The Uptime report

Uptime Report Research · Water Balance

Data Center Water Usage: Google’s Doubled in Three Years, Keeping Pace With Its Power, While Rivals Cut Water per kWh

Data center water usage is rising, but at most big operators it is rising more slowly than electricity. Google’s data centers consumed 10.5 billion gallons (39.8 billion liters) in 2025, up 102% from 2022, almost exactly in step with their 106% electricity growth. Microsoft’s water withdrawals rose 95% in five years against 244% for power, and Meta’s 38% against 159%.

10.5B galconsumed by Google data centers in 2025, 35% more than in 2024
+102% vs +106%Google data center water vs electricity, 2022 to 2025
1.35B galconsumed at Council Bluffs, IA, the largest site any company discloses
800B Lwater used at power plants for US data center electricity, 2023 (LBNL), about 12x on-site use
0.12 L/kWhAmazon’s 2025 water use effectiveness, down 52% since 2021, the lowest disclosed

The water balance: where data center water comes from and goes

A data center’s water account has two sides. Withdrawal is everything piped in; consumption is the part that does not come back, mostly evaporated in cooling towers. Google and Meta publish both for their data centers; Google also breaks out where the water came from and publishes consumption for every site. In 2025 its data centers withdrew 13,562 million gallons, discharged 3,039 million and consumed 10,523 million.

Google data centers, 2025 water balance (million gallons)

SourcesPotable: 10,949 million gallons (41.4 billion liters)Potable 10,949Non-potable: 1,695 million gallons (6.4 billion liters)Reclaimed: 918 million gallons (3.5 billion liters)WithdrawnWithdrawal: 13,562 million gallons (51.3 billion liters)Withdrawal 13,562Where it wentConsumed (mostly evaporated): 10,523 million gallons (39.8 billion liters)Consumed (mostly evaporated) 10,523Discharged: 3,039 million gallons (11.5 billion liters)Discharged 3,039Potable 10,949 + non-potable 1,695 + reclaimed wastewater 91878% of the water withdrawn was consumed; 22% was discharged to sewers or water bodies.

Source: Google 2026 Environmental Report, water use by data center location (p. 97). Excludes seawater and brackish water, which Google does not count.

19% of that water was non-potable or reclaimed wastewater, so 81% was drinking-quality supply. Google rates 13% of its freshwater withdrawals as coming from sources at high risk of depletion or scarcity, and 15% at medium risk. The other companies use different tests: Microsoft says 50% of its fiscal 2025 withdrawals came from water-stressed areas, Meta 30% of its 2024 withdrawals from high or extremely high baseline water stress, and Amazon 22% of its 2025 data center withdrawals.

Is data center water usage growing faster than electricity?

We tested this with each company’s own water and electricity series. At three of the four, water grew more slowly than power, so water per kilowatt-hour fell. Google is the exception.

Water vs electricity, indexed to the first reported year (= 100)

electricity water

Google data centers10020030020222025Google data centers, 2022: electricity index 100Google data centers, 2023: electricity index 116Google data centers, 2024: electricity index 149Google data centers, 2025: electricity index 206Google data centers, 2022: water consumed index 100Google data centers, 2023: water consumed index 117Google data centers, 2024: water consumed index 149Google data centers, 2025: water consumed index 202power 206water 202Microsoft, all operations (FY)10020030020202025Microsoft, all operations (FY), 2020: electricity index 100Microsoft, all operations (FY), 2021: electricity index 126Microsoft, all operations (FY), 2022: electricity index 169Microsoft, all operations (FY), 2023: electricity index 219Microsoft, all operations (FY), 2024: electricity index 277Microsoft, all operations (FY), 2025: electricity index 344Microsoft, all operations (FY), 2020: water withdrawn index 100Microsoft, all operations (FY), 2021: water withdrawn index 114Microsoft, all operations (FY), 2022: water withdrawn index 128Microsoft, all operations (FY), 2023: water withdrawn index 142Microsoft, all operations (FY), 2024: water withdrawn index 171Microsoft, all operations (FY), 2025: water withdrawn index 195power 344water 195Meta data centers10020030020202024Meta data centers, 2020: electricity index 100Meta data centers, 2021: electricity index 131Meta data centers, 2022: electricity index 160Meta data centers, 2023: electricity index 215Meta data centers, 2024: electricity index 259Meta data centers, 2020: water withdrawn index 100Meta data centers, 2021: water withdrawn index 114Meta data centers, 2022: water withdrawn index 121Meta data centers, 2023: water withdrawn index 129Meta data centers, 2024: water withdrawn index 138power 259water 138

Google: data center water consumption and data center electricity, 2022 to 2025 (data-center-only water starts with 2022 data). Microsoft: company-wide withdrawal and electricity, fiscal 2020 to 2025. Meta: data center withdrawal and electricity, 2020 to 2024.

