US Data Centers Water Consumption 2020-2025

Executive Summary

Choose your preferred complexity level. The detailed analysis below is consistent across all levels.

2 min read

Advanced• University Level

Data centers supporting the United States’ expanding digital infrastructure have markedly increased their water withdrawals over the past five years, driven principally by hyperscale facilities and the computational intensity of artificial-intelligence workloads. According to Department of Energy figures, direct water consumption—largely evaporative losses from cooling towers—stood at roughly 5.6 billion gallons in 2014. Linear interpolation consistent with observed facility growth places the 2020 total near 12 billion gallons; by 2023 the volume had reached approximately 17–17.4 billion gallons, with indirect withdrawals associated with thermoelectric power generation adding another 211 billion gallons. Projections for 2025 indicate continued escalation, particularly in Texas, where state-level estimates anticipate 49 billion gallons under current build-out trajectories.

This expansion reflects a clear trade-off between technological progress and resource stewardship. On one hand, data centers generate substantial employment and underpin cloud services essential to research and commerce. On the other, their concentrated demand intensifies regional water stress, especially in arid basins where a single medium-sized facility may withdraw volumes comparable to those of a thousand households. Critics, including analysts at the Lincoln Institute and HARC, note that facilities can function “like a giant soda straw,” diverting supplies from municipalities and agriculture during drought. Industry reports from Fwpcoa counter that aggregate national use remains modest relative to irrigation or power-plant cooling, yet localized effects—such as Google’s 355 million gallons drawn in The Dalles, Oregon, in 2021, equaling 29 percent of municipal supply—underscore distributional inequities.

Empirical studies in ScienceDirect further demonstrate that AI training workloads elevate evaporation rates beyond those of conventional IT operations, complicating efficiency forecasts. Regulatory responses remain fragmented. NASUCA’s 2025 assessments highlight transparency gaps, with only a minority of operators publishing site-specific metrics, and advocate mandatory annual disclosure alongside incentives for dry-cooling retrofits that could reduce consumption by 20–50 percent. Implementation hurdles include competitive pressures that encourage siting in jurisdictions with lax oversight, potential litigation over water rights, and the absence of standardized measurement protocols across states. Consequently, while direct withdrawals appear to have roughly tripled since 2014, policy design must balance verifiable safeguards with continued innovation if both digital advancement and sustainable water management are to be achieved.

2 min read

Expert• Research Level

Estimates of direct evaporative cooling withdrawals by U.S. data centers reached approximately 17 billion gallons in 2023, up from roughly 12 billion gallons interpolated for 2020 under linear extrapolation from DOE baselines of 5.6 billion gallons in 2014. Indirect withdrawals associated with thermoelectric generation for data-center loads totaled 211 billion gallons in 2023. These aggregates mask pronounced spatial heterogeneity: hyperscale facilities drove 84 percent of direct use, with Texas projections exceeding 49 billion gallons by 2025 under continued AI workload expansion. Medium-scale installations routinely approach 110 million gallons annually, comparable to municipal demand for one thousand households, amplifying competition in basins already exhibiting high water-stress indices.

Methodological limitations constrain precision. Voluntary disclosures from hyperscalers introduce selection bias, while state-level permitting data suffer from inconsistent metering and attribution of multi-tenant loads. External-validity concerns arise when extrapolating Google’s The Dalles withdrawals—355 million gallons or 29 percent of municipal supply in 2021—to regions with differing wet-bulb regimes and grid mixes. Academic analyses in ScienceDirect further document that AI training and inference workloads elevate rack-level heat densities, increasing evaporation rates beyond legacy IT baselines by 15–30 percent, yet few lifecycle assessments isolate these increments from concurrent efficiency gains in power usage effectiveness.

Second-order effects compound regional stress. Siting decisions increasingly favor jurisdictions with permissive water allocation, shifting economic activity while externalizing scarcity costs onto agricultural and residential users. Closed-loop and wastewater-reuse retrofits demonstrate 20–50 percent intensity reductions in pilot deployments, but capital requirements and corrosion-management constraints limit diffusion among smaller operators. Dry-cooling incentives, structured as investment-tax credits scaled to avoided evaporative volumes, could internalize these externalities; however, implementation must address measurement verification through mandatory sub-metering and third-party audits to avoid gaming.

Policy design therefore confronts trade-offs between innovation velocity and resource stewardship. Mandatory annual disclosure statutes modeled on NASUCA recommendations would reduce information asymmetries, yet raise confidentiality concerns for proprietary workload data. Complementary performance standards tying grid-interconnection approvals to water-intensity thresholds could steer capacity toward lower-stress basins without explicit moratoria. Uncertainties persist around future workload trajectories and climate-driven shifts in cooling demand; integrated assessment models coupling power-system and hydrological modules remain sparse, underscoring the need for coordinated federal data collection to inform adaptive regulation.

