As Gov. Gavin Newsom decides whether to sign legislation forcing data centers to reveal how much water they consume, OpenAI CEO Sam Altman is pushing back on the idea that artificial intelligence poses a serious water problem, insisting the concerns don’t hold up under scrutiny.
Speaking on the Sources Podcast with Alex Heath this week, Altman claimed it takes 38,000 ChatGPT queries to use as much water as growing a single almond in California. He compared a large modern data center’s water footprint to that of an ordinary office building, saying it’s roughly equivalent to what people use running sinks and toilets.
But experts who study the industry say Altman’s comparison is nearly impossible to verify, largely because tech companies release so little information about their operations.
“The information we have is so limited,” said Shaolei Ren, a professor of electrical and computer engineering at UC Riverside who has researched data center water consumption extensively.
Ren said there are simply too many variables to reduce data center water use to a single, tidy statistic. Location, outdoor temperature, cooling technology, the length of a query, and how much computational reasoning a task requires all shape how much water is consumed. Boiling that complexity down to a comparison with almonds, he said, oversimplifies a much murkier picture.
Michael Kiparsky, director of the Wheeler Water Institute at UC Berkeley’s Center for Law, Energy, and the Environment, agreed that the lack of public data is precisely why lawmakers are pushing for more transparency.
“There’s no published data — certainly not for California — and what is public seems all over the place,” Kiparsky said.
Two bills addressing that gap, authored by Assemblymember Diane Papan, D-San Mateo, cleared the Legislature in recent weeks and now await the governor’s signature. One would require data center operators to disclose their water sources and consumption levels when applying for business permits. The other would prevent local governments from approving new data center projects unless developers reveal their water usage plans and cover the cost of any necessary upgrades to local water infrastructure.
This marks Papan’s second attempt at requiring greater disclosure. Newsom vetoed a similar measure last year, saying he wanted more clarity on how such requirements might affect businesses and consumers before signing rigid reporting mandates into law.
Tech and business groups have again lined up against the proposals. But with communities across California increasingly clashing over proposed data center projects — and some cities moving to ban them altogether — pressure is mounting to get a clearer picture of their true water footprint.
“If we know what an almond uses, maybe we can figure out what a data center uses,” said Alex Biering, a policy advocate focused on water issues for the California Farm Bureau Federation.
Whatever the precise ratio between ChatGPT searches and almond production — a figure that fluctuates significantly depending on time and location, according to a 2019 study — federal researchers say one thing is clear: data centers are consuming far more water than they used to.
One study found that data centers directly used about 17 billion gallons of water in 2023, up sharply from 5.6 billion gallons in 2014. Another industry report projected that so-called hyperscale data centers could consume as much as 150 billion gallons of water between 2025 and 2030 — roughly equivalent to the annual water use of 4.6 million American households.
Ren cautioned that annual totals don’t tell the whole story. His research, currently under peer review and available as a preprint, found that data centers require significantly more water for cooling during the hottest days of the year, straining water systems at their most vulnerable moments.
“For community water systems, the peak capacity is more important because they need to be designed and built to accommodate the peak demand at all times, similar to our power grid,” Ren said.
His team estimated that data center cooling needs could add between 697 million and 1.45 billion gallons of peak daily water demand — a volume comparable to New York City’s average daily water supply. Unlike agricultural irrigation, which often relies on untreated water, most data centers draw from treated, drinkable municipal supplies, the researchers noted.
That distinction matters because many water systems, particularly in California, are aging, underfunded, and already stretched thin, according to the study, leaving them ill-equipped to absorb sudden surges in demand from large data centers. Ren’s team cited real-world examples in Georgia and Virginia, where local water utilities have warned they may not have enough supply to meet peak demand if multiple data centers draw from the system at the same time.
Ren noted that data centers could cut their water use by as much as half by switching to less water-intensive cooling technology, such as dry cooling systems. But that trade-off often comes at the cost of higher energy consumption — which, in turn, requires its own water resources to generate.
Original source: CalMatters