Google: data center water consumption rose 102% from 2022 to 2025 and withdrawal 105%, against 106% for data center electricity. Liters consumed per kWh barely moved: 0.96 in 2022 and 0.94 in 2025. In 2025 alone water rose 35% and electricity 38%. Google says water is a highly efficient means of cooling that requires less energy than air cooling, that it assesses local water risk before using it, and it reports a fleet-wide power usage effectiveness of 1.09.

Microsoft: withdrawals grew 95% and consumption 105% from fiscal 2020 to 2025, while electricity grew 244%. Withdrawal per kWh of electricity fell from 0.63 to 0.36 liters. These are company-wide totals, including offices.

Meta: data center withdrawals grew 38% from 2020 to 2024 against 159% for data center electricity, and Meta’s own water use effectiveness fell from 0.30 to 0.19 liters per kWh of IT load. Meta’s 2024 figures leave out another 1,019 megaliters used to build new data centers.

Amazon: no total water series before 2025, but its water use effectiveness fell from 0.25 liters per kWh in 2021 to 0.12 in 2025. In 2025 its data centers withdrew 9.4 billion liters (2.5 billion gallons).

Year Google DC withdrawal (M gal) Google DC consumption (M gal) Google L consumed per kWh Microsoft withdrawal (ML) Microsoft consumption (ML) Microsoft L withdrawn per kWh Meta DC withdrawal (ML) Meta WUE (L/kWh) Amazon WUE (L/kWh)
2020 n/a n/a n/a 6,794 3,990 0.63 3,000 0.30 n/a
2021 n/a n/a n/a 7,747 4,794 0.57 3,418 0.26 0.25
2022 6,621 5,220 0.96 8,698 5,329 0.48 3,618 0.20 0.19
2023 7,657 6,101 0.96 9,666 5,818 0.41 3,881 0.18 0.18
2024 9,866 7,787 0.96 11,590 6,693 0.39 4,145 0.19 0.15
2025 13,562 10,523 0.94 13,266 8,170 0.36 not yet not yet 0.12

M gal = million US gallons; ML = megaliters (million liters; 1 million gallons = 3.79 ML). Microsoft years are fiscal years ending June 30. “Not yet”: Meta has not published 2025.

Water use effectiveness: the numbers are not comparable

Water use effectiveness (WUE) is usually liters of water withdrawn per kWh of IT equipment load. The companies define and publish it differently, and Google does not publish it at all, so we computed Google’s ratios from its own totals. Because Google divides by total data center electricity rather than IT load, its true WUE on the usual definition would be slightly higher.

Company L/kWh Year Definition Source
Amazon (AWS) 0.12 2025 Liters withdrawn per kWh of IT load (company WUE) 2025 Sustainability Report
Meta 0.19 2024 Liters withdrawn per kWh of IT load (company WUE) 2025 Environmental Data Index
Microsoft 0.27 2025 Company-reported fleet average WUE Microsoft blog, June 24, 2026
Microsoft (our calculation) 0.36 FY2025 Liters withdrawn per kWh of total electricity, all operations 2026 Environmental Data Fact Sheet
Google (our calculation) 1.21 2025 Liters withdrawn per kWh of data center electricity 2026 Environmental Report
Google (our calculation) 0.94 2025 Liters consumed per kWh of data center electricity 2026 Environmental Report

Which data centers use the most water?

Google, Microsoft and Meta now publish water by site. We put all 77 disclosed sites with at least 1 million gallons of withdrawal in one table, ranked by withdrawal, the one metric all three report. Google’s sites dominate: its 14 largest, led by Council Bluffs, IA, Mayes County, OK, New Albany, OH, each withdrew more water than Microsoft’s largest site, Quincy (WA) (341 million gallons). The fastest-growing large Google site was Lockbourne, OH, where consumption rose from 143 to 360 million gallons in one year.

