OpenAI's Altman Claims ChatGPT Uses Less Water Than Almonds

OpenAI CEO Sam Altman has entered California's intensifying debate over data center water consumption with a striking claim: it takes 38,000 ChatGPT queries to use as much water as producing a single California almond.

Speaking on the Sources Podcast with Alex Heath, Altman argued that concerns over AI's water footprint don't hold "up to any scrutiny." He compared modern large data centers to office buildings in terms of water use for basic facilities like sinks and toilets.

The problem, according to researchers who study data center infrastructure, is that no one can verify those numbers. The data center industry remains largely opaque about its resource consumption, making independent fact-checking impossible.

Why it matters

As California faces ongoing water scarcity and AI facilities proliferate across the state, the lack of transparency around data center water use has become a flashpoint. Local governments are banning new facilities outright, and the state legislature has passed two bills requiring disclosure—now awaiting Governor Gavin Newsom's signature after he vetoed similar legislation last year.

The verification problem

"The information we have is so limited," said Shaolei Ren, a professor of electrical and computer engineering at UC Riverside. He explained that far too many variables affect data center water consumption to reduce it to a single convenient statistic.

Those factors include location, outside temperature, cooling system type, prompt length, computational complexity, and output size. All influence how much water a facility requires.

Michael Kiparsky, director of the Wheeler Water Institute at UC Berkeley's Center for Law, Energy, & the Environment, echoed that assessment. "There's no published data—certainly not for California—and what is public seems all over the place," he said.

What the available data shows

Despite the opacity, the Congressional Research Service reports that data center water use is climbing. One study found facilities directly consumed approximately 17 billion gallons in 2023, up from 5.6 billion gallons in 2014. Another report projected hyperscale data centers could use 150 billion gallons between 2025 and 2030—equivalent to the annual consumption of 4.6 million U.S. households.

Ren's research, currently under peer review, highlights a critical distinction: peak demand matters more than annual totals. On the hottest days when cooling requirements surge, data centers can strain local water systems designed around maximum capacity, similar to electrical grids.

His team calculated that data center cooling systems could require between 697 million and 1.45 billion additional gallons of peak water capacity daily—comparable to New York City's average daily supply. Unlike agricultural irrigation, this water typically comes from treated, potable municipal supplies.

The research identified examples in Georgia and Virginia where water suppliers expressed concern they "just don't have the water" to meet simultaneous peak demand from multiple data centers.

California's legislative response

Two bills by Assemblymember Diane Papan now await the governor's decision. One would require operators to report water sources and usage when seeking business permits. The other would block local governments from approving new data centers unless developers disclose water plans and pay for necessary infrastructure upgrades.

The tech industry and business groups opposed both measures. When Newsom vetoed Papan's similar 2025 bill, he cited reluctance to impose "rigid reporting requirements" without understanding full impacts on businesses and consumers.

Ren noted that water use could drop by as much as 50% if data centers adopted less water-intensive cooling systems like dry cooling, though that approach increases energy consumption—which itself requires water for power generation.

CalMatters first reported these details.