Executive summary

As the expansion of data centres for artificial intelligence accelerates across the United States, the pressure on water, energy, and land resources is rising rapidly. This report focuses on the significant, yet often overlooked, demand for water used in electricity generation to power these facilities. Ceres identifies that while data centre operators typically focus on direct water use for cooling, the indirect water required for power generation has an even greater impact. This reliance creates material risks for companies, investors, and communities, particularly as power plants are increasingly located in regions facing water stress or drought conditions. The analysis covers seven states representing approximately 50% of U.S. data centre capacity: Virginia, Texas, California, Illinois, Georgia, Ohio, and Arizona.

Background

AI expansion is driving rapid growth in data centre capacity, which is expected to nearly double by 2030. Electricity use from these facilities is projected to reach between 9% and 17% of total U.S. demand by 2030, up from roughly 4% to 5% today. This surge has profound implications for water resources, as power plants utilize significant volumes of freshwater for cooling and generation. In 2023, U.S. data centres consumed approximately 176 TWh of electricity, carrying an estimated indirect water dependency of 211 billion gallons. By 2030, water used in electricity generation is projected to account for 72% of total U.S. data centre water consumption. Operational constraints are already appearing; water strains and drought conditions have forced some coal and nuclear plants to curtail output, and local opposition regarding water and grid capacity disrupted $130 billion in data centre projects in the first quarter of 2026 alone.

Study area: U.S. data centre hotspots

The report examines seven key states that serve as data centre hotspots: Virginia, Texas, California, Illinois, Georgia, Ohio, and Arizona. These states were selected based on their significant share of the national market and their varied energy generation mixes. The regions also face diverse levels of water stress and drought. The analysis assumes that data centres source their electricity from the state in which they are located, and it focuses on water withdrawals rather than consumption to capture the full volume of water an asset depends on to operate.

Data analysis

To understand water use in electricity production, the analysis used power plant-level data from 2024, identifying 3,907 active plants across the study area. Of these, 1,434 plants (37%) were identified as being dependent on freshwater for generation. The methodology involved aggregating reported water volume data and estimating values for plants without reported data using technology-specific water-intensity factors. The study also mapped plant coordinates to water stress and drought data to characterize exposure to regional vulnerabilities. Limitations of the analysis include the use of state-level generation data as an approximation and the reliance on modeled estimates where reported data was unavailable.

Key findings

The report highlights that power generation makes up the bulk of data centre water use. Data centres in the seven analysed states depend on roughly 3.4 trillion gallons of freshwater annually for electricity, which is approximately 12X the annual water use of Los Angeles, Phoenix, and Washington D.C. combined. Furthermore, 78% of the electricity in these states came from power plants that require water to operate. California withdrew more water for energy generation than any other state, totalling 37.9 trillion gallons, primarily due to its large hydroelectric fleet. In contrast, Ohio, Georgia, and Virginia generated over 90% of their electricity from water-intensive thermal power plants.

The research also found that 66% of the electricity generated using water came from areas facing medium-high to extremely high water stress. Approximately 44% of this electricity came from plants in areas that faced three or more months of drought. Arizona and Illinois showed the greatest exposure to these compounding risks. In Arizona, 75% of generation was exposed to both water stress and at least three months of moderate drought in the same year. Despite these risks, most power producers have not yet fully considered the potential water risks tied to rising data centre demand, and most data centre operators do not account for the water risks linked to the electricity they purchase.

Recommended actions

Data centre companies are advised to integrate both direct and indirect water use into risk assessments and disclose this information for each facility. They should evaluate how power purchase agreements and renewable energy investments affect their water risk profiles and collaborate with power producers to understand regional vulnerabilities. Power producers should disclose water withdrawals and consumption at the individual plant level and integrate water risk from data centre growth into resource planning and infrastructure decisions.

Policymakers are encouraged to require standardised reporting on direct and indirect water use and incentivise the deployment of low water-intensive renewable energy. They should ensure that siting and permitting decisions account for basin-level water stress. Finally, investors should encourage comprehensive disclosures from both data centre companies and power producers to better understand water-related risks for investment decision-making and engage with companies to ensure these risks are appropriately mitigated.