Sam Altman, CEO of OpenAI, is pushing back on concerns about the water consumption of AI data centers, arguing the issue has been exaggerated and ignores recent advances in cooling technology. Speaking on the Sources Podcast with host Alex Heath, Altman addressed the growing backlash against AI development and offered a striking comparison to make his case.
According to Altman, producing a single almond in California uses roughly the same amount of water as 38,000 ChatGPT queries. He acknowledged he was recalling the figure from memory but said it reflected a complete water accounting rather than the usage of a single facility.
What Altman Said About Data Center Cooling
Altman referenced a widely circulated claim that each ChatGPT query consumes water equivalent to running a shower for six hours, water that is never recovered. He dismissed that framing as inaccurate. “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 said, describing it as “the full-on, total true water accounting.”
He also addressed the cooling methods used by large facilities. While Altman conceded that data centers once relied on evaporative cooling, he said that practice was abandoned long ago. A modern very large data center, he argued, uses about as much water as an office building running sinks and toilets. He called the water criticism “a robust meme” that is “difficult to disprove” but does not hold up to scrutiny.
How the Numbers Compare
Almonds are among the most water-intensive crops grown in the United States. Researchers estimate that it takes roughly 1.1 gallons of water to grow a single almond. Altman’s own past statements suggest a single ChatGPT response uses between 1 and 50 milliliters of water, or about 0.0002 to 0.013 gallons per query, figures tied to older evaporative cooling systems.
Those earlier estimates translate to roughly 85 to 5,500 queries per almond depending on the cooling method, well short of the 38,000 figure Altman cited. Training workloads can consume far more. Estimates suggest that GPT-3 used nearly 185,000 gallons of water on older systems.
Why AI Water Use Remains a Concern
The debate is not settled by newer cooling alone. The rising popularity of AI, and especially AI agents that consume far more tokens than standard chatbots, means data center demand continues to climb. As that demand grows, facilities could still draw more water than almond farming, keeping the environmental question in play even as cooling technology evolves. The comparison Altman drew places one almond at 1.1 gallons of water.
Source
Image: tomshardware.com
