AI’s water use has become a flashpoint across the American Southwest, where protest signs reading “Protect our water,” “Water for people not AI,” and “Don’t mess with our water” have appeared from Texas to New Mexico to Arizona. Residents concerned about water consumption, energy demand, and pollution are increasingly opposing the rapid buildout of data centers as tech companies race to expand capacity for artificial intelligence.
Many data centers rely at least partly on water to draw heat away from air or surfaces, cooling equipment as the water evaporates, much the way sweating cools the human body. Processors inside these facilities can reach internal temperatures as high as 176° Fahrenheit, for the same reason an overworked laptop heats up.
Putting the numbers in context
While data centers clearly consume large amounts of energy, their water use is harder to pin down. Claims circulating online range from articles asserting that chatbots use 500 milliliters of water per query to commentators dismissing the issue entirely. Water data disclosed by large tech companies is often incomplete and inconsistent.
Experts emphasize that overall data center water consumption is small compared with industries such as agriculture and certain kinds of manufacturing. Researchers estimated that data center cooling systems used 66 billion liters of water in 2023, less than 1 percent of total US consumption.
That figure could rise sharply with the rapid expansion of AI data centers. The impact may be minor in water-rich regions but could add significant strain in drought-affected areas like New Mexico and Arizona.
Efficiency gains and better siting
Recent advances in AI have already lowered water demand for many facilities. Shaolei Ren, an electrical and computer engineer at the University of California, Riverside, notes that his 2024 estimate—based on GPT-4, an earlier model behind ChatGPT—that drafting a short email would use 500 milliliters of water is now outdated because AI models have grown more efficient. That study was the origin of the half-liter-per-query figure. In 2025, Google estimated that a median-length query with its Gemini chatbot uses roughly five drops of water.
Even tiny amounts add up as AI use grows. A group led by Fengqi You, an energy systems expert at Cornell University, recently projected that by 2030, US data centers could consume between 731 billion and 1,125 billion liters annually. The higher figure is roughly equivalent to New York City’s yearly drinking water supply.
Experts anticipate that water use for cooling AI data centers, as well as for generating fossil fuel-powered electricity to run them, will climb as the number of facilities increases. A 2025 study modeled future US water consumption under different scenarios.
Paths toward lower water use
Engineers say there are numerous measures already being deployed to ease pressure on local water supplies. “There are many good ideas out there,” You said. Combined with efforts to replenish and restore natural water resources, he added, “there’s a good chance that, eventually, it could be net-zero water for the on-site cooling.” Options under consideration include shifting toward more renewable energy and choosing sites that reduce dependence on stressed local water sources.
Researchers estimate that data center cooling systems used 66 billion liters of water in the United States in 2023.
Source
Image: arstechnica.com
