Artificial intelligence has become the focus of growing environmental concerns amid warnings about its water and energy consumption. But according to a new report, the reality is far more nuanced, and the true environmental impact depends on what is actually being measured.
Claims about AI's water consumption vary dramatically. Google has said that a typical AI query uses about five drops of water, while OpenAI CEO Sam Altman has described a similar amount, estimating roughly one-fifteenth of a teaspoon per query.
The largest estimate is roughly 2,000 times greater than the smallest. According to CBS News, the conflicting figures are not necessarily false. Instead, they measure different aspects of AI's environmental footprint.
Some estimates account only for the water used to cool servers inside data centers. Others also include the enormous amount of water associated with generating the electricity required to power those facilities. In addition, many figures rely on older assumptions that do not reflect the rapid improvements in AI hardware and data center efficiency.
Researchers say there is no single number that accurately represents AI's water consumption because it varies based on several factors, including the AI model being used, where the data center is located, the local climate, and the energy sources supplying electricity. In some cases, water usage can differ by hundreds of times depending on these variables.
Scientists at the Lawrence Berkeley National Laboratory estimate that U.S. data centers consumed approximately 228 billion gallons of water in 2023. Of that total, around 17 billion gallons were used directly to cool servers, while roughly 211 billion gallons were indirectly consumed through electricity generation that powers those facilities.
Researchers project that total water consumption from U.S. data centers could rise to between 469 billion and 844 billion gallons annually by 2028 as AI demand continues to expand. However, AI itself is responsible for only a portion of that growth.
According to the International Energy Agency, artificial intelligence accounts for roughly 15% to 20% of data center electricity demand, while the majority supports other digital services such as video streaming, cloud computing, satellite data, internet searches, and social media platforms.
The CBS News analysis also places AI's water use into a broader national context, writing that "The portion of data centers' cooling water attributed to just AI processing is less than the water used for one year's worth of golf-course irrigation, vehicle washing, residential swimming pools and restaurant dishwashing, based on estimates derived by government and industry."
Agriculture remains by far the country's largest water consumer. Researchers estimate U.S. crop production required approximately 183 trillion gallons of water in 2019. Even when considering only irrigation water, agriculture consumes vastly more than AI infrastructure.
Food grown for livestock alone illustrates the scale. According to researchers at the University of Twente, fodder crops consume approximately 5.5 trillion gallons of irrigation water every year in the United States, dwarfing the estimated 228 billion gallons used by data centers.
Combined, nuts account for less than 1% of U.S. crop production but still consume an estimated 538 billion gallons of irrigation water, more than twice the total water associated with AI-powered data centers. Economist David Zetland told CBS News that the larger issue is that water remains significantly underpriced, encouraging excessive consumption across nearly every sector of the economy.
"The value of water is radically higher than the price we pay for it," Zetland said, arguing that the lack of sustainable pricing contributes to growing water shortages. Location also plays a major role in AI's environmental impact. CBS News found that many existing data centers are concentrated in water-stressed states such as California, Arizona, and New Mexico.
Researchers believe future facilities may be better suited to states including Texas, Nebraska, South Dakota, Louisiana, and Idaho because of greater access to wind and solar energy, which require relatively little water to generate electricity. However, those regions still face challenges expanding electric grid capacity fast enough to support rapid AI growth.
Water is only one piece of AI's environmental footprint. Data centers accounted for an estimated 4% of U.S. electricity consumption in 2023, with projections suggesting that share could reach 12% by 2028. Researchers also warn that increased electricity generation could worsen air pollution near power plants and data centers.
A study led by University of California, Riverside professor Shaolei Ren estimates AI-driven data center expansion could contribute to hundreds of thousands of asthma symptom cases and more than 1,000 premature deaths annually by 2028 because of increased pollution associated with electricity production.
