersheds, escalates regional water scarcity, and fuels community backlash, as seen in protests in Indonesia and pending moratoriums on data‑center projects.
What to watch next: Monitoring the adoption of low‑water cooling (closed‑loop, immersion) versus their electricity impact, corporate disclosures of indirect water use, and policy moves—such as California’s new reporting bills and Arizona’s water‑allocation cuts—that could force more transparent accounting and water‑efficient power generation for AI workloads. [[1, 2]]