The debate over data centers’ water use tends to focus on how much is needed to cool servers on-site. But that fails to capture the largest and least visible part of their overall use: the water used at the power plants fueling data centers’ rising energy needs.
Only one hyperscaler, Meta, reports its indirect water consumption embedded in purchased electricity. That figure was more than 20 times higher than the amount of water directly consumed by its data centers in 2024 — and grew every year since 2021, according to Meta’s most recent sustainability report.
Meanwhile, Amazon told Latitude Media that it tracks indirect water consumption from power generation, but hasn’t publicly disclosed the data yet because there isn’t an industry-wide reporting standard. That leads to widely variable figures, depending on what assumptions are used, so the company is working with experts to develop a more consistent framework. (Neither Google nor Microsoft said whether they track indirect water use.)
A new analysis by the sustainable investment group Ceres attempts to help fill the gap. Ceres modeled data centers’ indirect water use across seven states — Arizona, California, Georgia, Illinois, Ohio, Texas, and Virginia — which combined are home to about half the data centers in the U.S. The report estimates that these data centers collectively used about 3.4 trillion gallons of freshwater in 2024 for electricity, or about or 12 times the combined total of Los Angeles, Phoenix, and Washington, D.C.
That figure could rise to between 4 trillion and 8 trillion gallons by 2030, based on projections of data centers’ electricity demands.
About two-thirds of the power plants evaluated were in areas with medium-high to extremely high water stress, particularly those in Arizona and Illinois. Arizona is in a long-term drought, while Illinois relies a lot on nuclear power located in water-stressed regions of the Mississippi-Missouri river basin.
Nuclear and coal power need the most water for cooling, while gas plants need less. Renewables including solar and wind require almost none.
“Some of the states we looked at aren’t the first that come to mind when we think about water availability problems,” Kirsten James, senior director of Ceres’ water program, told Latitude Media in an interview. “When you tease out the data, you see that a lot of the electricity is being generated in areas exposed to water stress and drought. So this may be a concern in a broader geography than first thought.”
James said data center water use is becoming a larger focus for Ceres’ network of investors. However, because barely any companies — data center developers, energy utilities and other power producers — report on these risks, it is difficult to assess them holistically.
Lawrence Berkeley National Laboratory estimated that in 2023, U.S. data centers directly consumed 17 billion gallons of water. The facilities’ indirect water footprint that year, LBNL found, was 12 times as big. By 2028, the researchers projected that the totals could double.
However, a lot has changed in the last three years. That study was published well before the AI boom was fully underway, and it also didn’t capture water used for behind-the-meter generation that today is picking up steam at projects across the country. Amazon, Google, Meta, and Microsoft all have plans to build gas plants, with Amazon earlier this month confirming plans for a 7.65 gigawatt project in West Texas, Cleanview reported.
The tech companies, in trying to mitigate local opposition to data centers over concerns like water and energy use, have pointed out that data centers use a lot less water than other industries like U.S. agriculture and lawn care. In Arizona, for example, alfalfa farmers alone use about 423 billion gallons of water every year while nationwide the U.S. uses 3.2 trillion gallons to water lawns, according to federal data.
But those big numbers can obscure the fact that water availability is a hyperlocal issue, James said.
“When you’re going into these areas that are already water stressed, and there’s a lot of competing needs, data centers need to be part of this discussion,” she said. “Even if their direct and indirect water use is a relatively small amount, we need to know what any incremental withdrawal could mean to ensure resilience in that basin.”
The hyperscalers’ promises
The Big Four hyperscalers have all pledged to become “water positive” across their operations — meaning they plan to replenish more water than they consume — and invest in public infrastructure. Executives from these companies have also said they evaluate local watersheds before deciding what cooling technology to deploy to limit the impact, while acknowledging the public demand for more transparency.
“We recognize the connection between energy and water, and that’s why we’ve invested so heavily in renewable energy,” an Amazon spokesperson said. “Over the last decade, the water intensity of U.S. power generation has dropped by roughly 20%, and renewable energy investments are contributing to that shift. We’re also focused on making our operations more efficient.”
Ben Townsend, Google’s global head of infrastructure and sustainability, said in a statement that the company focuses its reporting on direct water use “where we have operational control to drive immediate conservation — while systematically reducing our indirect water footprint through our carbon-free energy procurements.”
Townsend added that sourcing clean energy like wind and solar is one of the most effective ways to save water, and Google averaged 65% carbon-free energy across its offices and data centers globally.
A Microsoft spokesperson said indirect water impacts from electricity use is an emerging area of focus across the industry, and that the company is focused on measuring and managing the water impacts of its operators “with rigor and transparency.”
Meta didn’t return a request for comment.
Ceres’ method
James acknowledged that Ceres’ analysis wasn’t a precise science. But the study “at least directionally” shows the scope and scale of data centers’ indirect water use.
The group used primary data wherever it could, including on the generation at individual power plants across the seven states and whether they used freshwater, which is reported by the U.S. Energy Information Administration. However, of the more than 1,400 plants identified for Ceres’ study, only 194 reported water use — and that sliver of data isn’t independently verified. For the rest, Ceres used the World Resources Institute’s generation-specific water-intensity factors.
To estimate the water use attributed to data centers, Ceres used 2024 state-level estimates of their share of electricity use as well as projections for 2030. That is a limitation, as the Ceres study points out, because the electricity delivered to a specific data center reflects the generation mix of a balancing authority that may span multiple states.
James said both data center companies and power producers have a role to play, and encouraged more robust disclosure about water risks in both industries.
“Given the exponential growth of this industry, we really have to take this indirect water piece seriously,” she said.
