US data centers used an estimated 17 billion gallons of water in 2023, roughly three times their consumption in 2014, according to a Congressional Research Service report published by the Library of Congress. The estimate covers a data-center buildout that was accelerating as demand for generative AI took off.

The number is not a complete national measurement. The federal government does not systematically track water consumption across the data-center industry, and much of the available information is collected by state and local agencies. Some municipal agreements also restrict disclosure of the rates and volumes charged to individual operators.

That makes the 17-billion-gallon figure an estimate rather than a facility-by-facility accounting. It also covers more than the water piped directly into server campuses.

Electricity is the larger water burden

The CRS distinguishes between direct water use, such as cooling equipment at a data center, and indirect water use associated with generating the electricity that powers it. Direct consumption represents a relatively small share of overall US water use—about 2%—but the report says indirect consumption through power generation accounts for more than 80% of a data center’s total water footprint.

AI systems are pushing operators toward denser deployments of specialized hardware. New AI halls generally consume more electricity than older server infrastructure, increasing their cooling requirements and the water associated with power production. The CRS data ends in 2023, the year the AI infrastructure boom began in earnest, so the report says consumption may have risen further since then.

The numbers also complicate claims that a data center is “waterless” simply because its building avoids evaporative cooling. Replacing on-site water consumption with a more power-intensive cooling design can shift part of the burden to the generation of that electricity rather than eliminate it.

Cooling systems trade water for power

Data centers can use several cooling approaches, and the choice depends on the facility’s size, location, climate and thermal requirements. Liquid cooling can be more efficient than air cooling, particularly for high-density equipment, but the CRS notes that it may not be cost-effective for smaller campuses.

Evaporation-based systems typically use less energy than air-cooled chillers and other waterless systems. Their trade-off is substantial water consumption: evaporation removes heat efficiently, but the consumed water must be continually replaced. Air-cooled systems reduce that direct demand while generally requiring more power.

Operators may source water from municipal drinking-water systems, treated wastewater, surface water or groundwater. Municipal systems supplied 97% of the water used by US data centers in the CRS estimate. That reliance puts data-center expansion in direct competition with other municipal and industrial demands, especially where new campuses are proposed in areas with constrained supplies.

The issue has reached litigation. A proposed 330 MW California data center is seeking 287 million gallons of water from fallowed farmland, a dispute we reported on in the California farm-water lawsuit. The CRS report does not resolve how individual projects should be allocated water; it instead identifies the lack of consistent data as a barrier to evaluating those decisions.

Federal data remains fragmented

Water provision is primarily a state-level responsibility, and reporting requirements differ by jurisdiction. Utilities may report customer usage to state or local agencies when required, but there is no federal assessment that consistently covers every US data center.

A federally supported effort led by researchers at Virginia Polytechnic Institute and State University has produced the United States Water Withdrawals Database. It compiles reported and estimated withdrawals across 42 states, including public-supply, industrial and commercial usage. But its coverage depends on the reporting obligations in each state, so it does not provide a uniform national registry of data-center consumption.

The CRS also identifies water-service agreements as a source of opacity. Contracts between operators and municipalities can define usage charges and contain provisions limiting the public release of those details. Without facility-level figures, communities may have difficulty comparing a proposed campus’s water needs with its promised economic or infrastructure benefits.

Legislation is moving slowly

Members of Congress have introduced bills that would improve water-use reporting, encourage water reuse and address other environmental effects of data-center construction. Most remain at the introduction stage, the CRS says.

The campaign site Data Center Water Leaks counts water-related data-center legislation in 36 of 52 state legislatures. It says California and Arkansas vetoed local bills, while 20 federal bills—14 in the House and six in the Senate—were active; only three had progressed beyond the referral stage.

The reporting picture is more advanced in the European Union. Recent EU legislation requires data-center operators to report annual freshwater consumption along with other facility metrics. Industry group CISPE has warned that strict water rules could push operators to place infrastructure outside the EU, and the European Commission has faced lobbying over a proposed environmental rating system covering energy and water efficiency.

For US operators, the issue is less a single mandated efficiency threshold than the absence of a consistent baseline. A campus can draw water from a municipal system while shifting additional consumption to its electricity supplier, and public records may show neither the full direct volume nor the indirect cost. Until those figures are reported under common rules, the industry’s water footprint will remain a national estimate assembled from incomplete state-level data.