OpenAI CEO Sam Altman has just waded into one of tech's hottest fights, telling the Sources Podcast, released on September 2, that the panic over AI guzzling water has been wildly blown out of proportion. He claimed that roughly 38,000 ChatGPT queries use the same amount of water as producing a single California almond, and that a modern, very large data center uses about as much water as an ordinary office building, mostly for things like sinks and toilets. He also took aim at viral posts comparing one ChatGPT question to hours of showering, calling the whole thing a meme that falls apart under scrutiny, and argued that the image of thirsty old-school cooling towers is outdated. Here's the twist: Altman admitted he was reciting the almond number from memory and that it might not be exactly right.

Fact-checkers have not let it slide. PolitiFact rated the comparison Mostly False, saying it leans on a very high estimate for almond water use and a low, unconfirmed estimate for ChatGPT, and noting that OpenAI does not publish details about its overall water use. Plug in Altman's own earlier figure of about 0.32 milliliters per query and a 2019 study's estimate of roughly 3.56 liters per almond, and the math lands near 11,000 queries, not 38,000, and that only counts cooling water on site, not the water used to generate the electricity behind each answer. CalMatters found that public data is too thin to verify the comparison either way, while the Congressional Research Service says data centers are using more water than they used to, and unlike farms, they often draw treated drinking water from public supplies. So is one chatbot question really just a drop in the bucket, or is the bucket getting a whole lot bigger as billions of prompts pile up?