
Fact check: Sam Altman says an almond uses as much water as 38K ChatGPT queries
OpenAI CEO Sam Altman claimed 38,000 ChatGPT queries use as much water as one California almond. PolitiFact ruled this "Mostly False," stating Altman used a low estimate for ChatGPT's water use and a high estimate for almonds. Actual estimates suggest 1,000 to 10,000 conversational AI prompts are needed to equal the fresh water consumed by one California almond.
OpenAI CEO Sam Altman's claim that 38,000 ChatGPT queries consume the same amount of water as the production of a single California almond has been fact-checked by PolitiFact and deemed "Mostly False." Altman made the comparison on the Sources Podcast, acknowledging he was unsure of the exact figure but believed it to be close. PolitiFact's analysis concluded that Altman's comparison relies on an unconfirmed, relatively low estimate for ChatGPT's water usage and a significantly high estimate for almond water consumption.
The fact-check reveals that Altman likely drew his figure from a 2025 blog post on ChatGPT's average water use, which reported 0.000085 gallons (0.32 milliliters) per query, a number for which OpenAI has provided no public details. For almonds, he used a 2019 study's estimate of 12 liters of water per almond, which included rainfall and "grey water" to dilute pollutants, not just direct fresh water irrigation. Closer estimates for fresh water consumption by an almond are around 6 liters.
When comparing the 6 liters of fresh water per almond to various estimates for AI prompts (ranging from 0.6 to 17 milliliters per prompt depending on factors like model and length), PolitiFact found that it would take anywhere from 1,000 to 10,000 conversational AI prompts to equal the water used for one California almond. Experts like Shaolei Ren, a University of California, Riverside, professor, emphasize the difficulty in precisely measuring AI's water demands due to varying data center locations, cooling types, power sources, and AI model efficiencies. While agriculture uses vastly more water nationally than the AI industry, the specifics of individual AI queries remain complex and difficult to verify publicly.