Sam Altman says concerns about the amount of water used by AI data centers are exaggerated, even though the perception has been difficult for the industry to change.
Speaking on the debut episode of Alex Heath’s Sources podcast, the OpenAI CEO said modern data centers are often portrayed as major water consumers, but argued that most use roughly the same amount of water as a large office building.
Altman described the idea that AI data centers are extreme water guzzlers as a persistent misconception that, in his view, does not hold up under closer scrutiny.
A 2024 report from a Virginia state commission found that although some data centers consumed significantly more water than others, most used an amount comparable to or lower than an average large office building.
Water consumption has become one of the main concerns surrounding the rapid expansion of data centers. A Gallup survey published in May found that 71% of Americans opposed having a data center built near them, while 70% expressed concern about their environmental impact. Half of respondents pointed to pressure on local resources, with 18% specifically mentioning water.
Previous reporting has also found that some data centers can place additional pressure on water supplies, particularly when they rely on water-intensive evaporative cooling systems. The Data Center Coalition says operators use a range of cooling technologies, including closed-loop and waterless systems, while considering local water availability.
Altman also compared ChatGPT’s water use with the amount required to grow almonds in California. He argued that producing a single almond can consume more water than thousands of ChatGPT queries.
Altman has previously said the average ChatGPT query uses about 0.32 milliliters of water. Google has estimated that an average Gemini text prompt uses roughly 0.26 milliliters. By comparison, a 2019 study involving U.S. Geological Survey researchers estimated that growing one California almond requires around 3.56 liters of water.
Although Altman believes concerns about AI water consumption are overstated, he said the technology industry still has work to do in changing public perceptions. He argued that the best way to win people over is by making AI products easier to use and delivering clear value to users.