Introduction
The global race to build artificial intelligence infrastructure has an increasingly visible casualty: water. Data centers--the physical backbone of the AI revolution--consume enormous quantities of water to keep their server racks cool and, less obviously, to generate the electricity that powers them. As generative AI accelerates demand for computing capacity, facilities are proliferating at a pace that has outstripped both public understanding and regulatory oversight. Data centers are now estimated to be among the top ten water-consuming industries in the United States, yet their water use remains understudied and largely unaccounted for in water resource planning [1].
The geography of this boom has sharpened the problem. Rather than gravitating toward cooler, wetter regions where cooling demands and freshwater withdrawals would be lower, data center developers have followed cheap power, tax incentives, and favorable regulations into increasingly arid territory. Roughly two-thirds of the data centers built since 2022 have been located in water-stressed regions, according to a Bloomberg News analysis, including hot, dry climates like Arizona [2]. A parallel trend is unfolding internationally, with Saudi Arabia and the United Arab Emirates welcoming more data centers than ever before, while in China and India an even greater proportion of facilities are situated in drier areas than in the United States [3].
The result is a growing set of conflicts--between tech companies and local utilities, between AI growth strategies and public water supply, and between economic development and environmental sustainability. "It is very much a growing issue -- and it is spreading everywhere," says Newsha Ajami, chief strategy and development officer for research at Lawrence Berkeley National Laboratory and founding director of the urban water policy program at Stanford University [3]. This article examines the anatomy of AI's water footprint, the siting conflicts now erupting in water-stressed regions, and the emerging policy frameworks attempting to manage an industry whose thirst has, until recently, been almost entirely invisible.
The Anatomy of AI's Water Footprint
Direct Use: The Cooling Problem
At the facility level, data centers consume water primarily for cooling. High-performance computing hardware--particularly the dense GPU clusters required for training and running large AI models--generates immense heat that must be continuously dissipated, most commonly through evaporative cooling systems that consume water directly from municipal supplies, private wells, or surface sources. The scale of individual facilities makes this consumption significant: the U.S. National Security Agency's Utah data facility, a 1.2 million-square-foot complex atop 250 acres of former sagebrush, requires millions of gallons monthly, drawing from a combination of private wells and a local utility [4].
Because most operators do not disclose facility-level consumption, researchers have had to reconstruct the picture. A forthcoming paper by Gianluca Guidi and Francesca Dominici of the Harvard T.H. Chan School of Public Health examines 472 large U.S. data centers and estimates approximately 300 billion liters of operational water use per year [5]. The study's central insight is that data center water use is not one problem, but two--and that conflating them has muddied both public debate and policy responses.
Indirect Use: The Hidden Thirst of Electricity
The larger and less visible share of AI's water footprint flows through the power grid. The Harvard researchers estimate that roughly three-quarters of operational water use at the facilities they studied is tied to electricity generation rather than on-site cooling [5]. The International Energy Association puts a similar figure on the pattern globally, estimating that 60% of data center water consumption is from indirect use [3]. Thermoelectric power plants--particularly those burning fossil fuels or relying on nuclear fission--consume vast quantities of water for cooling, and a 2021 paper found that nearly half of U.S. data center servers are fully or partially powered by water-hungry power plants located within water-stressed regions [3].
This distinction matters because the two pathways create different demands in different places. Cooling creates intensely local pressure, especially in water-stressed basins in the western and south-central United States, while electricity-related water use is more regional and concentrated around power generation hubs [5]. "As AI infrastructure scales, this kind of granular, evidence-based assessment is what turns 'AI uses a lot of water' into an actionable roadmap for utilities, regulators, and operators to actually reduce it," Dominici notes [5].
The water footprint of artificial intelligence: Emerging solutions and governance imperatives - ScienceDirect
Siting Conflicts: Where AI Ambitions Meet Water Stress
The Geography of Misalignment
The core of the emerging conflict is a mismatch between where data centers want to be and where water can support them. As tech firms have pushed to develop new facilities to support cutting-edge AI systems, they have increasingly turned to states and countries with ample energy resources and favorable regulations--places that often lack abundant water supplies. The result, researchers and journalists observe, is that data centers threaten local water supplies, agriculture, and energy production [3].
