A report from the CRS, part of the Library of Congress, has quantified the sharp rise in water consumption by American datacenters over the past decade, attributing much of the increase to the surge in AI infrastructure following the release of ChatGPT in late 2022. The CRS estimates that direct water use—primarily for cooling servers—reached 17 billion gallons in 2023, though this represents only about 2% of total U.S. water consumption.

The majority of a datacenter's water footprint, over 80%, comes indirectly from electricity generation, which often relies on water-intensive cooling processes at power plants. The report notes that new datacenters housing specialized AI hardware consume significantly more power and require more cooling than older facilities, driving up both direct and indirect water use.

Water usage varies by cooling technology. Evaporation-based systems are more energy-efficient but consume more water, while air-cooled chillers use less water but more electricity. Operators often choose systems based on local climate and cost. Most datacenters draw water from municipal supplies, accounting for 97% of direct water use, though some utilize treated wastewater or direct surface water.

The CRS points to a lack of systematic federal data collection on datacenter water use, as water management is primarily a state-level responsibility. Water usage data is often reported to state or local agencies, but the federal government has not conducted a comprehensive assessment of the industry's consumption. Additionally, water service agreements between municipalities and operators may include confidentiality clauses limiting disclosure.

This regulatory gap has led to legal challenges. Last month, a water resources scientist filed a lawsuit against Amazon Web Services, alleging the company made false claims about its water use in Northern Virginia. The scientist obtained billing records through Freedom of Information Act requests to local water utilities.

Efforts to improve data transparency include the United States Water Withdrawals Database, developed by researchers at Virginia Tech, which compiles withdrawal data for 42 states, covering public supply and industrial use. However, the CRS notes that sector-specific datacenter water use remains poorly understood at the federal level.

As AI demand continues to grow, the report warns that water consumption by datacenters is likely to increase further, raising questions about sustainability and the need for updated regulations.