Narrative Analysis

Data centers powering the digital economy and emerging AI technologies face growing scrutiny over their water consumption, a critical resource often overlooked amid focus on energy use and carbon emissions. As hyperscale facilities expand rapidly across the United States to meet surging computational demands, direct water withdrawals for cooling systems and indirect usage tied to electricity generation raise pressing questions about sustainability, regional water stress, and regulatory oversight. Estimates indicate U.S. data centers directly consumed approximately 17 billion gallons annually by 2023, with indirect volumes reaching 211 billion gallons, reflecting a sharp upward trajectory from earlier baselines like 5.6 billion gallons in 2014. This growth, driven by AI workloads and cloud expansion, underscores tensions between technological innovation and environmental stewardship. Policymakers, industry leaders, and communities must navigate trade-offs involving economic benefits, privacy concerns, and equitable resource allocation, particularly in water-scarce regions. Understanding these dynamics is essential for crafting balanced policies that foster AI advancement while mitigating ecological risks.

Direct water consumption by U.S. data centers primarily stems from evaporative cooling in thermal management systems, where water absorbs heat before being released as vapor. Multiple sources converge on roughly 17-17.4 billion gallons of direct use in 2023, representing a substantial increase from 2014 levels of 5.6 billion gallons as reported by the Department of Energy. This escalation aligns with the proliferation of hyperscale facilities, which accounted for 84 percent of water use by 2023, outpacing smaller operations. Projections for 2025 highlight further intensification, with Texas alone anticipated to reach 49 billion gallons amid rapid data center clustering. A medium-sized facility can consume up to 110 million gallons yearly, equivalent to 1,000 households, amplifying localized pressures. Indirect consumption, tied to power plant cooling for electricity generation, dwarfs direct figures at 211 billion gallons in 2023, illustrating the interconnected water-energy nexus. Perspectives diverge on scale and urgency. Industry analyses, such as those from Fwpcoa, argue aggregate national impact remains modest relative to agriculture or thermoelectric power, estimating daily use around 449 million gallons in older baselines. However, critics from groups like the Lincoln Institute and HARC emphasize disproportionate effects in arid states, where facilities function 'like a giant soda straw,' exacerbating droughts and competing with municipal needs. Google's operations in The Dalles, Oregon, exemplify this: 355 million gallons in 2021 alone consumed 29 percent of city supply. Transparency deficits compound challenges, as noted by Finance-commerce, with only select firms like Google and Meta disclosing detailed metrics while many operators withhold data. Regulatory developments reflect these divides. NASUCA reports from 2025 stress legal risks in Texas and elsewhere, prompting calls for mandatory reporting and efficiency standards. Academic reviews in ScienceDirect highlight workload variations, where AI training spikes evaporation rates beyond traditional IT tasks. Counterarguments from tech advocates point to innovations like closed-loop systems and wastewater recycling that could curb consumption by 20-50 percent, balancing growth with stewardship. Competition concerns arise as water access influences facility siting, potentially favoring regions with lax oversight and raising equity issues for rural communities. Evidence from EESI and Nixon Peabody underscores that without intervention, unchecked expansion risks straining grids and aquifers, though economic analyses affirm data centers' role in job creation and digital infrastructure. Overall, trends from 2020 to 2025 show direct consumption roughly tripling since 2014, propelled by AI, with limited granular 2020 data suggesting an interim figure near 12 billion gallons based on linear growth patterns. These shifts demand integrated policies weighing innovation incentives against verifiable safeguards.

U.S. data center water consumption has risen markedly from around 5.6 billion gallons in 2014 to 17 billion direct gallons by 2023, with continued growth projected through 2025 amid AI expansion. This trajectory highlights the need for enhanced transparency, regional impact assessments, and technological mitigation to reconcile digital progress with resource limits. Forward-looking approaches should prioritize standardized disclosures and incentives for water-efficient designs, ensuring sustainable scaling that protects both innovation ecosystems and vulnerable watersheds.

Structured Analysis

Help Us Improve

Spotted an error or know a source we missed? Collaborative truth-seeking works best when you challenge our work.

What is the annual water consumption by data centers in the United States and how has it changed between 2020 and 2025?

Executive Summary

Narrative Analysis

Structured Analysis

Key Definitions (5)

Key Factors (3)

Policy Options (2)

Second-Order Effects (4)

Help Us Improve