KNOWABLE: How much of a problem is AI’s water use?
Public protests are growing across the US due to the significant water and energy consumption of rapidly expanding AI data centers, especially in water-stressed regions like New Mexico and Arizona. While data center water use is currently a small fraction of national consumption, it is projected to rise considerably by 2030. Advances in cooling technologies, strategic siting, and shifts to renewable energy are proposed as solutions to mitigate this environmental impact and help companies meet sustainability goals.
Locals from Texas to New Mexico and Arizona are increasingly protesting the rapid expansion of data centers across the United States, citing concerns over water use, energy consumption, and pollution, as tech companies rush to expand AI capacity. Data centers often rely on water-based cooling systems to manage the extreme heat generated by processors, a necessity that intensifies with energy-hungry AI hardware.
While data center water consumption currently accounts for less than 1 percent of the nation's total, experts like Fengqi You of Cornell University and Shaolei Ren of the University of California, Riverside, warn that this figure could substantially increase with the rise of AI. You's group predicts U.S. data centers could consume up to 1,125 billion liters annually by 2030, roughly equivalent to New York City’s annual drinking water supply. This consumption is particularly problematic in drought-prone regions.
To address this, engineers are developing and implementing water-saving cooling technologies. Companies such as Amazon, Microsoft, and Google are adopting liquid cooling, a more efficient, closed-loop system for direct processor cooling. NVIDIA, a major processor designer, is developing chips that run at higher temperatures, reducing the need for water-based cooling. Additionally, experts recommend siting data centers in water-rich areas with ample renewable energy, like Montana or parts of Texas, rather than water-stressed regions. Companies like Amazon, Google, and OpenAI have also set ambitious water positive and replenishment goals, and some are exploring using waste heat for district heating. The future water footprint of AI data centers will largely depend on these technological advancements, strategic siting decisions, and policy choices, according to Eric Masanet of the University of California, Santa Barbara.