How much of a problem is AI’s water use?
Public protests are growing across the US against the rapid expansion of data centers, driven by concerns over water use, energy consumption, and pollution. While data centers' overall water use is currently small compared to other industries, experts predict a significant increase by 2030, particularly with the growth of AI. Tech companies and engineers are developing solutions like liquid cooling and strategic siting to mitigate environmental impact.
Across the United States, locals are increasingly protesting the rapid expansion of data centers, citing concerns over water use, energy consumption, and pollution, particularly as tech companies expand capacity for AI. Protests have been observed from Texas to New Mexico to Arizona, with signs declaring messages like "Water for people not AI.
While data centers consume significant energy, their water consumption is a more complex issue. Experts like Fengqi You of Cornell University and Shaolei Ren of the University of California, Riverside, acknowledge that overall data center water use (estimated at 66 billion liters in 2023) is currently less than 1% of the nation's total. However, this figure is expected to rise considerably with AI's rapid buildout, potentially reaching 731 billion to 1,125 billion liters annually by 2030, an amount comparable to New York City's annual drinking water supply. This surge could exacerbate water stress in drought-prone regions such as New Mexico and Arizona.
Engineers and companies are actively pursuing solutions to reduce this strain. Advances in AI technology have made models more water-efficient, reducing consumption per query. Furthermore, shifting to renewable energy sources like solar and wind can significantly cut the water used in electricity generation for data centers. Experts also advocate for strategic siting in water-rich areas with ample renewable energy, such as Montana or parts of Texas, rather than in water-stressed regions. Cooling technologies are evolving, with many tech companies, including Amazon, Microsoft, and Google, adopting more efficient liquid cooling systems. NVIDIA's Josh Parker notes that their latest processors can be cooled with warmer water, reducing the need for traditional water-based cooling.
Companies like Amazon Web Services, Google, and OpenAI have set ambitious sustainability goals, including becoming water positive by 2030 or replenishing more water than they consume. Eric Masanet of the University of California, Santa Barbara, emphasizes that the future of AI's water usage hinges on collective decisions by tech companies, engineers, and policymakers regarding data center location, energy infrastructure, climate considerations, and design.