Where both electricity and water are published, the ratio shows how each site is cooled. Among Microsoft sites using at least 100,000 MWh, Singapore withdrew 1.25 liters per kWh and Dublin, Ireland 0.01. Among Meta’s, Gallatin (TN) withdrew 0.57 liters per kWh and Forest City (NC) 0.03.

# Site Company Year Withdrawn (M gal) Withdrawn (ML) Consumed (M gal) L withdrawn per kWh County in drought (avg)
1 Council Bluffs, IA Google 2025 1,746.4 6,611 1,346.0 n/a 29%
2 Mayes County, OK Google 2025 1,396.7 5,287 1,081.4 n/a 2%
3 New Albany, OH Google 2025 1,008.1 3,816 835.7 n/a 8%
4 Berkeley County, SC Google 2025 988.5 3,742 901.7 n/a 34%
5 St. Ghislain, Belgium Google 2025 714.4 2,704 450.8 n/a outside US
6 Papillion, NE Google 2025 677.8 2,566 548.2 n/a 40%
7 The Dalles, OR Google 2025 654.3 2,477 469.0 n/a 37%
8 Eemshaven, Netherlands Google 2025 532.2 2,015 405.7 n/a outside US
9 Montgomery County, TN Google 2025 499.5 1,891 424.7 n/a 17%
10 Lockbourne, OH Google 2025 460.8 1,744 359.5 n/a 9%
11 Douglas County, GA Google 2025 440.6 1,668 367.1 n/a 32%
12 Henderson, NV Google 2025 401.7 1,521 239.4 n/a 94%
13 Lenoir, NC Google 2025 378.3 1,432 344.3 n/a 15%
14 Bristow, VA Google 2025 359.6 1,361 267.6 n/a 57%
15 Quincy (WA) Microsoft FY2025 341.3 1,292 not disclosed 0.94 40%
16 Lancaster, OH Google 2025 307.1 1,162 135.8 n/a 8%
17 Quilicura, Chile Google 2025 278.2 1,053 143.0 n/a outside US
18 Midlothian, TX Google 2025 263.6 998 220.2 n/a 4%
19 Phoenix (AZ) Microsoft FY2025 259.2 981 not disclosed 1.03 87%
20 Sterling, VA Google 2025 258.4 978 199.8 n/a 70%
21 Omaha, NE Google 2025 241.8 915 183.9 n/a 32%
22 Leesburg, VA Google 2025 238.1 901 190.7 n/a 70%
23 Jackson County, AL Google 2025 191.0 723 173.6 n/a 13%
24 Clonee (Ireland) Meta 2024 150.8 571 not disclosed 0.53 outside US
25 Singapore Microsoft FY2025 111.7 423 not disclosed 1.25 outside US
26 San Antonio (TX) Microsoft FY2025 111.0 420 not disclosed 0.29 97%
27 Boydton (VA) Microsoft FY2025 95.6 362 not disclosed 0.12 22%
28 Inzai, Japan Google 2025 94.6 358 73.2 n/a outside US
29 Fredericia, Denmark Google 2025 88.8 336 71.8 n/a outside US
30 Prineville (OR) Meta 2024 86.6 328 not disclosed 0.19 52%
31 Fort Worth (TX) Meta 2024 82.2 311 not disclosed 0.28 16%
32 Odense (Denmark) Meta 2024 77.1 292 not disclosed 0.51 outside US
33 Ashburn, VA Google 2025 76.5 290 53.2 n/a 70%
34 Des Moines (IA) Microsoft FY2025 69.7 264 not disclosed 0.12 8%
35 Red Oak, TX Google 2025 67.0 254 50.4 n/a 4%
36 Storey County, NV Google 2025 66.7 252 6.7 n/a 20%
37 Los Lunas (NM) Meta 2024 66.6 252 not disclosed 0.22 18%
