Home / Research / AI Data Center Water Usage Statistics

AI Data Center Water Usage Statistics: 2026 Report

AI runs in large data centers, and many of them use water to keep their equipment cool. Large cloud and data center operators now publish yearly figures on how much water they use. But each company reports them in its own way, which makes them hard to compare.

Our analysts studied six of them: Google, Microsoft, Meta, Amazon Web Services (AWS), Equinix and Digital Realty. We took the figures from each company's latest report as of October 8, 2026, from the Google 2026 Environmental Report (2025 data) to the Digital Realty 2025 Impact Report. Our main measure is water withdrawal, the total water a company takes in during a year, stated in megaliters (ML, one million liters). Google's data centers withdrew 51,338 ML in 2025, 1.33 times the 38,679 ML reported by the other five companies combined.

In the following sections, we compare withdrawal and consumption by company, track withdrawal over time, compare water usage effectiveness, and look at water use in water-stressed areas.

AI Data Center Water Withdrawal by Company

The AI Industry Reviews research team compared how much water each company withdrew from local supplies in its latest reported year, with every figure converted to megaliters.

AI Data Center Water Withdrawal, Latest Reported Year

CompanyPeriodScopeReported FigureWithdrawal (ML)
Google2025Data centers13,562 million gallons51,338
Google2025Company-wide14,689 million gallons55,604
MicrosoftFY25 (July 2024 to June 2025)Company-wide (offices, datacenters and labs)13,266 ML13,266
MicrosoftFY25Owned datacenters, 29 listed sites4,717 ML (sum of site figures)4,717
AWS2025Leased, owned and shared data centers9.4 billion liters9,400
Equinix2025Company-wide6,040 ML6,040
Digital Realty2025Managed and operational sites1,539,657 thousand gallons5,828
Meta2024Data centers4,145 ML4,145
Meta2024Company-wide5,637 ML5,637
Horizontal bar chart of water withdrawal in megaliters, latest reported year: Google data centers 51,338 (2025), Microsoft company-wide 13,266 (FY25), AWS data centers 9,400 (2025), Equinix company-wide 6,040 (2025), Digital Realty 5,828 (2025), Microsoft owned datacenters, listed sites 4,717 (FY25), Meta data centers 4,145 (2024).
AI Data Center Water Withdrawal, Latest Reported Year

AI Data Center Water Consumption

Some of the water a data center takes in goes back to rivers and treatment plants, and the rest evaporates in cooling or is otherwise not returned. That second part is called consumption, and it is the water a local area does not get back. AWS reports withdrawal only. Digital Realty reports water by source and labels the total as consumption in its assurance statement, but it reports no discharge, so its figure is shown here as withdrawal.

AI Data Center Water Consumption, Latest Reported Year

CompanyPeriodScopeWithdrawal (ML)Consumption (ML)Consumption as Share of Withdrawal
Google2025Data centers51,33839,83477.6%
Equinix2025Company-wide6,0404,60076.2%
Meta2024Data centers4,1452,97471.7%
MicrosoftFY25Company-wide13,2668,17061.6%
AWS2025Data centers9,400Not reportedNot available
Digital Realty2025Managed and operational sites5,828Not reportedNot available

AI Data Center Water Withdrawal Over Time

Five companies report more than one year of withdrawal. AWS reports a single year.

AI Data Center Water Withdrawal Trend

CompanyScopeFirst Year (ML)Latest Year (ML)Change Over PeriodChange in Latest Year
GoogleCompany-wide2021: 23,8372025: 55,604+133.3%+33.4%
MicrosoftCompany-wideFY20: 6,794FY25: 13,266+95.3%+14.5%
MetaData centers2020: 3,0002024: 4,145+38.2%+6.8%
EquinixCompany-wide2023: 5,9702025: 6,040+1.2%+11.0%
Digital RealtyManaged and operational sites2023: 5,9862025: 5,828-2.6%+4.3%
Line chart of water withdrawal in megaliters from 2020 to 2025: Google company-wide rising from 23,837 in 2021 to 55,604 in 2025, Microsoft company-wide from 6,794 in FY20 to 13,266 in FY25, Meta data centers from 3,000 in 2020 to 4,145 in 2024, Equinix 6,040 and Digital Realty 5,828 in 2025.
Water Withdrawal Over Time, 2020 to 2025

Data Center Water Usage Effectiveness

Total water use rises as a company builds more data centers, so operators also report an efficiency figure that shows how much water they use for each unit of electricity their computers draw.

