Overview
- OpenAI CEO Sam Altman said on the Sources podcast that about 38,000 ChatGPT queries use roughly the same water as producing one California almond and he acknowledged he was recalling the number from memory.
- Using Altman’s earlier public figure of about 0.32 millilitres per query and a 2019 estimate of 3.56 litres per almond produces a different ratio of roughly 11,000 queries per almond, which fact‑checkers cite to show the 38,000 number is approximate.
- Researchers and engineers warn that single‑number comparisons are unreliable because data‑center water use depends on cooling technology, local climate, workload and whether treated potable water is used.
- California lawmakers have passed two disclosure bills requiring operators to report water sources and plans and to tie permitting to water commitments, and both measures now await Governor Newsom’s signature.
- Independent studies show aggregate data‑center water use has risen and can create large peak demands that strain community supplies, so choices about cooling systems carry trade‑offs for local water systems, energy use and residents’ access.