Last month, journalist Karen Hao posted a Twitter thread in which she acknowledged that there was a substantial error in her blockbuster book Empire of AI. Hao had written that a proposed Google data center in a town near Santiago, Chile, could require “more than one thousand times the amount of water consumed by the entire population”—a figure which, thanks to a unit misunderstanding, appears to have been off by a magnitude of 1,000.
In the thread, Hao thanked Andy Masley, the head of an effective altruism organization in Washington, DC, for bringing the correction to her attention. Masley has spent the past several months questioning some of the numbers and rhetoric common in popular media about water use and AI on his Substack. Masley’s main post, titled “The AI Water Issue Is Fake,” has been linked in recent months by other writers with large followings, including Matt Yglesias and Noah Smith. (Hao said in her Twitter thread that she would be working with her publisher to fix the errors; her publicist told me she was taking time off and was unavailable to chat with me for this story.)
When I called him to talk more about AI and water, Masley emphasized that he’s not an expert, but “just some guy” interested in how the media was handling this topic—and how it was shaping the opinions of people around him.
“I would sometimes bring up that I used ChatGPT at parties, and people would be, like, ‘Oh, that uses so much energy and water. How can you use that?’” he says. “I was a little bit surprised when people would be talking so grimly about just a little bit of water.”
As local and national opposition to data centers has grown, so, too, have concerns about their environmental impacts. Earlier this week, more than 230 green groups sent a letter to Congress, warning that AI and data centers are “threatening Americans’ economic, environmental, climate and water security.”
The AI industry has started fighting back. In November, the cochairs of the AI Infrastructure Coalition, a new industry group, authored an op-ed for Fox News that touched on environmental worries. “Water usage? Minimal and often recycled—less than America’s golf courses,” they wrote. One of the authors of the op-ed, former Arizona senator Kyrsten Sinema, is currently advocating in favor of a data center project in the state that has prompted local pushback, including because of concerns about water use. The coalition also approvingly retweeted a post from Masley on the impact of AI on energy prices. (Masley maintains an exhaustive disclaimer on his Substack refuting allegations that he’s being paid by industry to share his opinions.)
It’s true that much of the discussion around water use and data centers lacks nuance. While carbon emissions are a zero-sum game—we need to cut greenhouse gases as much as possible, period, and climate change’s impacts will touch us all, regardless of where emissions come from—water use is much more complex, and geographically varied. A project that can wreak havoc on one region’s water supply may be a great match for an area with healthier reservoirs or fewer thirsty industries.
Experts I spoke to agreed that people often have a muddled understanding of how data centers use water, and that their overall consumption, in many places, is less of a risk than the public may think. But as the number of data centers continues to grow across the country—and as the Trump administration rolls back environmental protections to encourage more development—it’s worth understanding what, exactly, data centers are using water for, and how popular estimates are produced. And it’s worth having a bigger conversation about how and why we’re choosing to use water to cool data centers in the first place.
How AI Uses Water
You may have seen estimates of how much water a ChatGPT prompt uses, including the statistic that writing an email with AI consumes an entire bottle of water. But calculating such a figure is more complex than simply slapping a metric on an “average” query, experts say.
Onsite at a data center, water is mostly used for cooling. Processors in data centers run hot, and circulating water through them is one way to keep them at the right temperature; the water that absorbs the heat is then transferred to a cooling tower, where some of it evaporates. Salty and brackish water can corrode machinery, so many companies use potable water, drawing directly from municipal supplies. (Some big companies, like Amazon, Meta, and Apple, are increasingly using municipal wastewater that has been treated.)
The amount used depends heavily on the individual data center. Using more water means that data centers can avoid running electric cooling systems. Using more electricity, by contrast, lessens the water footprint, but ups the power bill—and causes more greenhouse gas emissions. (There are technologies that use special kinds of cooling liquids to cut down on both electricity and water use, but they have the potential to introduce forever chemicals into the mix. That’s made some Big Tech companies skittish of investing too heavily in them.) Cooling needs intensify in the summer when the weather is hotter, so data centers may use more water—or power—then.
“Every location and every state is different,” says Fengqi You, a professor in energy systems engineering at Cornell and an author of a recent analysis on the most sustainable places to put data centers. “How much water you will need for the same amount of AI depends on the climate, depends on the technology used, depends on the [energy] mix.”
Complicating things even further, some calculations around AI and water also include indirect water use—mainly from the massive power generation needed for data centers—to estimate their total water footprint. These numbers are generally much bigger than onsite use, but the calculations themselves are region-dependent.
This type of calculation is standard when talking about indirect greenhouse gas emissions—it’s known as Scope 2 emissions accounting—but it’s relatively rare to use the same calculations for water, says computing researcher Jonathan Koomey, coauthor of a recent paper from the Lawrence Berkeley National Laboratory that crunched numbers around AI and water. Koomey says he is increasingly convinced that offsite water use from energy shouldn’t factor into data-center water footprints, simply because we don’t tend to count this use when we talk about other industries.