38 Altoona (IA) Meta 2024 63.9 242 not disclosed 0.15 40%
39 Ashburn (VA) Microsoft FY2025 63.1 239 not disclosed 0.29 85%
40 New Haven, IN Google 2025 62.2 235 37.8 n/a 36%
41 Huntsville (AL) Meta 2024 55.2 209 not disclosed 0.24 52%
42 Gallatin (TN) Meta 2024 54.2 205 not disclosed 0.57 15%
43 Chicago (IL) Microsoft FY2025 53.9 204 not disclosed 0.55 29%
44 Cheyenne (WY) Microsoft FY2025 49.7 188 not disclosed 0.17 80%
45 Stanton Springs (GA) Meta 2024 38.6 146 not disclosed 0.12 15%
46 Sarpy (NE) Meta 2024 37.5 142 not disclosed 0.11 63%
47 Eagle Mountain (UT) Meta 2024 35.1 133 not disclosed 0.12 0%
48 DeKalb (IL) Meta 2024 27.7 105 not disclosed 0.28 23%
49 Henrico (VA) Meta 2024 24.3 92 not disclosed 0.10 19%
50 New Albany (OH) Meta 2024 22.7 86 not disclosed 0.16 29%
51 Jakarta, Indonesia Microsoft FY2025 22.2 84 not disclosed 3.79 outside US
52 Mesa, AZ Google 2025 19.4 73 9.7 n/a 96%
53 Mesa (AZ) Meta 2024 15.1 57 not disclosed 2.31 65%
54 Middenmeer, Netherlands Google 2025 13.6 51 9.3 n/a outside US
55 East Wenatchee (WA) Microsoft FY2025 13.5 51 not disclosed 0.17 45%
56 Hollands Kroon, Netherlands Microsoft FY2025 12.2 46 not disclosed 0.04 outside US
57 Kansas City (MO) Meta 2024 10.8 41 not disclosed 1.79 n/a
58 Luleå (Sweden) Meta 2024 7.7 29 not disclosed 0.06 outside US
59 Kuala Lumpur, Malaysia Microsoft FY2025 6.1 23 not disclosed 1.15 outside US
60 Atlanta (GA) Microsoft FY2025 5.0 19 not disclosed 1.62 21%
61 Queretaro, Mexico Microsoft FY2025 5.0 19 not disclosed 2.99 outside US
62 Dublin, Ireland Microsoft FY2025 4.8 18 not disclosed 0.01 outside US
63 Hamina, Finland Google 2025 4.3 16 0.4 n/a outside US
64 Forest City (NC) Meta 2024 4.2 16 not disclosed 0.03 6%
65 Madrid, Spain Microsoft FY2025 4.0 15 not disclosed 0.66 outside US
66 Gävle-Sandviken, Sweden Microsoft FY2025 3.7 14 not disclosed 0.02 outside US
67 Hanau, Germany Google 2025 3.4 13 1.9 n/a outside US
68 Manassas (VA) Microsoft FY2025 2.9 11 not disclosed 0.28 65%
69 Wilmer, TX Google 2025 2.6 10 0.3 n/a 1%
70 Melbourne, Australia Microsoft FY2025 2.4 9 not disclosed 0.70 outside US
71 Sydney, Australia Microsoft FY2025 2.1 8 not disclosed 1.04 outside US
72 Busan, South Korea Microsoft FY2025 1.8 7 not disclosed 0.04 outside US
73 Milan, Italy Microsoft FY2025 1.6 6 not disclosed 0.13 outside US
74 Sydney, Australia Google 2025 1.4 5 0.2 n/a outside US
75 Frankfurt, Germany Google 2025 1.2 5 0.2 n/a outside US
76 Taipei, Taiwan Microsoft FY2025 1.1 4 not disclosed 0.05 outside US
77 Vienna, Austria Microsoft FY2025 1.1 4 not disclosed 0.29 outside US