The siting calculus has traditionally rewarded cheap land, proximity to airports and major roads, local tax breaks, inexpensive power, and lax utility connection rules--factors that led the NSA to Utah, where its facility now sits in a region grappling with chronic drought [4]. "Wherever they choose to put a data center, it is like a giant soda straw sucking water out of that basin," says Colohan of the Lincoln Institute. "And when you take water from a place, you have to reduce demand or put water back in that same place, there's no other solution" [2].
A Case Study in Policy Incoherence: The United Kingdom
Nowhere are the tensions more explicit than in the United Kingdom, where a recent government-commissioned analysis found that AI's water footprint is escalating at an alarming rate and is largely unaccounted for in national water resource planning [6]. The UK's designation of data centers as Critical National Infrastructure (CNI)--intended to bolster digital resilience--creates what the report describes as a direct policy tension: the government's drive for AI growth and its heightened responsibility for data center operational resilience potentially conflict with existing water scarcity challenges and the ability of water companies to manage supply equitably in stressed regions [6].
The geography makes the risk concrete. The overlap between areas designated as "seriously water stressed" and existing or proposed data center clusters has created predictable "conflict hotspots." The first government-designated "AI Growth Zone" in Culham, Oxfordshire, sits just seven miles from the site of a planned new reservoir at Abingdon intended to supply customers in the Thames Valley, London, and Hampshire. Thames Water, the utility responsible for the area, is itself classified as seriously water stressed [6]. The concentration of data centers in regions such as Greater London further intensifies exposure to localized climate risks like heatwaves and water stress, and the report warns that policy incoherence could lead to unintended consequences for public water supply and environmental flows [6].
Global Hotspots
The pattern repeats worldwide. Arid Gulf states are courting data center investment as part of AI diversification strategies, while India and China are concentrating facilities in drier regions even as climate change heightens water scarcity [3]. In the American West and Southwest, water-stressed areas are facing particularly intense demands as facility concentrations grow [7]. Even as operators adopt water reuse technologies or onsite water sources, these measures do not change the fact that water stresses are hitting different regions with greater frequency and intensity [7].
Data centers' hidden water footprint is linked to the grid | Latitude Media
The Strain on Communities and Water Systems
Fragmented Utilities, Concentrated Demands
The burden of managing these impacts falls largely on a patchwork of water utilities. In the United States, water infrastructure is overseen by more than 50,000 local water systems--a hyper-fragmented landscape in which individual systems, often with limited technical and financial capacity, must suddenly negotiate with some of the world's largest corporations [7]. The challenges facing these systems include significant and ongoing water use, expanded infrastructure costs such as new distribution lines to reach exurban facilities, and regulatory and financial pressures [7].
The concentration effect compounds the strain. Because facilities cluster in particular regions, some areas face far more intense demands than others--particularly water-stressed areas in the West and Southwest [7]. Each approved development raises the baseline pressure on the next, compounding a problem that researchers argue is not even fully understood yet [8]. There is also an equity dimension: data centers can inadvertently drive up electricity prices for residential ratepayers as utilities build new generation capacity, and they compete with residential users and agriculture for finite water resources [4].
The Case for Regional Thinking
Policy analysts increasingly argue that the solution lies in scaling up governance to match the scale of the demand. Addressing the growing footprints of data centers alongside mounting water infrastructure investment needs requires more regional thinking and planning, which can help spread financial risk, build greater institutional capacity, and seize economic development opportunities [7]. Some rural communities with acute capacity constraints are already considering such regional approaches, while leaders in larger urban regions are recognizing the benefits of formal and informal regional collaborations [7]. Community advocates similarly emphasize that local officials must engage in discussions about data centers early on, with a coordinated, holistic understanding of existing resources and potential impacts on the energy grid and the watershed [2].
Emerging Policy Frameworks
The United Kingdom: Planning Under Scarcity
The UK report represents one of the most direct attempts by a national government to confront the tension between AI strategy and water security. Its authors conclude that the rapid expansion of AI presents a profound challenge to the country's water security and environmental sustainability, and that the CNI designation heightens the government's responsibility to ensure these facilities operate sustainably [6]. The implicit message is that data center location policy is now inseparable from water resource planning--and that decisions about where facilities go are critical to avoiding the worst conflicts [6].
California: Building a Regulatory Playbook
In the United States, California has emerged as a testbed for regulation. A report from the Berkeley Law Center for Law, Energy & the Environment examines how the state--a major data center hub that also faces increasing water stress--might better assess and manage facilities' direct and indirect water consumption [1]. Among its key insights is that siting matters enormously: placing a data center in a warmer, water-stressed area can create a direct conflict with other local water needs, whereas locating facilities in low-stress areas or near alternative water sources--such as a wastewater treatment plant that can provide recycled water--can reduce water-related impacts [1]. The report also catalogs the many non-water factors that typically drive siting decisions, illustrating why water considerations are so often an afterthought [1].
