Data Centres Use Water. So Does Your Lawn. Compared to What?
A popular claim says golf courses use about three times the water of data centres. We checked it against the primary reports. The answer swings from about 2 times to about 30 times depending on a choice most people never state, and the sources measure different things, so they cannot settle it.
Dragonfly Lens · 28 October 2026 · Checked against the golf industry’s 2024 survey report, Lawrence Berkeley National Laboratory’s data-centre report, USGS and EPA documents. We did not trace the claim to a single original post.
The short version
Golf: about 531 billion gallons a year applied to turf in 2024 (a survey projection).12
Data centres: about 17 billion gallons consumed on site and about 211 billion more at power plants, in 2023 (a model).4
The problem: applied water and consumed water are different things, in different years. The 2.3x and 30x ratios are illustrations, not a range.
What is missing: where and when the water is used. That is where harm happens, and no national total shows it.
Plain words, so you can follow along
Acre-foot: the amount of water that covers one acre a foot deep, about 325,851 US gallons. Water reports use it because the numbers are huge.
Applied, withdrawn, consumed: put on, taken from a source, and not returned. Section 1 explains why the difference matters.
Hyperscale and colocation: hyperscale means the giant data centres run by the biggest cloud companies; colocation means buildings where many companies rent space for their own servers.
Direct and indirect water: direct is the water a data centre itself uses, mostly for cooling; indirect is the water used at power plants to make the electricity it draws.
mgd: million gallons a day. One billion gallons a day is 1,000 mgd.
Modelled estimate: a number produced by a computer model from other data, not a count taken at the source.
1. Three different things get called “water use”
Water gets withdrawn (taken from a river or well), applied (put on a lawn or fairway) or consumed (not returned to the local water system, mostly through evaporation). A comparison is only fair when both sides use the same one, for the same year and the same place. That is the whole problem with this claim.
2. The golf side
The best national source is the 2024 water survey of US golf courses. It was commissioned by the Golf Course Superintendents Association of America, administered by the National Golf Foundation, and analysed by two named researchers.12 Responses came from 1,687 facilities, about 12 percent of the roughly 13,950 US golf facilities, and the results are weighted to represent the country. Reports based only on estimated volumes were left out. The survey projects that golf facilities applied 1,631,915 acre-feet of water in 2024. At about 325,851 US gallons per acre-foot, that is roughly 531 billion gallons a year, or about 1.45 billion a day averaged over 2024. It is applied water: what was put on the turf. It is a survey projection, not a nationwide meter reading.1
Two things to know about the trend. Projected applied water is down about 31 percent since 2005 and about 3 percent since 2020.13 But the number of facilities also fell, from about 16,050 in 2005 to about 13,950 in 2024, so part of the decline is courses closing, not courses using less. And recycled water supplied about 18 percent of the water applied in 2024 (296,454 acre-feet; the report’s source chart rounds it to 19 percent).1
3. The data-centre side
Lawrence Berkeley National Laboratory’s December 2024 report on US data centres gives modelled estimates for 2023, not metered totals.4 Direct consumption, on site and mostly for cooling, was 66 billion litres, about 17.4 billion gallons. Indirect consumption, the water used to generate the electricity the data centres draw, was nearly 800 billion litres, about 211 billion gallons. The indirect figure is allocated from the regional power mix where the data centres are, so it includes things like evaporation from hydroelectric reservoirs, and it is not a count of any one facility’s supply. It is about twelve times the direct figure. Both are consumption. A June 2026 “2025 Update” to the report revises electricity estimates; we read its text and found no new water figures.5
4. Putting them together
Same word, different water
US, billions of gallons a year. The bars are NOT the same kind of water: applied vs consumed, different years.
Source: GCSAA 2024 golf water survey; Lawrence Berkeley National Laboratory, Dec 2024 (2023 estimates). Dragonfly Lens arithmetic.
Same word, different water - US, billions of gallons a year. The bars are NOT the same kind of water: applied vs consumed, different years.
