Aerial drone overview on internet infrastructure, data centers in Middenmeer, The Netherlands. (© Sepia100 - stock.adobe.com)
Data Center Water Risk May Fall on Homes and Farms
In A Nutshell
- About 25% of data centers worldwide (2,315 of 9,347) sit in regions where water demand outstrips supply for at least one month a year.
- At 3 degrees Celsius of warming, that share could rise to one in three.
- Data centers can usually secure water during shortages, so the risk may fall mostly on neighbors sharing the supply.
- The study measured location, not each facility’s water use, so it shows exposure, not proven harm.
Every chatbot question, streamed movie, and cloud backup lands in a data center somewhere, and roughly one in four of those buildings sits in a region where water demand outstrips supply for at least one month a year. That is the finding of a new study that matched nearly every known facility on Earth against local water supplies.
Published in Environmental Science & Technology, the study placed 2,315 of 9,347 facilities in such regions, using water conditions from 2001 to 2010. If the planet warms 3 degrees Celsius (about 5.4 degrees Fahrenheit) above late 1800s levels, that share could climb to one in three.
Those buildings often share rivers, reservoirs, and underground water reserves, called aquifers, with nearby homes, farms, and factories. Estimates cited in the paper put their water use at 313 billion to 764 billion liters a year, possibly more than doubling by 2030 to roughly what a country like the Netherlands withdraws each year for household and city use.
Mapping 9,347 Data Centers Exposes Worldwide Water Shortfalls
Researchers started with a list of operating data centers, among the most complete counts available. A private company, Saoradh Enterprise Partners LLC, supplied it, and the authors say it likely misses some facilities, especially small or new ones built during the current boom.
For each site, the team calculated monthly “water gaps,” meaning months when a region uses more water than nature replaces. A water gap resembles overspending a paycheck: the shortfall comes out of savings, which here means groundwater, rivers, and lakes.
Two runs were made, one with estimated water conditions from 2001 to 2010 and another with projections for a 3-degree-warmer world, a possibility if emissions stay high. Only today’s facilities were counted, so the future numbers show how current locations would fare, not where new ones will go.
Water use at individual buildings was not measured, because consistent worldwide data on cooling systems and water sources are not available. Instead, the team checked whether sites sit in places already under water pressure, and which other users, such as farms, factories, and city water systems, use the most water nearby.
South Asia and the Middle East Stand Out for Data Center Water Stress
Nearly four in 10 data centers, 3,619 in all, are in the United States. Water gaps are minimal in the East, home to dense clusters of facilities, and 79% of American data centers face none. The other 21% face at least one month a year, with parts of the West and South hit for four to eight months and 102 facilities nationwide for nine to twelve. At 3 degrees of warming, the affected share would rise to 35%.
Northern and central Europe, home to most of the region’s facilities, sees little to no stress. Conditions are harsher in South Asia and the Middle East, which the authors say stand out as the most water-stressed regions. In India, 73% of data centers face at least one month of shortage a year, and 33% face nine to twelve months. In the United Arab Emirates, 93% sit in areas with some shortage. In Spain, 64% are affected, a share projected to reach 80% under the warmer scenario.
Then there are the neighbors. Worldwide, 41% of data centers sit where household and city water use is the biggest draw on local supply, and 28% where irrigated farmland leads. In the United States, city use leads for 54%.
Other Water Users May Bear the Data Center Water Risk
Water is a local problem in a way electricity is not. Power can travel long distances, but water use is tied to the nearby river or aquifer, and data centers run around the clock, so they cannot easily cut back during a drought.
Even so, water gaps rarely force a data center to shut down for good. Operators typically have the money and priority status to secure water whenever any is available, and the authors note that pricing, water-sharing rules, and legal priorities often favor cities, industry, and critical infrastructure, data centers included, over farms and natural ecosystems during shortages.
That could leave farmers with less irrigation water, households under watering restrictions, or rivers running thinner. In the authors’ words, “the primary risk may not be to data center operations themselves but to other water users within the same basins.” Basins, in this case, are the shared river and groundwater systems.
None of this proves a fight over water is underway. Sharing a region with heavy water users does not automatically mean competition, the authors caution, since supplies may be sufficient in many places and utilities typically assess supply capacity and regulatory requirements before approving new large water users. Their gap estimates are also a worst-case upper limit, because the model leaves out tools such as moving water between regions and turning seawater into drinking water.
Still, the study shows where trouble could start. Where data centers get built, and how openly they report their water use, could help decide whether neighbors end up carrying the strain.
Disclaimer: This article is for general informational purposes only and summarizes a single peer-reviewed study. Its findings describe estimated exposure to water shortages, not proven harm from any specific data center, and should not be taken as professional, legal, or policy advice.
Paper Notes
Limitations
The authors list several limits. Their data center list came from a private source and likely undercounts facilities, particularly undisclosed, smaller, or newly built ones. The study measures whether data centers sit in water-stressed areas, not how much water individual facilities use, since consistent global data on cooling technology, facility size, and water sourcing is unavailable. Water gap estimates are an “upper-bound estimate” because the model does not represent water transfers between basins, desalination, managed aquifer recharge, reservoir operations, or water rights systems. The baseline period (2001 to 2010) was chosen because the most comprehensive globally validated water-use data set covers those years, and the baseline and future runoff come from different hydrological models, which adds some uncertainty. Water use in sectors other than irrigation was held constant in the future scenario, and results for India are sensitive to how climate models handle monsoons. The 25% figure is modeled exposure under 2001 to 2010 conditions applied to today’s facilities, not a verified 2026 shortage at each site. The authors also say that sharing a region with a major water user does not necessarily mean active competition for water, and that current assessments may underestimate effects on less economically or politically powerful water users. The water use and 2030 projection figures in the article come from earlier studies the paper cites, not from this analysis.
Funding and Disclosures
The authors declare no competing financial interest. The paper lists no specific funding source. The authors thank Saoradh Enterprise Partners LLC of Boulder, Colorado, for providing the data center facilities dataset, which came from the company’s Hydrogen Market Model.
Publication Details
The study, “One Quarter of Global Data Centers are Located in Water-Stressed Areas,” appeared in Environmental Science & Technology (2026, Volume 60, Issue 39, pages 27780 to 27791) as part of the special issue “60th Anniversary of Environmental Science and Technology.” Authors are Lorenzo Rosa (corresponding author; Carnegie Institution for Science and Stanford University), Matteo Sangiorgio (Politecnico di Milano), Bridget R. Scanlon (The University of Texas at Austin), Marco Mazzotti (ETH Zurich), Joseph Berry (Carnegie Institution for Science), and Anna M. Michalak (Carnegie Institution for Science). The paper was received May 28, 2026, revised August 5, 2026, accepted September 15, 2026, and published September 24, 2026, under a CC-BY-NC-ND 4.0 open access license. DOI: 10.1021/acs.est.6c08088. Study results are available at https://zenodo.org/records/22912911.







