AI data centers use 12 times more water than tech giants disclose

The water consumed by AI data centers is as much as 12 times higher than what major tech companies report, because most firms exclude the water used to generate the electricity that powers them.

A 2024 analysis by Lawrence Berkeley National Laboratory found that indirect water consumption for US data centers has historically been about 12 times as great as the amount they directly consume. Among the largest technology companies, only Meta tallies water used at the power stations that feed its data centers, in addition to the water used on-site.

"The lack of transparency and widespread use of NDAs by many data-center builders has only drawn more suspicion and distrust," said Alex de Vries-Gao, a researcher at VU Amsterdam who published a paper earlier this year quantifying Google's indirect water consumption at roughly three times its direct use. "In a lot of cases, whenever such a project is being launched, the only information you're getting is an extremely tiny part. You're just getting the tip of the iceberg."

Google's 2025 sustainability report said the company consumed 10.9 billion gallons of water, a 34% increase from 2024, almost all for data-center cooling. Meta reported 19 billion gallons of indirect water use in 2024 — more than 20 times its direct consumption. Amazon said its data centers use water seven times as efficiently as the industry average and is 75% of the way to its goal of replenishing one gallon for every one it draws, though that target excludes indirect use.

The scale of the buildout is staggering. Tech companies including Microsoft, Google and Amazon are spending an estimated $1 trillion on AI infrastructure this year and last. No law obligates them to report the full scope of their water consumption, both direct and indirect. And many new data centers are being built in precisely the regions least able to spare the resource.

Where the Water Goes

About two-thirds of new data center construction in the US is in water-stressed areas such as Phoenix, according to analyses by the Guardian and Bloomberg. A 2025 report from Ceres, a nonprofit sustainability advocacy organization, found that the total direct and indirect water demands of data centers in Phoenix amount to about 3% of the city's annual water use today. By 2031, that figure could exceed 20%, approaching the amount consumed by all of Phoenix's residential lawns and landscaping.

"Cheap land and cheap power put data centers in the high water-stress areas," said Matthew Pine, chief executive of Xylem, a US water-technology company that supplies utilities across the country. Water-hungry, fossil-fueled relics such as coal-fired power plants are now being kept running past their intended retirement dates to address AI demand, he added.

The problem extends beyond the desert Southwest. In Homer City, Pennsylvania, a data-center complex is being built on the site of a decommissioned coal-fired power plant, complete with a natural-gas power plant. Though the location will generate more energy, it is expected to draw about the same amount of water as before, including what is needed to cool its data centers, said Jonathan Burgess, director of the Pittsburgh Water Collaboratory at the University of Pittsburgh.

Closed-Loop Solutions Gain Traction

Nvidia recently unveiled a closed-loop cooling system that requires no additional water once filled, claiming to have solved the data-center water issue. The design eliminates direct water use while also reducing the total amount of energy required for cooling, said Kenneth Gillingham, a professor of environmental economics at Yale.

"It's totally accurate that, as we generally solve the water footprint at the data-center operational level, the water footprint for AI becomes potentially driven by electricity generation," said Josh Parker, head of sustainability at Nvidia.

Microsoft has committed to similar closed-loop technology, announcing in 2024 that all its new data centers would use it starting in 2027. But most existing data centers use evaporative cooling systems that are energy-efficient but water-hungry, according to the Lawrence Berkeley report. Retrofitting those could be prohibitively expensive.

The controversy has real financial consequences. Climate consulting firm Carbon Direct estimates that $170 billion of AI data-center capacity has been blocked, stalled or canceled since 2024, partly due to community opposition over resource use. For investors, the water issue introduces a new layer of regulatory and reputational risk for hyperscalers already spending heavily on infrastructure. Nvidia's closed-loop system, if widely adopted, could reduce direct water demand but does nothing to address the indirect consumption tied to power generation — the far larger share of the problem.

This article is for informational purposes only and does not constitute investment advice.