Label
Value
Note
Golf courses, applied to turf (2024)
531.0 bn gal, applied
GCSAA survey projection, 1,631,915 acre-feet
Data centres, consumed incl. power plants (2023)
228.0 bn gal, consumed
LBNL model: 17.4 direct + about 211 indirect
Data centres, consumed on site only (2023)
17.4 bn gal, consumed
LBNL model, 66 bn litres
Golf’s applied water against the data centres’ direct plus indirect consumption comes out at about 2.3 times. Golf against direct consumption alone comes out at about 30 times. Those are the 2 and 30 that circulate.
Neither is a fair comparison. One side is applied water in 2024 and the other is modelled consumption in 2023, and the larger figure also adds the power plants’ water on one side only. Treat the two ratios as illustrations of how much the choice of measure moves the answer, not as a range.
Part of golf’s applied water soaks back into the ground or runs off, and some of the water that soaks in is later taken up by the turf. We found no national estimate of how much of golf’s irrigation is actually consumed. The USGS’s older 2015 data, from the 36 states that reported golf separately, put golf-course withdrawals at 1,039 million gallons a day, under 1 percent of irrigation withdrawals, and identify some county-level consumptive use, but give no national consumed share.10 If the consumed share is some fraction, call it f, then the 30-times figure is really about 30 times f. We do not know f, so we will not guess it.
5. What the national totals hide
Water is local. A national total says very little about whether a particular river or town is under strain.
Direction. Golf’s applied water has been falling. The Berkeley report’s scenarios have hyperscale data centres consuming 60 to 124 billion litres directly in 2028.4 We cannot say how fast that is growing, because the report’s 84 percent share for 2023 covers hyperscale and colocation together and hyperscale alone is not separated out. But the direction is up.
Place. Only about one in five US golf facilities is in the Pacific, Southwest or Mountain regions, so golf is not mostly in dry places. But the Southwest alone, with about 8 percent of facilities, accounted for about a quarter of all applied water, because courses there use about twice as much water as courses elsewhere.1 For data centres we did not find a national map of where their water comes from or how stressed those sources are.
Source. About a fifth of golf’s applied water was recycled. Some data centres also use reclaimed water: a 2023 audit report of the City of Quincy, Washington describes a plan to let Microsoft “and other data centers” use treated wastewater for cooling, with the reuse facility started up in May 2021.9 We have not seen a national share.
Timing. We do not have peak-day or monthly figures for either. A hot week can matter more than an annual total.
6. Other uses, for scale only
The US Geological Survey’s 2020 estimates, which are computer-modelled and not the older county surveys, put crop irrigation at about 108 billion gallons a day withdrawn, power plants at about 80 billion and public water supply at about 36 billion.6 Different USGS releases give slightly different numbers for these, so treat them as rounded. Golf’s 1.45 billion a day is about 1.3 percent of the crop-irrigation figure, but it compares a 2024 applied number with a 2020 withdrawal number for the lower 48 states, so it is a sense of scale, not a share.
The Environmental Protection Agency’s WaterSense page says residential outdoor water use is nearly 8 billion gallons a day.7 The page does not say how that was measured. An EPA technical document traces it to a per-person estimate from 2015 USGS data and a 2016 study of indoor use, so it is an old derived estimate, not a recent meter count.8 By that figure, home outdoor use is on the order of five times golf’s applied water, with the same caveats.
All of it side by side, per day
Here is everything above in one place, converted to billions of US gallons a day so the sizes are easy to compare. The right-hand column is the important one: it says what kind of water each number is, because numbers of different kinds should not be divided by each other.
