AI DATA CENTER WATER RISK

How much water do AI data centers use? Data center water risk, explained

How much water AI data centers consume, where it comes from, why siting in water-stressed regions matters, and how to assess a site’s long-term water risk.

By Argos Tellus · Updated October 1, 2026

KEY TAKEAWAYS

U.S. data centers consumed about 17 billion gallons (66 billion liters) of water directly in 2023, according to Lawrence Berkeley National Laboratory (LBNL).
LBNL projects direct water consumption by hyperscale data centers alone could reach 16–33 billion gallons a year by 2028 as AI workloads grow.
Data centers used about 4.4% of U.S. electricity in 2023 and could use 6.7–12% by 2028 — and the power plants behind that electricity consume water too.
About two-thirds of U.S. data centers built or in development since 2022 sit in areas of high water stress, per a Bloomberg analysis of WRI Aqueduct data.
The real risk is local: whether the aquifer or utility under one specific site can keep supplying water for the 20–30 year life of the facility. That is a groundwater question, not a corporate-average one.

How much water does a data center use?

Data centers use water in two ways. Direct (on-site) water is mostly cooling: evaporative cooling towers and adiabatic coolers reject heat by evaporating water, and the evaporated water is consumed — it does not return to the local supply. Indirect water is the water consumed to generate the electricity the facility draws from the grid, mainly at thermoelectric power plants.

At national scale, LBNL’s 2024 United States Data Center Energy Usage Report estimated direct water consumption of about 66 billion liters (roughly 17 billion gallons) in 2023, with hyperscale and colocation facilities responsible for most of it. LBNL’s earlier work put typical evaporatively cooled facilities at about 1.8 liters of water per kWh of site energy.

At facility scale the range is wide. The World Resources Institute (WRI) notes that a mid-size data center can use up to about 300,000 gallons a day, and large campuses up to several million gallons a day — comparable to the demand of a small city.

Metric	Figure	Source
U.S. data center direct water consumption, 2023	≈66 billion L (≈17 billion gal)	LBNL 2024
Projected hyperscale direct water use, 2028	16–33 billion gal/yr	LBNL 2024
Data center share of U.S. electricity, 2023	4.4% (176 TWh)	LBNL 2024 / DOE
Projected share of U.S. electricity, 2028	6.7–12% (325–580 TWh)	LBNL 2024 / DOE
Typical evaporative cooling water intensity	≈1.8 L per kWh of site energy	LBNL 2016

Why AI makes data center water use a bigger problem

AI training and inference run on dense GPU clusters that produce far more heat per rack than conventional servers. More heat means more cooling, and in hot or dry climates the cheapest way to reject that heat is usually evaporation. LBNL’s projections show data center electricity demand roughly doubling or tripling between 2023 and 2028, and water demand rising with it.

Closed-loop and direct-to-chip liquid cooling can cut on-site water use sharply, but they typically raise electricity use, which shifts water consumption upstream to power generation. There is no free option: every design trades on-site water, grid water, energy cost and capital cost against each other — see data center cooling and water.

Where the water comes from: utilities, surface water and groundwater

A data center gets water from a municipal utility, from surface water rights, from on-site wells, or from reclaimed wastewater. In much of the arid and semi-arid U.S. — including large parts of Texas, Arizona and the Great Plains — the utility’s own supply is groundwater, so a data center on city water is often still drawing on an aquifer.

That matters because aquifers respond slowly and locally. A wellfield pumping a few million gallons a day creates a cone of depression that can lower water levels in neighboring wells, reduce pump submergence, and, over years, draw down the saturated thickness that the whole area depends on. Many aquifers in the Southwest and High Plains are already in long-term groundwater depletion.

What “data center water risk” actually means

Corporate sustainability reports describe water at portfolio scale. Investors, lenders, utilities and neighbors care about the site. For a specific campus, water risk breaks into a handful of questions:

Physical availability — can the aquifer or source sustain the required pumping rate for the facility’s design life without unacceptable drawdown?
Competition and interference — what other wells, irrigators and municipal systems draw on the same aquifer, and how will their pumping interact with yours?
Regulatory and permit risk — is groundwater regulated by a local district (for example, a Texas groundwater conservation district), and could production limits tighten during drought?
Water quality — does the source need treatment (salinity, hardness, contaminants such as PFAS) before it can be used for cooling?
Reputational and community risk — data centers in water-stressed regions face growing local opposition when residents see their own wells decline.

How to assess water risk for a data center site

A defensible assessment starts from the aquifer, not from a national water-stress map. Water-stress indices such as WRI Aqueduct are useful screens, but they average over large basins and say nothing about the specific formation, saturated thickness and neighboring wells under a parcel.

The groundwater due-diligence checklist for data center sites walks through the inputs: aquifer identification, historical water-level trends, sustainable yield, interference with existing wells, permit limits, and a forward projection of water levels across the facility’s design life.

Frequently asked questions

How much water do U.S. data centers use?

Lawrence Berkeley National Laboratory estimated U.S. data centers consumed about 66 billion liters (about 17 billion gallons) of water directly in 2023, mostly for cooling. That excludes the water consumed to generate their electricity.

Does AI increase data center water use?

Yes. AI hardware is denser and hotter than conventional servers, and LBNL projects data center electricity use rising from 4.4% of U.S. electricity in 2023 to 6.7–12% by 2028. Hyperscale direct water use alone could reach 16–33 billion gallons a year by 2028.

Why do data centers use water?

Mainly for cooling. Evaporative cooling towers and adiabatic systems reject heat by evaporating water, which is consumed rather than returned. Water is also consumed indirectly at the power plants that supply the facility’s electricity.

Are data centers built in water-stressed areas?

Often. A Bloomberg analysis of WRI Aqueduct data found about two-thirds of U.S. data centers built or in development since 2022 are in areas of high water stress, many in California, Texas and Arizona.

How do you evaluate groundwater risk for a data center?

Identify the aquifer under the site, review historical water levels, model drawdown and interference from the planned pumping rate, check groundwater district permit rules, and project water levels across the facility’s design life. See the data center groundwater due-diligence guide.

HOW ARGOS HELPS

Phreatic — groundwater availability

Per-aquifer water-level forecasts, cone-of-depression and interference modeling, and sustainable pump rates for a site’s design life.

CARI — aquifer reasoning agent

Ask questions about availability, drawdown risk and groundwater district rules for a site in plain English.

SOURCES

Figures are taken from the primary sources above and dated as of the update shown. Concentrations are expressed in mg/L. Spot an error? Email support@argostellus.com.

RELATED GUIDES

Need a site-specific answer?

Argos models groundwater availability, drawdown and contamination for individual sites and portfolios.

CONTACT ARGOS