Google: calendar 2025, consumption also disclosed. Microsoft: fiscal 2025, owned data centers only (sites under 1% of its total are excluded by Microsoft). Meta: 2024, owned data centers (leased capacity reported only as a total). L per kWh uses each site’s total electricity. Drought: average weekly share of the site’s county in D1 to D4 during the reporting year, from the US Drought Monitor; we assigned each site to a county from its published name.

Are the biggest water users in drought areas?

Mostly not. We matched each disclosed US site to its county and averaged the US Drought Monitor’s weekly share of that county in moderate drought or worse over the same year as the company’s figures. Counties that averaged at least half their area in drought held 16% of the disclosed US withdrawal. The largest water users there were Google’s Henderson, NV (402 million gallons, 94% in drought), Google’s Bristow, VA (360 million gallons, 57% in drought), Microsoft’s Phoenix (AZ) (259 million gallons, 87% in drought).

Disclosed US data center sites: water withdrawn vs county drought

Google (2025) Microsoft (FY2025) Meta (2024)

COUNTY HALF OR MORE IN DROUGHT, ON AVERAGE0.111010010000%25%50%75%100%Average share of the county in moderate drought or worse (D1 to D4) during the reporting yearWater withdrawn, million gallons (log scale)Google Council Bluffs, IA (2025): 1,746.4 million gallons withdrawn; Pottawattamie County, IA averaged 29% of its area in droughtGoogle Mayes County, OK (2025): 1,396.7 million gallons withdrawn; Mayes County, OK averaged 2% of its area in droughtGoogle New Albany, OH (2025): 1,008.1 million gallons withdrawn; Licking County, OH averaged 8% of its area in droughtGoogle Berkeley County, SC (2025): 988.5 million gallons withdrawn; Berkeley County, SC averaged 34% of its area in droughtGoogle Papillion, NE (2025): 677.8 million gallons withdrawn; Sarpy County, NE averaged 40% of its area in droughtGoogle The Dalles, OR (2025): 654.3 million gallons withdrawn; Wasco County, OR averaged 37% of its area in droughtGoogle Montgomery County, TN (2025): 499.5 million gallons withdrawn; Montgomery County, TN averaged 17% of its area in droughtGoogle Lockbourne, OH (2025): 460.8 million gallons withdrawn; Franklin County, OH averaged 9% of its area in droughtGoogle Douglas County, GA (2025): 440.6 million gallons withdrawn; Douglas County, GA averaged 32% of its area in droughtGoogle Henderson, NV (2025): 401.7 million gallons withdrawn; Clark County, NV averaged 94% of its area in droughtGoogle Lenoir, NC (2025): 378.3 million gallons withdrawn; Caldwell County, NC averaged 15% of its area in droughtGoogle Bristow, VA (2025): 359.6 million gallons withdrawn; Prince William County, VA averaged 57% of its area in droughtMicrosoft Quincy (WA) (FY2025): 341.3 million gallons withdrawn; Grant County, WA averaged 40% of its area in droughtGoogle Lancaster, OH (2025): 307.1 million gallons withdrawn; Fairfield County, OH averaged 8% of its area in droughtGoogle Midlothian, TX (2025): 263.6 million gallons withdrawn; Ellis County, TX averaged 4% of its area in droughtMicrosoft Phoenix (AZ) (FY2025): 259.2 million gallons withdrawn; Maricopa County, AZ averaged 87% of its area in droughtGoogle Sterling, VA (2025): 258.4 million gallons withdrawn; Loudoun County, VA averaged 70% of its area in droughtGoogle Omaha, NE (2025): 241.8 million gallons withdrawn; Douglas County, NE averaged 32% of its area in droughtGoogle Leesburg, VA (2025): 238.1 million gallons withdrawn; Loudoun County, VA averaged 70% of its area in droughtGoogle Jackson County, AL (2025): 191.0 million gallons withdrawn; Jackson County, AL