OpenAI CEO Sam Altman has dismissed concerns over artificial intelligence’s water consumption as a widely repeated “meme,” noting that modern data centers generally use far less water than their public image suggests.
Altman addressed the issue during the premiere episode of the Sources podcast with Alex Heath, published Tuesday, as he was asked about the growing resistance to AI and the environmental concerns surrounding the rapid expansion of data center infrastructure.
AI companies have faced increasing scrutiny over the electricity, land and water required to build and operate the computing infrastructure needed to train and run powerful models. Altman acknowledged that the industry has developed a reputation for consuming large quantities of water, but disputed the scale of that impact.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
“That has been a robust meme and difficult to disprove, but I don’t think holds up to any scrutiny,” Altman said.
He argued that modern data centers typically consume roughly as much water as an office building. That claim is broadly consistent with some available data, although water consumption varies significantly depending on a facility’s cooling technology, size, location and local climate.
A 2024 report by a Virginia state government commission, examining the state’s rapidly expanding data center industry, found that water use differed substantially between facilities. However, most of the data centers examined consumed amounts of water comparable to or below those of an average large office building.
The issue remains one of the most contentious elements of the backlash against AI infrastructure. Data centers can require substantial quantities of water when they use evaporative cooling systems, particularly in hot or water-stressed regions.
A Gallup survey published in May found that 71% of Americans opposed the construction of a data center in their local area. About 70% of respondents said they were concerned about the environmental impact, while half cited the effect on local resources. Water specifically was identified by 18% of respondents.
A Business Insider investigation last year also linked data center development to worsening water pressures in parts of the U.S. West, citing the use of water-intensive evaporative cooling systems among the factors contributing to consumption.
The industry, however, has increasingly sought to reduce its dependence on fresh water. The Data Center Coalition, an industry trade group, says developers use a range of cooling technologies, including closed-loop and waterless systems, and consider local water availability when designing facilities.
Altman also sought to put AI’s water consumption into perspective by comparing ChatGPT queries with California’s almond industry, one of the state’s more water-intensive agricultural sectors.
He said growing a single California almond requires more water than thousands of ChatGPT queries.
“The people that are scarfing down 12 almonds at a time don’t feel like they’re doing something horrible from a water perspective, for the most part,” Altman said.
The comparison has limitations, however, because the water footprint of an AI query and the agricultural water requirements of a crop are calculated in very different ways. It also does not account for differences between data centers, cooling systems, locations, or the source of the water being consumed.
Still, available estimates support the broad magnitude of Altman’s comparison.
Altman has previously estimated that an average ChatGPT query consumes about 0.32 milliliters of water. Google has estimated that an average text prompt on its Gemini app uses about 0.26 milliliters.
By comparison, a 2019 study by researchers affiliated with the U.S. Geological Survey estimated that producing a single California almond required an average of about 3.56 liters of water, or slightly less than a gallon.
Using those figures, the water associated with one almond would be equivalent to roughly 11,000 ChatGPT queries at Altman’s stated estimate.
The larger debate, however, is less about the water consumed by an individual AI prompt than about the aggregate demand created by billions of queries and the construction of increasingly large data centers.
A small amount of water per interaction can become a significant resource requirement when multiplied across massive user bases, while the location of the infrastructure can determine whether that consumption places meaningful pressure on local water supplies.
This is where Altman’s argument faces its most important test. Saying that AI uses relatively little water on a per-query basis does not necessarily resolve concerns about the cumulative impact of rapidly expanding computing infrastructure, particularly in regions already facing water scarcity.
Altman nevertheless argued that the industry should focus less on defending AI against criticism and more on demonstrating its usefulness.
“I think the right way to get people to like something is to deliver them value,” he said.
“The industry has got work to do in terms of how we make these products easy to use and easy for people to get a lot of value out of.”
AI companies are investing hundreds of billions of dollars in computing capacity while attempting to convince governments and communities that the economic and technological benefits of that infrastructure outweigh its environmental costs.