Finding out details about water use in a specific data center isn’t always easy: A lot of companies use nondisclosure agreements to hide even basic information about projects from the public. A city in Oregon dragged state newspaper The Oregonian through a monthslong legal battle in 2022 to avoid disclosing how much water a Google data center used, arguing that it was a “trade secret.” (Following the lawsuit, Google started disclosing how much water its data centers use in its annual sustainability report.)
If the complexity makes measuring the water use of a given data center difficult and contingent, isolating the effects at the level of a single user or prompt is nearly impossible. Understanding the environmental impacts of specific LLMs is almost entirely dependent on sustainability disclosures from Big Tech, and while some have gotten more transparent, a lot of questions still remain. When OpenAI CEO Sam Altman mentioned in a personal blog post this summer that an “average” ChatGPT query used “roughly one fifteenth of a teaspoon” of water, he gave some parameters for understanding the company’s water and energy use—but also didn’t clarify key details, like the definition of an “average” query and whether or not the figure includes the energy and water cost of training an AI model.
We Use a Lot of Water Without Thinking
One of Masley’s main arguments in his popular Substack post is that there are industries that currently use much more water than AI, and that context needs to be part of the conversation. This is undoubtedly true. A single burger takes more than 400 gallons of water to produce; a humble cotton T-shirt takes more than 700. The United States’ 16,000 golf courses, meanwhile, each have the potential to use on average between 100,000 to 2 million gallons of water per day. (For comparison, Google says its thirstiest data center in Iowa consumed about 2.7 million gallons per day in 2024; most of the company’s data centers used substantially less.)
Arizona, one of the areas in the US where data center growth is exploding, has more than 370 golf courses. I can understand some of Masley’s points when I think about all the water that has, for decades, been going to help people play golf in the middle of the desert, with seemingly no one making a fuss.
But experts caution against dismissing concerns about water outright. “In the near term, it’s not a concern and it’s not a nationwide crisis,” says Cornell professor You. “But it depends on location. In locations that have existing water stress, building these AI data centers is gonna be a big problem.”
Koomey made a similar point. While he notes that people have a tendency to exaggerate the environmental impacts of computers, data centers’ water use is “ not something where you can just hand-wave it away,” he says. “Each situation has to be evaluated in the context of the specific design of the facility that’s being proposed. You just can’t say a priori that it’s always not an issue.”
It’s obviously crucial for journalists to get their numbers right. But it’s also clear that data centers do not have a negligible impact on an area’s water supply when water is scarce. While Hao’s numbers on the data center in Chile were likely substantially off, a single facility requesting more than 100 percent of the water consumed by a city’s residents is still nothing to sneeze at. (Google paused and then halted the project Hao cited last year after a court ordered the company to reconsider the potential impacts of climate change on the aquifer. More than a dozen other data centers have been constructed or are planned in the Santiago region.) Chile is nearing its 15th straight year of an unprecedented drought, and water supplies near Santiago have reportedly been endangered by other industries, including lithium mining.
Why Water Use and Data Centers Is a Big Deal to Some People
These considerations are running up against an inconvenient truth: The American public needs to seriously reconsider how it thinks about water as a resource. Droughts across the American West, juiced up by climate change, are showing in real time that the way the US economy has oriented itself around a seemingly unending water supply is pretty quickly becoming unsustainable. A 2023 New York Times investigation found that groundwater reservoirs across the country—not just in areas experiencing drought—are being overpumped, threatening both drinking water supplies and economic activity.
Koomey says that the concern about AI and water reflects an age-old tension over how we appropriately price public resources for private use, especially when the scarcity of that resource has changed over time. “Part of what we’re seeing with water is that the rules and the norms and the prices are set based on a previous reality,” he says. “It does all come back to this question, of what is the value of the service being delivered?”
This seems like a pretty accurate analysis of what’s driving a lot of the reaction to water and AI. People who don’t think twice about eating a burger or buying a new T-shirt are angry about LLMs and water because they are rejecting the entire premise that AI is worth the price of its water use. A societal mass value judgement on AI is, fairly or unfairly, playing out in real time in the uproar around data centers. Part of the visceral environmental shame that folks like Masley—and maybe some readers of this newsletter—feel from others about their AI use is probably less about the specific water footprint of a ChatGPT search than about the acceptance of a culture where AI is woven into everyday life, regardless of its environmental impact.
And I don’t think it’s entirely unjustified to differentiate the conversation around AI and water from other thirsty industries, merely because of how much importance is being placed on the breakneck growth of AI right now—and how big the promises being made about this technology already are. After all, people who run golf courses aren’t dining with the Trump administration, gaining massive policy concessions and economic gifts in order to reshape society solely for their putting greens, which they say will simultaneously solve all of our problems while also putting most of us out of work. Big Golf isn’t making headlines about how it’s upending entire industries, or causing mass psychosis, or keeping coal plants open—or how we could all be facing an economic catastrophe if the market for golf bottoms out.
It is correct to question the environmental trade-offs of a technology that is being presented as inevitable. And it’s essential to demand more environmental transparency from companies that are reshaping the economy to achieve their goals. It’s also just as important to double-check the statistics along the way.
This is an edition of Steven Levy’s Backchannel newsletter. Read previous newsletters here.