Sam Altman (샘 알트먼), chief executive of OpenAI, pushed back against concerns over water use by AI data centres, saying an exaggerated perception is spreading.

On Sept. 2 (local time), Business Insider reported that Altman set out his position on the issue of water consumption, cited as one reason for backlash against AI, on the first episode of the podcast "Sources".

Altman said modern data centres are known as facilities that use excessive water, but actual usage is similar to office buildings. He said the perception has hardened like a "strong meme" and is not easy to rebut, but argued it does not hold up even with a little scrutiny.

The remarks came as community opposition grows over the spread of data centres. A May Gallup survey cited by the outlet showed 71 percent of Americans oppose construction of a data centre near their home. Of them, 70 percent cited concerns about environmental impact and half mentioned resource use. Those who specifically cited water use were 18 percent.

Actual water use by data centres varies by facility. A 2024 report by a Virginia state government commission said some facilities used far more water than others, but most data centres recorded levels the same as, or lower than, an average large office building. Virginia is considered a major U.S. hub for data centres.

It was also raised that controversy over water use is not based only on unfounded claims. In the western United States, studies have already found that water-intensive cooling technologies such as evaporative cooling are worsening water shortages. The Data Center Coalition, an industry group, said developers use various methods including closed-loop circulation and waterless cooling, and prioritise efficient operations that reflect local water conditions.

Altman also compared ChatGPT's water use to California almond farming. He said, "The water needed to grow one almond in California is more than the water used for thousands of ChatGPT queries." He added, "People who eat 12 almonds at a time generally don't feel like they're doing a terrible thing from a water perspective."

The outlet said the comparison has not been independently verified, but broadly points in the same direction as publicly available figures. Altman said average water use per ChatGPT query is about 0.32 millilitres, and Google has put average water use per text prompt on the Gemini app at 0.26 millilitres. By contrast, a 2019 study by researchers affiliated with the U.S. Geological Survey found producing one California almond requires an average of 3.56 litres of water.

Altman said concerns about water consumption are exaggerated but acknowledged the industry has responsibility for changing public opinion. "The right way to make people like something is to provide value," he said, adding, "The industry still has the task of making products easier to use and ensuring users get greater value."

In this environment, the debate over data centre water use is expected to remain an issue linked to community acceptance of AI infrastructure beyond simple comparisons of numbers. Data centre companies face pressure to explain cooling methods and region-specific resource-use strategies in more detail, and AI companies are left with the task of managing environmental burden controversies while demonstrating the utility of their services.