Jay Lund, an emeritus professor at UC‑Davis, argues that AI’s water footprint in California is being blown out of proportion. Data centers powering large language models sit on roughly 15 million square feet of floor space, converting electricity to heat that must be cooled with water. Rough physics suggest evaporative cooling could consume between 32,000 and 290,000 acre‑feet annually.

Estimates derived from simple energy‑to‑water calculations yield a broad range, but four AI tools—ChatGPT, Claude, Gemini and Co‑Pilot—converge on a narrower band of roughly 20,000 acre‑ft per year. Even the high end represents less than 0.1 % of California’s total human water use, roughly 40 million acre‑feet, making AI a relatively efficient consumer.

The takeaway, according to Lund, is to treat AI water use as a data point, not a crisis. Transparent reporting can curb speculation, while policymakers should weigh it against larger agricultural demands. In practice, the modest consumption suggests that fears of AI draining California’s reservoirs are unfounded.

Lund’s calculations also reveal that evaporative cooling uses 25–150 times more water per square meter than irrigated agriculture, yet the total volume remains a fraction of statewide use. This contrast underscores that AI’s water demand, while not negligible, is dwarfed by traditional sectors and should be managed with proportionate policies.