How much water does AI really use?
Estimates for AI's water consumption vary widely due to different measurement methods, including whether water for cooling data centers and generating electricity is included. While data centers consume a significant amount of water, overall U.S. water demands for agriculture, lawns, and daily life are substantially higher. The impact on water supplies and electricity grids depends heavily on the data center's location and local infrastructure.
The amount of water used by Artificial Intelligence (AI) queries is a subject of widely varying estimates, ranging from a few drops to a half-liter bottle. This discrepancy arises because different calculations include or exclude the water used for cooling data centers, generating the electricity that powers them, and the rapid pace of technological change. Researchers from Lawrence Berkeley National Laboratory estimate U.S. data centers used approximately 228 billion gallons of water in 2023, with 211 billion gallons tied to electricity generation alone. This figure is projected to rise significantly by 2028.
However, AI accounts for only 15-20% of this demand, with the rest supporting other digital activities. When placed in context, data center water consumption, even including AI's share, is considerably less than that used for agricultural irrigation (particularly for animal feed), watering lawns, or other daily human activities. For instance, U.S. crop production used an estimated 183 trillion gallons in 2019, with fodder crops alone consuming 5.5 trillion gallons of irrigation water annually. Economist David Zetland suggests that comparing different water-intensive activities for a 'villain' is misleading; the core problem is that water is undervalued and its price does not reflect sustainable management costs.
The location of AI data centers also plays a critical role in their environmental impact due to variations in local electricity grids, water resources, and climate. While states like Texas, Nebraska, South Dakota, Louisiana, and Idaho are identified as favorable for future development due to access to wind and solar power, they still face challenges in expanding electric grid capacity. Beyond water, electricity consumption is a major concern, with data centers projected to reach 12% of U.S. electricity consumption by 2028. This growth contributes to grid strain, higher utility bills, and increased pollution, with potential for significant health impacts such as asthma and premature deaths, according to Professor Shaolei Ren of the University of California, Riverside. Concerns are highly localized in areas like northern Virginia, a major data center hub, regarding impacts on water wells, air, noise, and light pollution.