US datacenters tripled water use in a decade. AI makes it worse

America's datacenters drank 17 billion gallons of water in 2023, roughly triple what they consumed a decade earlier, according to a new Congressional Research Service report. That estimate, published through the Library of Congress, captures the industry's footprint just as the AI boom was taking off, meaning the real number today is almost certainly higher. The CRS points directly at the generative AI wave that began with ChatGPT's release in late 2022, which triggered a massive surge in datacenter construction and a wave of demand for large language models. Newer facilities packed with specialized AI hardware typically burn far more power and require more cooling than older infrastructure, and that cooling is where the water goes.

For context, the 17 billion gallon figure is still a relatively small slice of total US water consumption, roughly 2 percent when measured as direct on-site use. But the CRS found that indirect water consumption, the water used to generate electricity that powers the facilities, accounts for more than 80 percent of a datacenter's total water footprint. That distinction matters because it means the industry's true water impact is much larger than the meter at the facility gate, and it is largely invisible to current federal oversight. The CRS notes that the federal government does not systematically assess all water use by the industry, partly because water provision is a state-level responsibility and data is often collected by state bodies, if at all.

The report lands at a moment when water is becoming a board-level issue for operators, investors, and the communities hosting these facilities. The CRS highlights that municipal water systems supply 97 percent of the water used by US datacenters, and the terms of that supply are often locked inside water service agreements between operators and municipalities. Those agreements can include provisions that limit disclosure of water usage information, which helps explain why sector-specific data remains scarce. Water utilities may report usage to state or local agencies when required, but there is no comprehensive federal picture, and the CRS says the federal government has not conducted a systematic assessment of the industry's water use.

That data gap is already producing legal and political friction. The Register reported last month that a water resources scientist filed a lawsuit against Amazon Web Services, alleging the company published false and misleading statements about its water use in Northern Virginia. The evidence was drawn from billing and water consumption records obtained through Freedom of Information Act requests sent to water utilities in the region. The case illustrates how the absence of standardized reporting creates room for disputes, and how FOIA requests are becoming a tool for holding operators accountable when official data is thin.

At the policy level, the response is fragmented. The CRS lists a number of bills introduced in this Congress that would address datacenter water use through various mechanisms, from improving information availability to incentivizing water reuse, but most have not moved past introduction. The campaign site Data Center Water Leaks counts 36 of 52 state legislatures that have introduced some form of water legislation related to server farms, while California and Arkansas vetoed local bills. At the federal level, 20 bills are currently active, 14 in the House and 6 in the Senate, though only three have advanced beyond the referral stage. The picture is one of growing attention but slow action, leaving operators to navigate a patchwork of state and local rules.

Outside the US, the regulatory trajectory is different. Recent European Union legislation requires datacenter operators to explicitly report annual freshwater consumption along with other facility metrics. But the European Commission has faced pushback from the industry, with trade body CISPE warning last year that onerous water regulations could push operators to shift infrastructure outside the EU. More recently, the Commission was reported to be weakening its proposed environmental rating system for datacenter energy and water efficiency in response to industry lobbying. That dynamic, regulation pushing against industry resistance, is likely to play out in the US as well, especially as AI infrastructure investment continues to climb.

For executives, the strategic stakes are clear. Water is no longer just an operational input or a sustainability talking point; it is a siting constraint, a disclosure risk, and a potential source of legal liability. The CRS report notes that datacenter operators may draw water from municipal drinking water systems, treated wastewater, or direct withdrawals from surface water or groundwater, and the choice of cooling system involves trade-offs between water consumption and power usage. Evaporation-based cooling is generally more energy efficient than air-cooled chillers or waterless systems, but it uses far more water. Location and climate often determine which system makes sense, which means water strategy is inherently local.

Federal funding has supported some efforts to close the data gap, including the United States Water Withdrawals Database developed by researchers at Virginia Polytechnic Institute and State University, which compiles water withdrawal data and estimates for 42 states. But the CRS makes clear that the federal role remains limited and reactive. For CEOs and CFOs in the datacenter, cloud, and AI infrastructure space, the message is that water data is coming, whether through state legislation, FOIA-driven lawsuits, or EU-style reporting mandates. The companies that treat water as a strategic constraint today, rather than a compliance issue to be managed later, will have more options on siting, cooling, and community relations when the regulatory wave arrives.