Overview
- OpenAI CEO Sam Altman told the Sources podcast that about 38,000 ChatGPT queries use the same water as producing one California almond and said he was recalling the number from memory so it could be off.
- Altman has previously estimated a single ChatGPT query uses roughly 0.32 millilitres of water, and using that figure with a 2019 USGS‑linked estimate of 3.56 litres per almond produces about 11,000 queries per almond, illustrating the arithmetic sensitivity of these comparisons.
- Experts and state reports show data‑centre water use varies widely because cooling method, facility design and local water sources determine direct withdrawals and total water footprint.
- Companies point to newer cooling approaches, such as closed‑loop and liquid cooling, that cut direct water use, but documented heavy draws in some Georgia projects and a Gallup survey showing 71 percent of Americans oppose a data centre near their home keep public concern high.
- The exchange is likely to prolong scrutiny of AI infrastructure as regulators, communities and firms press for clearer disclosure of water metrics and for design choices that limit withdrawals in water‑stressed regions.