The Energy and Water Footprint of AI
This article discusses the energy and water consumption of AI-driven data centers, highlighting advancements in cooling technologies that significantly reduce water usage, potentially to zero. It analyzes global and US electricity demand for data centers, noting that while demand is projected to double, computational efficiency is improving even faster. The piece concludes by framing data centers as central to an energy transition, urging California to adopt 'smart regulations' to foster the industry.
The rapid expansion of artificial intelligence (AI) has led to a surge in data center construction, raising concerns about their environmental impact, particularly concerning energy and water consumption. However, a recent report by economist Richard Evans suggests that advanced cooling innovations are dramatically reducing data center water usage, with modern closed-loop systems consuming as little as 0.12 liters per kilowatt-hour, and some new facilities, like those by Microsoft and Oracle in states such as New Mexico, Michigan, Texas, and Wisconsin, achieving zero water evaporation.
The article also delves into the electricity footprint of data centers, noting the complexity of accurate estimation. While the International Energy Agency (IEA) and Gartner project global data center electricity consumption to double by 2030, other sources like the Statistical Review of Global Energy (using S&P data) reported 788 TWh worldwide in 2025, with 313 TWh from the U.S. This constitutes 2.4% of global and 6.6% of U.S. electricity production, a share considered less onerous in the U.S. due to higher per capita electricity availability.
Despite the significant energy demands, the article emphasizes that computational efficiency (FLOPs per TWh) is projected to improve by 34% annually until 2030, meaning data center capacity could increase 8.6 times by then, even with a doubling of electrical input. Edward Ring, director of water and energy policy for the California Policy Center, argues that data centers are vital for the broader energy transition and automation. He advocates for California to implement 'smart regulations' to enable the industry to thrive within the state, rather than driving it to other regions.