The aggregate water footprint of generative AI is small relative to common consumer products (a year of heavy ChatGPT use is on the order of 300 to 900 litres, against 15,415 litres per kilogram of beef and 2,495 litres per cotton t-shirt; data centers globally remain a small single-digit fraction of one percent of freshwater withdrawals, with agriculture at about 70% and industry at about 20%); the real governance concern is geographic concentration of consumption in water-stressed regions (Microsoft 2024 disclosure: 42% of consumption in water-stressed areas; Microsoft West Des Moines, Iowa Jul 2022 about 11.5M US gallons / about 6% of the city that month per AP 9 Sep 2023), not global query volume.

Claim is scoped to the aggregate-vs-local distinction that is collapsed in most public AI-water discourse. The aggregate-is-small reading is anchored on three data sources: (a) the Li et al. arXiv:2304.03271 (2023) 500 ml-per-10-to-50-queries figure with the stated 5x geographic variance band, (b) the Water Footprint Network / Mekonnen-Hoekstra (2010, 2012) consumption-based water footprint dataset for food, beverage, and textile products, and (c) the UNESCO World Water Development Report 2024 / FAO AQUASTAT / USGS Circular 1441 aggregate sectoral split (agriculture about 70%, industry about 20%, municipal about 10%). The locational-concern reading is anchored on (d) Google 2025 Environmental Report and Microsoft 2024 Environmental Sustainability Report own-disclosure water-stress percentages, and (e) AP 9 Sep 2023 reporting on West Des Moines, Iowa. The claim does not assert that AI water use is zero, that it will not grow, or that no locally significant facility loads exist. It asserts that the aggregate framing collapses under comparison to the household and procurement footprint while the locational framing does not. 90-day review cadence is calibrated to the slow pace of water-footprint dataset updates and to the annual rhythm of hyperscaler environmental reports. Trigger conditions: (1) a peer-reviewed update to the Mekonnen-Hoekstra dataset or to Li et al. with materially different numbers would move the claim toward Partial; (2) an IEA, IPCC, or UNESCO publication that revises global sectoral water-use shares (agriculture, industry, municipal) by more than 10 percentage points would move toward Partial; (3) a single hyperscaler disclosing total water consumption that materially closes the gap to global industrial consumption (currently roughly 800 cubic kilometres per year) would move toward Partial or Not holding; (4) a published EU AI Act, CSRD, or US SEC enforcement action invoking aggregate-AI water use (rather than locational concentration) as the material concern would move toward Partial; (5) emergence of a credible peer-reviewed challenge to the consumption-based water-footprint methodology that the comparison rests on (notably the green-plus-blue water footprint convention) would require methodological note rather than status change. Sibling claim on the local-concentration governance question is planned under AM-008 revise.