averaged 13% of its area in droughtMicrosoft San Antonio (TX) (FY2025): 111.0 million gallons withdrawn; Bexar County, TX averaged 97% of its area in droughtMicrosoft Boydton (VA) (FY2025): 95.6 million gallons withdrawn; Mecklenburg County, VA averaged 22% of its area in droughtMeta Prineville (OR) (2024): 86.6 million gallons withdrawn; Crook County, OR averaged 52% of its area in droughtMeta Fort Worth (TX) (2024): 82.2 million gallons withdrawn; Tarrant County, TX averaged 16% of its area in droughtGoogle Ashburn, VA (2025): 76.5 million gallons withdrawn; Loudoun County, VA averaged 70% of its area in droughtMicrosoft Des Moines (IA) (FY2025): 69.7 million gallons withdrawn; Polk County, IA averaged 8% of its area in droughtGoogle Red Oak, TX (2025): 67.0 million gallons withdrawn; Ellis County, TX averaged 4% of its area in droughtGoogle Storey County, NV (2025): 66.7 million gallons withdrawn; Storey County, NV averaged 20% of its area in droughtMeta Los Lunas (NM) (2024): 66.6 million gallons withdrawn; Valencia County, NM averaged 18% of its area in droughtMeta Altoona (IA) (2024): 63.9 million gallons withdrawn; Polk County, IA averaged 40% of its area in droughtMicrosoft Ashburn (VA) (FY2025): 63.1 million gallons withdrawn; Loudoun County, VA averaged 85% of its area in droughtGoogle New Haven, IN (2025): 62.2 million gallons withdrawn; Allen County, IN averaged 36% of its area in droughtMeta Huntsville (AL) (2024): 55.2 million gallons withdrawn; Madison County, AL averaged 52% of its area in droughtMeta Gallatin (TN) (2024): 54.2 million gallons withdrawn; Sumner County, TN averaged 15% of its area in droughtMicrosoft Chicago (IL) (FY2025): 53.9 million gallons withdrawn; Cook County, IL averaged 29% of its area in droughtMicrosoft Cheyenne (WY) (FY2025): 49.7 million gallons withdrawn; Laramie County, WY averaged 80% of its area in droughtMeta Stanton Springs (GA) (2024): 38.6 million gallons withdrawn; Newton County, GA averaged 15% of its area in droughtMeta Sarpy (NE) (2024): 37.5 million gallons withdrawn; Sarpy County, NE averaged 63% of its area in droughtMeta Eagle Mountain (UT) (2024): 35.1 million gallons withdrawn; Utah County, UT averaged 0% of its area in droughtMeta DeKalb (IL) (2024): 27.7 million gallons withdrawn; DeKalb County, IL averaged 23% of its area in droughtMeta Henrico (VA) (2024): 24.3 million gallons withdrawn; Henrico County, VA averaged 19% of its area in droughtMeta New Albany (OH) (2024): 22.7 million gallons withdrawn; Licking County, OH averaged 29% of its area in droughtGoogle Mesa, AZ (2025): 19.4 million gallons withdrawn; Maricopa County, AZ averaged 96% of its area in droughtMeta Mesa (AZ) (2024): 15.1 million gallons withdrawn; Maricopa County, AZ averaged 65% of its area in droughtMicrosoft East Wenatchee (WA) (FY2025): 13.5 million gallons withdrawn; Douglas County, WA averaged 45% of its area in droughtMicrosoft Atlanta (GA) (FY2025): 5.0 million gallons withdrawn; Fulton County, GA averaged 21% of its area in droughtMeta Forest City (NC) (2024): 4.2 million gallons withdrawn; Rutherford County, NC averaged 6% of its area in droughtMicrosoft Manassas (VA) (FY2025): 2.9 million gallons withdrawn; Prince William County, VA averaged 65% of its area in droughtGoogle Wilmer, TX (2025): 2.6 million gallons withdrawn; Dallas County, TX averaged 1% of its area in droughtGoogle Phoenix, AZ (2025): 0.5 million gallons withdrawn; Maricopa County, AZ averaged 96% of its area in droughtGoogle Pflugerville, TX (2025): 0.4 million gallons withdrawn; Travis County, TX averaged 82% of its area in droughtCouncil BluffsMayes CountyNew AlbanyBerkeley CountyHendersonBristow

