UC Davis civil engineering professor Jay Lund calculates California AI data center water use at roughly 20,000-290,000 acre-ft/year, under 0.7% of California’s 40 million acre-ft annual human water use.
Key Takeaways
California has ~15 million sq ft of data center floor space; physics-based estimates yield 32,000-290,000 acre-ft/year assuming all-evaporative cooling, a worst-case ceiling.
A conservative cross-model estimate of ~20,000 acre-ft/year puts AI at roughly 0.055% of California’s total annual human water use.
Per-unit-area, data centers evaporate 25-150x more than irrigated agriculture, but total footprint is tiny compared to California’s 7 million irrigated acres.
Four AI models (ChatGPT, Claude, Gemini, Copilot) gave ranges spanning 2,300-400,000 acre-ft/year; the author treats this as validation that quick AI-assisted estimation is viable for policy scoping.
Outside the arid West, new data center water demand could offset revenue losses from urban conservation, since water systems have excess capacity.
Hacker News Comment Review
Commenters pushed back on the evaporative-cooling assumption: closed-loop and air-cooled systems are common, meaning the article’s upper-bound estimates may be inflated even further beyond reality.
The mandatory-vs-optional framing generated debate. Some argued comparing AI to agriculture is unfair since food water use is non-negotiable; others countered that beer production in Arizona already exceeds data center use, undermining any necessity hierarchy.
Local concentration effects were flagged as the real issue: a single Google facility in one municipality demanding 2-8 million gallons of drinking water per day stresses local infrastructure even if statewide totals are trivial.
Notable Comments
@Springtime: cites a lawsuit revealing a single Google data center near local supply limits at 2-8 million gallons of drinking water per day, illustrating local vs. aggregate risk.
@bee_rider: notes evaporated cooling water returns as rain, unlike toxic industrial discharge, questioning whether evaporation loss framing is meaningful at all.