Water usage effectiveness (WUE) is liters of water per kilowatt-hour (kWh) of electricity used by IT equipment. Each company defines the water side differently, so the figures cannot be ranked against each other as equals.

Data Center Water Usage Effectiveness as Reported

CompanyPeriodWUE (L/kWh)Prior YearWater Counted
AWS20250.120.15Total withdrawal
Meta20240.190.18Total withdrawal
MicrosoftFY250.270.30Cooling and humidification water at owned datacenters operating 12 months
Digital Realty20250.590.70Weighted average of managed sites operating 12 months at 20% utilization or more
Equinix20250.910.95Company-defined, entire data center portfolio
Google2025Not publishedNot publishedNot available
Horizontal bar chart of water usage effectiveness in liters per kWh of IT electricity as each company reports it: Equinix all data centers 0.91 (2025), Digital Realty stabilized managed sites 0.59 (2025), Microsoft owned datacenters 0.27 (FY25), Meta data centers 0.19 (2024), AWS data centers 0.12 (2025).
Data Center Water Usage Effectiveness as Reported

AI Data Center Water Use in Water-Stressed Areas

Each company also reports how much of its water comes from areas it classes as water-stressed, meaning places where demand for water is high compared with the supply available.

Each company uses its own water-stress screen and cutoff, so the shares below measure different things.

Share of Water From Stressed Areas, Latest Reported Year

CompanyPeriodMeasureShare
MicrosoftFY25Withdrawal from areas with high or extremely high baseline water stress, company-wide50%
Digital Realty2025Water from areas with high or extremely high water stress37.6%
Equinix2025Withdrawal from areas with high baseline water stress34.9%
Meta2024Withdrawal from areas with high or extremely high baseline water stress, company-wide30.2%
AWS2025Withdrawal in high or extremely high water stress regions, leased and owned data centers22%
Google2025Freshwater withdrawal from sources at high risk of depletion or scarcity, company-wide13%

What AI Data Center Water Usage Trends Mean for 2027 and Beyond

AI data centers use most of their water in a few places, and those places feel it. A single large campus can draw as much as a small town, and some of that water comes from regions already short on it. How a site is cooled matters more than anything else, because sites that cool by evaporation use far more water than those that do not.

As AI capacity grows, water will play a bigger role in where companies can build. Local officials are starting to weigh water supply in permit decisions, and towns in dry regions may push back or demand cooling designs that use less. Efficiency gains help, but not enough to offset the growth in demand.

So if a data center is proposed near you, ask how it will be cooled and where its water will come from, because that decides how much it takes from the local supply.

Data Collection and Methodology

These figures come from the companies' own reports, and they differ in timing and scope.

Periods differ: Meta's latest public data is for 2024, Microsoft's is for the year ended June 2025, and the rest are for 2025. Microsoft's withdrawal includes offices and labs, its site table covers owned datacenters only, and leased capacity is missing from that table. Equinix and Digital Realty totals are company-wide. AWS does not report consumption, and Digital Realty does not report discharge. Microsoft estimates part of its datacenter figures because most of its sites lack discharge meters. None of the companies splits AI workloads from other computing, so these are fleet totals.

Scope differs by company. Google, AWS and Meta report a data center figure, while Equinix and Digital Realty run data centers almost entirely. Consumption is shown only for the four companies that report both withdrawal and discharge. Gallons are converted to liters at 3.785411784 liters per gallon.

Sources