Drought statistics: U.S. Drought Monitor, county percent of area, traditional (cumulative) statistics. The U.S. Drought Monitor is jointly produced by the National Drought Mitigation Center at the University of Nebraska-Lincoln, the United States Department of Agriculture, the National Oceanic and Atmospheric Administration and the National Aeronautics and Space Administration.

Drought is a short-term weather measure, not a test of long-run supply, and a county average can hide a utility with ample reserves. For disaster exposure by state, see our disaster declarations by state data; for where data centers are concentrated, our data centers by state atlas.

The bigger water footprint is at the power plant

Water used to generate a data center’s electricity does not show up in on-site figures. LBNL’s 2024 report for Congress estimates that the electricity used by US data centers in 2023 consumed nearly 800 billion liters of water at power plants, about 12 times the direct total, or 4.52 liters per kWh against 4.35 for US electricity in general (report page 57). Of the four companies, only Meta reports this: water embedded in its purchased electricity was 72,207 megaliters in 2024, 24 times its data centers’ on-site consumption. Our data center statistics cover LBNL’s on-site series and its 2028 range.

On-site water vs water at the power plant

US data centers, 2023 (LBNL)US data centers, 2023 (LBNL): direct 66.0 billion literson site 66.0US data centers, 2023 (LBNL): indirect, at power plants 800.0 billion litersat power plants 800.0Meta, 2024 (company report)Meta, 2024 (company report): direct 3.0 billion literson site 3.0Meta, 2024 (company report): indirect, at power plants 72.2 billion litersat power plants 72.2Billion liters of water consumed. Each pair is drawn on its own scale.

LBNL’s indirect estimate assumes each data center uses its regional grid mix and ignores power purchase agreements. Meta’s figure uses the grid electricity it bought; Meta reports a separate figure for the water embedded in its contracted renewable energy, 5,075 megaliters.

That makes energy efficiency a water measure too: on both estimates, most of the water behind a data center is used where its electricity is generated. Our report on AI energy consumption tracks how fast that electricity use is growing.

How much water does an AI prompt use?

Company claims, not audited figures. Google says a median Gemini Apps text prompt consumed 0.26 mL of water in May 2025, estimated from energy per prompt times its fleet-wide water use effectiveness. OpenAI’s chief executive wrote that an average ChatGPT query uses about 0.000085 gallons (0.32 mL). Mistral’s lifecycle study, which counts manufacturing and training as well as use, puts a 400-token response from its Le Chat assistant at 45 mL. The three figures count different things, so they cannot be compared directly.

What this means for MSPs

Use site-level data in client questions. Clients in drought-prone states may ask about the water behind their cloud region. Google, Microsoft and Meta now publish water by site, and the table above links each to its county drought record. Point to that data, with its year, instead of generic per-data-center estimates.

Ask colocation providers for WUE, and for its definition. Published WUE runs from 0.12 to about 1.2 liters per kWh depending on the company and on whether the figure counts withdrawal or consumption, IT load or total load. When a provider quotes WUE, ask which numerator and denominator it uses before comparing.

Count electricity as water. For a client’s environmental reporting, the water at the power plant can be an order of magnitude larger than on-site use, according to LBNL and Meta. Cutting server energy, through consolidation and right-sizing, cuts that indirect water as well.

Frequently asked questions

How much water does a data center use?

It ranges from almost nothing at air-cooled sites to millions of gallons a day. Google’s largest disclosed site, Council Bluffs, IA, withdrew 1,746 million gallons in 2025, about 4.8 million gallons a day, and consumed 1,346 million. Microsoft’s largest, Quincy (WA), withdrew 341 million gallons in fiscal 2025, and Meta’s largest, Clonee (Ireland), 151 million gallons in 2024.

Do data centers really use a lot of water?

Google’s data centers consumed 10.5 billion gallons (39.8 billion liters) in 2025, which Google compares to irrigating about 70 golf courses in the southwestern US. But more water is used at the power plants that supply data centers: LBNL estimates 800 billion liters of indirect water for US data centers in 2023, about 12 times their direct use.

How many gallons of water a day does a data center use?

From the companies’ site tables: Google’s Council Bluffs, IA averaged 4.8 million gallons a day withdrawn in 2025, and the median Google site in its table withdrew 652 thousand gallons a day. Several air-cooled sites, such as Dublin and Phoenix, used almost none.

Is data center cooling water reused?

Partly. 19% of the water Google’s data centers withdrew in 2025 was non-potable or reclaimed wastewater, and 22% was discharged back rather than consumed. Microsoft reports 74% non-potable water at Quincy (WA).

Is data center water use growing faster than electricity use?

Not at three of the four companies. Microsoft’s water withdrawals grew 95% from fiscal 2020 to 2025 while its electricity grew 244%; Meta’s data center withdrawals grew 38% against 159% for power. Amazon’s water per kWh fell 52% since 2021. Google is the exception: its data center water consumption rose 102% from 2022 to 2025, close to its 106% electricity growth.

How much water does an AI prompt use?

By the companies’ own figures: 0.26 mL for a median Gemini Apps text prompt (Google, May 2025) and about 0.32 mL for an average ChatGPT query (OpenAI CEO Sam Altman). Mistral’s lifecycle study, which includes manufacturing and training, puts a 400-token Le Chat response at 45 mL. All three are unaudited company claims.

