Sam Altman’s almonds and AI comparison doesn’t hold water

OpenAI CEO Sam Altman made a statement on the Sources Podcast comparing the water usage of ChatGPT queries to that of growing California almonds. He claimed 38,000 ChatGPT queries use the same amount of water as one California almond. Altman prefaced this by noting he wasn't sure of the exact number but believed it was close. His comparison appears to be based on an unverified 2025 blog post estimate for ChatGPT's water use and a 2019 study on California almond farming, which included rainwater and "grey water" in its calculations.

PolitiFact analyzed Altman's claim and rated it "Mostly False," arguing it overstates almond water consumption and uses a potentially low and unconfirmed estimate for ChatGPT. While the 2019 almond study cited 3.2 gallons (12 liters) per almond by including all water types, a 2021 study and revised estimates indicate the direct freshwater consumption for an almond is closer to 1.5 gallons (6 liters).

Regarding AI's water demands, experts like University of California, Riverside professor Shaolei Ren highlight the difficulty in precise measurement due to varying factors like data center location, cooling methods, AI model, and prompt lengths. While data centers nationally used about 228 billion gallons of water in 2023, agriculture consumed nearly 30 trillion gallons for irrigation in 2020, indicating a vast difference in overall sectoral water usage.

Estimates for ChatGPT's water use per prompt also vary widely, with some studies suggesting 0.6 to 17 milliliters per prompt for its 4o model. Based on more accurate almond water usage and various AI estimates, it would likely take 1,000 to 10,000 conversational AI prompts to equal the 6 liters of fresh water used for one California almond, significantly undercutting Altman's original 38,000 queries figure.