1 almond’s water equals to 38,000 ChatGPT queries says Sam:What does comparison mean, and how much water does AI data centre actually use

AI’s water use has become a growing concern as more data centres are being built around the world. OpenAI CEO Sam Altman has now offered an unusual comparison to explain how much water an individual ChatGPT query may use. According to Altman, the water needed to produce a single California-grown almond could be enough for thousands of ChatGPT queries. However, the exact comparison is different when the figures he previously shared are used. Speaking on the Sources podcast, Altman discussed concerns about the environmental impact of AI and said the amount of water used by individual AI queries is often exaggerated. Altman explained: I hear more from people, I was able to get a cure to this horrible disease than, you know, I think that ChatGPT is using up all the water in the world. Altman compares almonds with ChatGPT queries While discussing AI’s water consumption, Altman said that producing one almond in California uses roughly the same amount of water as 38,000 ChatGPT queries. Altman said: For every 38,000 ChatGPT queries, that is the same amount of water that is used in the production of a single almond in California. He also pointed to the everyday consumption of almonds, saying, “People who are scarfing down 12 almonds at a time don’t feel like they’re doing something horrible from a water perspective for the most part.” However, Altman later acknowledged that he was recalling the 38,000 figure from memory and that it was only an approximate number. So how much water does one almond use? Research has estimated that producing a California almond requires around 3.56 litres of water, or 3,560 millilitres. This figure includes the water needed to grow the almond crop. The number is useful for comparison, but it does not mean that 3.56 litres of water is physically contained inside an almond. It represents the water associated with growing and producing it. How much water does one ChatGPT query use? In June, Altman said that an average ChatGPT query uses around 0.32 millilitres of water. Using these two figures gives a different result from Altman’s 38,000-query comparison. 3,560 ml ÷ 0.32 ml = about 11,125 queries That means, based on these figures, the water associated with producing one almond would be equivalent to roughly 11,000 ChatGPT queries, rather than 38,000. The difference highlights why the exact assumptions behind estimates of AI’s water use matter. Altman says AI’s water use has become a ‘robust meme Altman also criticised what he described as exaggerated claims about AI and water consumption. He said the perception around data centre water use had become a “robust meme” that was difficult to correct. “I saw this thing going around about the water usage of ChatGPT. And it was like every time you run a single ChatGPT query, it’s like, you know, you run your shower for like six hours, and the water never comes back,” he said. According to Altman, modern data centres are also different from some older facilities that relied heavily on evaporative cooling. He argued that a very large modern data centre can use an amount of water comparable to an office building for routine activities such as sinks and toilets, although water consumption varies considerably between facilities and cooling systems. “I don’t think it holds up to any scrutiny,” Altman said of some of the claims circulating online. Why are people worried about data centre water use? The debate is not simply about how much water is used by one ChatGPT query. AI models require powerful computers housed in data centres. These machines generate significant amounts of heat and need cooling systems to keep them operating safely. As AI demand grows, more data centres are being built. This has raised concerns about their impact on local electricity supplies, water resources and the environment. A May Gallup survey cited by Business Insider found that 71% of Americans opposed having a data centre built near their home. Environmental impact was one of the reasons given by respondents, while some specifically mentioned water use. Data centre water use can vary Not every data centre uses the same amount of water. The amount can depend on factors such as the cooling technology being used, the local climate, and how efficiently the facility operates. A 2024 report from a Virginia state government commission found that water consumption differed significantly between facilities, although many data centres used amounts comparable to or lower than those of large office buildings. At the same time, researchers have warned that water-intensive cooling systems can put additional pressure on water supplies in already dry regions. Companies are trying to reduce water consumption Technology companies are also working on ways to reduce the amount of water their data centres require. Amazon reported that the water used by its data centres fell by 2% in 2025 compared with the previous year, although the company still used more than 9 billion litres of water during the year. Microsoft has also said that the water efficiency of its newer data centres has improved significantly compared with its earliest facilities. Almonds vs AI: what does the comparison really tell us? Altman’s almond comparison is meant to put the water use of an individual AI query into perspective. But comparing one almond with one ChatGPT query does not tell the whole story.
An individual query may use a very small amount of water, but billions of queries and the construction and operation of large data centres can add up to significant resource use. Similarly, agriculture consumes enormous amounts of water because crops are produced on a massive scale. The bigger question is therefore not simply “How much water does one ChatGPT query use?” but “How much water does the rapidly growing AI industry require, and where does that water come from?”