US Datacenters Triple Water Use by 2023
US datacenters used about 17 billion gallons of water in 2023, roughly triple their 2014 use, according to a Congressional Research Service estimate reported by The Register on August 24. The CRS found that the federal government does not systematically assess the industry's water consumption, while the reported figures predate much of the recent AI infrastructure construction surge.
US datacenters used approximately 17 billion gallons of water in 2023, about three times their consumption in 2014, according to a US Congressional Research Service (CRS) estimate reported by The Register. The increase was measured before much of the current wave of AI-focused data center construction, which surged after demand for generative AI and large language models expanded from late 2022.
The CRS report examines how data centers consume water, options for reducing use, and the federal role in oversight. The Register reports that specialized AI hardware generally consumes more power than older infrastructure, increasing cooling requirements.
A fragmented measurement picture
The CRS found that the federal government does not systematically assess all data center water use. Water-use data is often collected by state-level bodies, according to The Register, leaving no single federal accounting system for the sector.
The report distinguishes between direct, on-site water used for thermal management and indirect water use associated with electricity generation. Direct data center water use represents a relatively small share of total US water consumption, estimated at about 2 percent, The Register reported.
That accounting boundary matters for infrastructure analysis. Comparable facility assessments can reach substantially different results depending on whether they count only cooling-water withdrawals at a campus or also water consumed in producing the electricity that powers servers and cooling equipment.
Cooling choices carry efficiency tradeoffs
Data centers use different cooling approaches based on facility size, location, and IT thermal requirements, according to The Register's account of the CRS report. It reports that liquid cooling can be more efficient than air cooling, although it may not be cost-effective for smaller campuses. The article also notes a tradeoff between water and power consumption: evaporation-based systems are generally more energy-efficient than air-cooled alternatives but consume water through evaporation.
For ML infrastructure teams, those tradeoffs are becoming more relevant as accelerator-heavy clusters raise rack power densities. Across comparable deployments, capacity planning increasingly needs to account for electrical supply, thermal design, local water availability, and the reporting methodology used to measure each factor.
The Lincoln Institute's 2025 overview of data center development similarly identifies water, electricity, and land as linked local constraints during the AI buildout. The CRS estimate does not establish the sector's current water footprint after 2023, and the lack of standardized national reporting limits direct comparison across operators and regions.
Key Points
- 1CRS estimates US datacenters used about 17 billion gallons in 2023, three times 2014 consumption, establishing a pre-AI-boom baseline.
- 2The federal government does not systematically assess industry water use, CRS found, leaving state-level collection without a unified national operational picture.
- 3AI hardware's higher power density increases cooling demand; across comparable facilities, water and electricity efficiency choices create engineering tradeoffs.
Scoring Rationale
The CRS estimate provides a consequential baseline for evaluating the environmental resource demands of US data center growth. It is particularly relevant to practitioners working on AI infrastructure, cooling, capacity planning, and sustainability measurement, although it does not introduce a new model, tool, or policy mandate.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
