AI’s Hidden Bill: Vanishing Water

AI’s growth is set to double data centers’ water demand within a few years, right as climate stress tightens supplies.

Story Snapshot

United Nations-linked researchers project data-center water use could nearly double by 2030.
Government analysis warns AI could drive billions of cubic meters of global water withdrawals by 2027.
Two-thirds of new U.S. data centers are rising in water-scarce regions, magnifying risk.
Local governments are already pausing projects over water and power concerns.

Data centers are a fast-rising industrial water user

United Nations University researchers, summarized by Reuters, reported data centers used about 1.2 trillion gallons of water in a single year and could double both water and power use by 2030 as artificial intelligence drives expansion. A United Kingdom government-commissioned report adds scale to that trend, estimating that artificial intelligence tools could drive between 4.2 and 6.6 billion cubic meters of global water withdrawals by 2027, much of it tied to cooling and electricity generation. Those numbers frame the stakes that utilities and communities now face.

Industry siting choices raise the heat. Bloomberg reported that two-thirds of new U.S. data centers are being built in regions already short on water, with sector-wide water use projected to more than double by 2030. That strategy piles new demand onto stressed basins that also face hotter, drier years. Climate change tightens supply through drought and heat, while artificial intelligence boosts demand. This is not theory to residents living on watering limits or to utilities planning for the next multi-year dry spell.

Direct and indirect water use both matter

Lawrence Berkeley National Laboratory-based estimates, cited by the Environmental Law Institute, show direct U.S. data-center water use rose from about 21.2 billion liters in 2014 to 66 billion liters in 2023, while indirect use through electricity generation reached about 800 billion liters in 2023. The difference is key. Cooling towers and chillers draw water on-site. Power plants upstream draw far more to make the electricity that servers burn. Clear accounting of both is the only honest way to manage risk.

That accounting remains patchy. The Environmental Law Institute notes data often mixes direct and indirect use, which can blur attribution and invite spin. The Wall Street Journal reports no law compels technology companies to report full direct and indirect water use, which leaves the public relying on partial disclosures and estimates. Conservative common sense says what gets measured gets managed. Requiring transparent totals, by watershed, is the baseline for any serious plan.

Local backlash is shaping approvals

City halls are not waiting for national rules. CBS Chicago reported Naperville rejected a proposed data center after residents raised water and energy concerns, and neighboring Aurora set a 180-day moratorium while it drafted water and power standards for future projects. These moves show how fast public priorities shift when the utility bills and the aquifer maps show up. Jobs and tax base matter. So does a steady tap and a grid that does not buckle in a heat wave.

Developers have options if they plan ahead. Closed-loop cooling, recycled wastewater, and siting near non-potable sources can cut stress. Aligning operations with off-peak power and adding on-site thermal storage can also help. Yet these steps only work if companies share real numbers and pick sites that match local water budgets. Building massive new loads in dry corridors and expecting someone else to figure it out later is not stewardship. It is a gamble with other people’s wells.

Climate stress is a force multiplier, not a scapegoat

Climate change does not sign the construction permits. It does load the dice. Hotter summers drive higher electricity demand. Longer droughts tighten river and aquifer flows. When artificial intelligence accelerates data-center growth in those same basins, the margin for error shrinks. The United Kingdom executive summary warns that artificial intelligence’s water footprint spans cooling and electricity alike, which means heat and drought raise both sides of the ledger at once. Prudent planning prices in that compound risk.

Policy can move as fast as concrete. Require facility-level water and power disclosure by source watershed. Tie approvals to drought contingency plans that protect households first. Reward designs that use recycled water and avoid freshwater draws. Favor siting in cooler, wetter regions with robust grids. None of this slows innovation; it filters out bad bets. When facts show demand will surge, planning should not lag. Communities deserve growth that keeps faith with basic resources.