# AI Growth Increases Data Center Water Demand > Source: > Published: 2026-07-04 01:10:30+00:00 # AI Growth Increases Data Center Water Demand **U.S.** data centers expanded for AI workloads were consuming nearly **one trillion litres** of water a year by 2025, according to a **July 4, 2026** SpaceDaily analysis of investor and media estimates, with cooling systems that shed heat by evaporating water into the air the main driver. For AI infrastructure teams, that water demand is now an operational constraint alongside electricity, shaping site selection and cooling-system design, and it is compounded by weak transparency: water use is measured unevenly and often reported only at the company level rather than per site. A **2021** academic estimate had already put U.S. data-center water use at over 600 billion litres a year, so the AI boom intensified an existing issue rather than creating a new one. Rising compute density and larger AI workloads are turning water availability and cooling architecture into practical risk factors for operations and sustainability teams, not just electric-power planning, and weak disclosure makes that risk harder to manage than the equivalent electricity numbers. ### What happened SpaceDaily reported on July 4, 2026 that, by 2025, estimates drawn from investor and media reporting put North American data-center water use at approximately one trillion litres a year, with the United States accounting for the bulk of new AI-driven capacity. SpaceDaily notes the figure is not a single official meter reading: data-center water use is measured unevenly, disclosed inconsistently, and often masked by company-level reports that skip site-level detail. That estimate builds on a longer trend rather than a sudden jump; a 2021 peer-reviewed estimate by David Mytton in npj Clean Water put U.S. data-center water consumption at about 1.7 billion litres a day, or roughly 620 billion litres a year, at a time when fewer than a third of operators even measured their water use. Separately, the IEA's 2025 Energy and AI report found data centers used about 415 terawatts-hours of electricity in 2024, projected to more than double by 2030, and a 2026 preprint by Gianluca Guidi and colleagues estimated 403 U.S. hyperscale facilities consumed roughly 68 to 99 terawatt-hours between May 2024 and April 2025, with more than half of attributed generation coming from fossil fuels, a reminder that a facility's water footprint also depends on the water intensity of the electricity behind it. ### Technical context The driver is higher heat density per server rack from AI accelerators and training workloads, which increases the heat cooling systems must remove. Evaporative and open-loop cooling methods are electricity-efficient but scale water use directly, since water leaves the system as vapor rather than recirculating; closed-loop liquid cooling and reclaimed-water systems reduce direct consumption but do not eliminate the underlying tradeoff between cooling efficiency and water intensity. ### For practitioners Infrastructure and MLOps teams should treat cooling water as a resource to model during capacity planning and cost forecasting, not an externality. Companies operating in water-constrained regions face potential operational and permitting friction; where site-level water metrics are unavailable, teams should instrument cooling systems, track water use per kWh or per useful compute unit, and weigh closed-loop or low-water cooling options in total-cost-of-ownership comparisons alongside electricity. ### What to watch Regulatory scrutiny or local permitting limits for large water users in drought-prone jurisdictions are likely to tighten as data-center clusters concentrate in water-stressed regions. More granular, site-level disclosure from operators or third-party audits would materially improve on today's company-wide reporting. Adoption rates for low-water cooling technologies, such as closed-loop chillers versus higher water-intensity evaporative systems, are a useful signal of whether operators are responding to the tradeoff before regulators force the issue. ## Key Points - 1U.S. AI data centers were consuming nearly one trillion litres of water a year by 2025, per investor and media estimates cited by SpaceDaily. - 2Rising rack density from AI accelerators drives more evaporative cooling, while inconsistent site-level disclosure obscures true water intensity. - 3Infrastructure teams should model cooling water alongside electricity in capacity planning, since low-water cooling adoption is an operational differentiator. ## Scoring Rationale A well-corroborated synthesis of AI data centers' growing water footprint (SpaceDaily estimate plus the Mytton 2021 baseline, IEA electricity data, and a 2026 hyperscale preprint), with clear practitioner relevance for capacity planning and site selection, though it is a trend explainer rather than a discrete news event. ## Sources 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](/problems)