Water and the AI Infrastructure Boom

Data centers are a critical chokepoint in the competition to build the computing infrastructure required for frontier AI, cloud services, and other emerging technologies. Their development depends not only on advanced chips and abundant electricity but also on sufficient water for cooling and power generation. Given AI’s growing importance to national security and economic competitiveness, factors that render data centers unsustainable, unduly burdensome for host communities, or politically difficult to build pose long-term strategic risks. The lack of transparency around data centers’ water usage does all three.

Water remains one of the most misunderstood inputs along the AI value chain. High-profile estimates of data centers’ aggregate water consumption have heightened concerns about the environmental impact of data centers and, in some cases, contributed to political opposition to their expansion. These tensions are especially acute in water-stressed regions, including Tucson, Arizona, and Aragón, Spain, where community groups have fought against proposed projects due to concerns about their effect on local water demand.

Technical analysts warn that widely cited statistics can risk overstating water consumption while obscuring the conditions under which data centers endanger water sustainability. That is because the pressure a data center places on local water supplies varies by site. Differences in climate, cooling technologies, energy sources, water availability, and other factors all play a role. Assessing them, however, requires facility-level information that is often difficult for policymakers and local communities to obtain. The resulting uncertainty complicates sustainability assessments and fuels resistance to new AI infrastructure.

A more informed policy discussion would consider how data centers use water and seek to overcome barriers to accessing that information. Stronger reporting policies, and technologies such as smart metering, can provide both.

Facility-Level Transparency Matters

Data centers use cooling technologies that vary significantly by water demand. Traditional evaporative cooling generally consumes more water but less electricity, while newer approaches such as closed-loop and immersion cooling can substantially reduce water consumption. Climate and seasonal weather patterns also matter, as cooling data centers is often more water intensive in hot climates than in cooler ones. That is in part because data centers also indirectly consume water from the electricity generation used to power the centers. As a result, two facilities with similar computing capacity may require vastly different amounts of water.

Findings must also be put into context. US data centers consumed an estimated total of 228 billion gallons of water in 2023, far less than the 2.9 trillion gallons that Americans used for watering their lawns that year. That is not to say that data center water usage is trivial, but it makes the issue hyper-local. A data center that consumes a modest amount of water on a national scale can severely undermine local water sustainability if the facility relies on water-stressed aquifers, rivers, and watersheds.

Accordingly, policymakers and local communities need facility-level data if they are to make informed decisions about data center projects. They must be able to determine if a proposed facility is suited to local conditions.

Murky Waters

Data transparency is, unfortunately, lacking in Europe and the United States because governments on both sides of the Atlantic generally do not mandate facility-level reporting. In addition, existing reporting is not standardized, and implementing reporting requirements must overcome hurdles.

In the United States, federal agencies face legal and procedural constraints when collecting new data from private entities. Requirements for defining, collecting, and standardizing data vary widely across states and localities. Responsibility is fragmented across federal, state, and local agencies, and among public and private utilities, while different water sources may be governedby distinct legal and regulatory regimes. The entities involved also have different funding, staffing, and technical capacity to implement reporting requirements consistently. This non-standardized landscape complicates aggregating data, comparing usage across jurisdictions, anddeveloping a complete picture of an individual facility’s impact on local water supply.Authorities subsequently often rely on voluntary disclosures.

In the EU, reporting requirements are more formalized, but legislation enacted in 2024 includes a provision that classifies all facility-level information on data centers as confidential for competitiveness reasons. (Multiple legal experts have noted that the provision contradicts the bloc’s commitments under the Aarhus Convention, the leading international agreement on environmental democracy.)

Technical constraints on and limited incentives for data center operators and utilities to disclose water use accurately and consistently are additional challenges. Utilities operating older infrastructure may lack the necessary metering systems. Wider adoption of smart meters could improve the precision of water-use data and improve water conservation by, for example,identifying previously undetected leaks or using data from remote sensing satellites to track the health of watersheds.

Data center operators in fact have incentives to resist disclosing water use. They argue that facility-level data constitutes sensitive commercial information that competitors could use to infer operational details. Reporting requirements may impose little direct financial cost relative to profits, but they still create administrative burdens without generating revenue. Operators may also fear that headline figures will be taken out of context and fuel opposition to data center construction. These concerns explain why many companies encourage national governments to classify such data as proprietary and why they ask local governments and utilities to sign nondisclosure agreements.

Recommendations

Make transparency a priority. Governments should require standardized, facility-level reporting of data center water use. Proposed data centers should provide credible estimates of their anticipated water demand, which policymakers can compare with reliable data on local water availability and infrastructure capacity before approving a project. Existing data centers should report their actual water use over time, enabling operators and public authorities to identify and address unsustainable levels of water consumption. Greater transparency at both stages would help policymakers evaluate proposed data centers in relation to their likely impact on local water supplies while allaying legitimate public concerns about sustainability.

Make reporting mandatory. Existing voluntary disclosures are incomplete and may disadvantage companies that report more than their competitors. Uniform requirements would create a level playing field. Data center operators that stand to make significant profits should bear the costs of facility-level compliance, while public funding should help utilities and government agencies collect complementary data on the capacity of local water supplies.

Standardize reporting across jurisdictions. Governments should establish common definitions, reporting thresholds, and data formats. In federal systems such as that in the United States, a single mandate may not be practical, but the national government can provide a model framework for state and local adoption.

Incentivize use of new technologies. Governments and utilities should expand the use of smart meters and other technologies that enable accurate facility-level reporting. Public funding should help smaller and under-resourced utilities modernize aging infrastructure. Remote sensing and other monitoring tools can supplement metering, identify leaks, and improve understanding of watershed conditions. Private companies should consider pursuing public-private partnerships to address societal concerns about energy and water sustainability by investing in upgraded regional water and energy infrastructure.

Improve measuring indirect water consumption. Drawing inspiration from the “Scope 2” framework for identifying indirect drivers of greenhouse gas emissions, governments, industry, and philanthropies should support research on standardized methods for measuring water used to generate electricity consumed by data centers. Once developed, these methods should be incorporated into reporting frameworks and used to identify grids that depend heavily on water-intensive generation. In the meantime, governments and utilities should prioritize less water-intensive energy sources, such as wind and solar, especially in water-stressed regions.

The views expressed herein are those solely of the author(s). GMF as an institution does not take positions.