Use
Billion gal / day
What kind of number
Crop irrigation
about 108
withdrawn, 2020, modelled
Power plants
about 80
withdrawn, 2020, modelled
Public water supply
about 36
withdrawn, 2020, modelled
Homes, outdoors
nearly 8
old derived estimate (EPA)
Golf courses
about 1.45
applied, 2024, survey projection
Data centres, all-in
about 0.6
consumed incl. power plants, 2023, modelled
Data centres, on site
about 0.05
consumed directly, 2023, modelled
Our arithmetic: the golf and data-centre annual figures divided by the days in the year. For scale, dividing the public-supply figure by the US population of about 331 million gives about 107 gallons per person per day, which is what an average person’s share of the public water system looks like. Reading down the table, data centres look small and crop irrigation looks enormous; reading the right-hand column, you can see the table is a sense of scale, not a ranking.
7. What this does and does not show
The best primary sources measure different things: applied irrigation on one side, modelled consumption on the other, in different years, with different boundaries. They do not establish the viral ratio, and they do not rank which use does more harm where you live. They do not show that data-centre water is not a problem, and they do not show that golf water is fine.
The honest worries are growth and location, and for those you need local data: a utility’s monthly demand, its supply mix, drought rules, and what happens to the water afterwards.
8. What we could not confirm
A national estimate of how much golf irrigation is consumed; a national map of data-centre water by stressed watershed; how much of data centres’ water is reclaimed; hyperscale-only water use in 2023; and why the golf survey’s own documents give slightly different facility counts and survey lengths (we noted the differences, not resolved them). We did not trace the viral claim to an original post.
9. Five questions to ask of any “X uses more water than Y” claim
Applied, withdrawn or consumed?
Is the power plant’s water counted on both sides, or neither?
Same year, same place, same unit?
Is there a primary source for each side, and is each a measurement or a model?
Is the basis written next to the ratio?
Our call
About 40 percent: by 31 December 2027, Lawrence Berkeley National Laboratory or the US Department of Energy will publish new national estimates of data-centre water consumption for 2024 or later. Starting state, 28 October 2026: the latest water estimates are for 2023, in the December 2024 report; the June 2026 “2025 Update” has no water figures. Resolution source: the laboratory’s data-centre publications page. We will publish the result either way.
Sources
Golf Course Environmental Profile, Phase IV, Volume I: Water Use and Conservation on U.S. Golf Courses (GCSAA, 2025) — gcsaa.org (PDF). Figures read: Tables 1, 3 and 16.
Shaddox, Unruh et al., golf water-use survey paper, HortTechnology 35(5), published online 12 September 2025 — journals.ashs.org (read by our reviewer, not in full by us).
GCSAA news release, “Golf courses reduce water usage by 31 percent according to national survey”, 30 December 2025 — gcsaa.org. An industry-association release.
Shehabi et al., 2024 United States Data Center Energy Usage Report, Lawrence Berkeley National Laboratory, LBNL-2001637, December 2024 — eta-publications.lbl.gov (PDF). Direct, indirect and 2028 water figures read in the report text.
Smith et al., United States Data Center Energy Usage Report: 2025 Update, Lawrence Berkeley National Laboratory, 18 June 2026 — escholarship.org (36 pages; no water figures found).
USGS, “National water use: 2020, average daily water withdrawals by use type and region” (4 April 2025) — usgs.gov. USGS describes its newer estimates as modelled.
EPA WaterSense, “Outdoors” (page last updated 20 July 2026) — epa.gov.
EPA, WaterSense Specification for Soil Moisture-Based Irrigation Controllers: Supporting Statement, February 2021, footnote 1 — epa.gov (PDF).
Washington State Auditor, Financial Statements Audit Report: City of Quincy, 1 January 2020 to 31 December 2021, published 3 April 2023 — quincywashington.us (PDF). A municipal project record, not a water-savings measurement.
USGS, Water Use Across the Conterminous United States, Water Years 2010–20 (Professional Paper 1894-D, 2025), section on golf-course irrigation (2015 data from 36 states) — pubs.usgs.gov (PDF).
Dollar and gallon figures are US. Acre-foot conversion uses 325,851 US gallons. Ratios are our arithmetic and are illustrations, not measurements. Nothing here is advice.