Download the data. data-center-water-by-site-2026.csv (CSV). Licensed under CC BY 4.0: free to reuse with attribution to The Uptime Report.

Methodology

Sources and retrieval. In October 2026 we downloaded Google’s Environmental Reports (2023 to 2026 editions), Microsoft’s 2026 Environmental Data Fact Sheet, Meta’s 2025 Environmental Data Index and Amazon’s 2025 Sustainability Report, and parsed the water and electricity tables from the PDF text. Each Google data center year comes from the edition that first reports it (2022 from the 2023 edition, 2023 from the 2024 edition, 2024 from the 2025 edition, 2025 from the 2026 edition), because Google publishes the data-center-only split for the latest year only. We checked that the per-site rows sum to Google’s stated data center total. Drought data are weekly county statistics from the U.S. Drought Monitor web service, January 2024 to December 2025.

Definitions. Withdrawal: water taken from utilities, surface or ground sources. Consumption: withdrawal minus discharge, mostly evaporated. Google and Meta report data centers separately; Microsoft’s totals include offices, while its per-site table covers owned data centers. Liters per kWh for Microsoft and Meta sites is site withdrawal divided by site electricity. County drought is the mean weekly share of county area in D1 (moderate) to D4 (exceptional), using cumulative statistics, over each company’s reporting year (Google 2025, Microsoft July 2024 to June 2025, Meta 2024). Counties were assigned by us from the published site names (for example Council Bluffs to Pottawattamie County, Iowa); sites on county lines may draw water from a neighboring county.

Limits and anomalies. Company figures are self-reported; each says an outside accounting firm reviewed selected metrics. Meta’s 2021 data center consumption (162 ML) and office consumption (2,406 ML) appear to be swapped in its own table, so we do not use 2021 Meta consumption. Amazon publishes no site data. Google’s figures exclude seawater and brackish water. LBNL’s indirect estimate is a model, not a measurement. Per-prompt water figures are company claims.

Not advice. This is an analysis of public disclosures, not engineering or environmental advice.

Cite as:The Uptime Report. "Data Center Water Usage: Google's Doubled in Three Years, Keeping Pace With Its Power, While Rivals Cut Water per kWh." October 2026. https://theuptimereport.com/data-center-water-usage/

References

  1. Google, 2026 Environmental Report, water tables pp. 96 and 97, electricity table p. 93. Accessed October 2026.
  2. Google, 2025 Environmental Report, water use by data center location, pp. 110 to 111. Accessed October 2026.
  3. Google, 2024 Environmental Report, data center water, p. 80. Accessed October 2026.
  4. Google, 2023 Environmental Report, data center water, p. 95. Accessed October 2026.
  5. Microsoft, 2026 Environmental Data Fact Sheet, Tables 5, 8, 14 and 15. Accessed October 2026.
  6. Microsoft, “Inside Microsoft’s two-decade push to cut water intensity while scaling for growth,” June 24, 2026. Accessed October 2026.
  7. Meta, 2025 Environmental Data Index, sections 2.1 and 3.1 to 3.4. Accessed October 2026.
  8. Amazon, 2025 Sustainability Report, Water section. Accessed October 2026.
  9. Shehabi, A., et al., 2024 United States Data Center Energy Usage Report, Lawrence Berkeley National Laboratory, LBNL-2001637, pp. 55 to 57. Accessed October 2026.
  10. U.S. Drought Monitor, data services (county statistics by percent of area), National Drought Mitigation Center, USDA, NOAA and NASA. Accessed October 2026.
  11. Elsworth, C., et al. (Google), “Measuring the environmental impact of delivering AI at Google Scale,” arXiv:2508.15734. Accessed October 2026.
  12. Altman, S., “The Gentle Singularity,” June 10, 2025. Accessed October 2026.
  13. Mistral AI, “Our contribution to a global environmental standard for AI”. Accessed October 2026.

Research library

More original data reports from The Uptime Report. Every report includes its methodology and a free CSV download. Browse all research

Threats and breaches

Infrastructure and uptime

Regulation and disclosure

Workforce and pay

MSP market and spending

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *

More posts