Regulatory scholars have proposed a complementary set of tools: mandating efficiency standards that incentivize water-efficient and closed-loop cooling systems, prohibiting the cost-shifting of water infrastructure expenses onto other ratepayers, and requiring comprehensive environmental reviews demonstrating minimal impact on aquifers and watersheds, particularly in water-stressed regions [4].
Technical Standards and the Feedback Loop
Engineering frameworks are evolving alongside regulation. The ASHRAE AI Data Center Energy Performance Framework calls on developers to evaluate energy and water availability and regional resource constraints when selecting cooling strategies, particularly in water-stressed regions, and to balance energy efficiency with responsible water use while incorporating sustainability considerations early in site planning [9]. Researchers have also mapped the systemic dynamics at play--AI data centers consume water for cooling, water scarcity constrains siting, and, notably, AI tools can in turn improve water system efficiency--a feedback loop that suggests both risk and opportunity in how the industry and water sector co-evolve [10].
AI is thirsty: Fresh water has become a key resource of the digital economy -- The Insider
Pathways Toward Sustainable AI Infrastructure
The emerging consensus among researchers, utilities, and advocates is that the water crisis of AI is tractable--but only with deliberate intervention across technology, siting, and governance.
Cooling technology. Innovative cooling systems can reduce freshwater use by up to 75%, while nascent technologies can virtually eliminate evaporative water use altogether. However, these systems are currently too expensive to be adopted at scale without regulatory incentives, which is why scholars argue frameworks should mandate efficiency standards rather than merely encourage them [4].
Smarter siting. Locating data centers in cooler, wetter regions reduces both the volume of water required and the impact of freshwater withdrawals. Yet since 2022, the industry has largely moved in the opposite direction [2]. Guidi and Dominici's two-pathway analysis offers a guide for action: some places need better cooling systems, reclaimed water, and drought planning, while others need cleaner, less water-intensive electricity [5].
Replenishment and reuse. In some cases, major data center developers have begun to recognize the replenishment imperative--putting water back into the basins from which they draw--and are actively engaging in water replenishment where it counts [2]. Siting near recycled water sources offers another route to decoupling facilities from potable supplies [1].
Integrated planning. At the broadest level, experts argue that integrated land use planning--with sustainable land, water, and energy practices--is the only way to sustainably achieve the virtuous circle needed to reap AI's benefits and the economic growth associated with it [2]. In the Indo-Pacific, a joint initiative by the Australian Water Partnership and the Alliance for Global Water Adaptation is working to equip cities and regions with the policy frameworks and tools needed not just to survive an AI boom but to thrive in it, arguing that policymakers must incentivize water efficiency and infrastructure investment, investors need to price in long-term resource risk, and the industry must prioritize water resilience alongside energy and infrastructure [8]. Community benefit agreements, in this framing, should be seen as frameworks for sustainable growth--because if taps run dry, so does business [8].
Conclusion
The water footprint of artificial intelligence has moved from an obscure technical footnote to a frontline policy conflict, playing out in reservoir planning battles in Oxfordshire, drought-stressed basins in Utah and Arizona, and regulatory hearings in Sacramento. The research now converges on a clear diagnosis: AI's water use is large, growing, geographically concentrated in precisely the places least able to spare it, and split between visible on-site cooling and a larger hidden footprint in electricity generation [6][3][5].
What remains contested is the response. The tools exist--closed-loop cooling, recycled water partnerships, regional utility collaboration, mandatory disclosure, environmental review, and community benefit agreements [5][7][4][1][8]. What has been missing is the coordinated will to deploy them at the pace of AI-driven construction. Governments that have designated data centers as critical national infrastructure now bear a heightened responsibility to ensure these facilities operate sustainably, lest the pursuit of digital resilience undermine the water security on which everything else depends [6]. The choice facing policymakers, investors, and communities is stark but tractable: plan for water as an integral constraint on AI growth, or watch each new approved development compound a crisis that is only beginning to be understood. As the Alliance for Global Water Adaptation puts it, water resilience must be treated as critical to business resilience--because in the age of AI, the two are no longer separable [8].
References
- 1.
- 2.
- 3.
- 4.
- 5.
- 6.
- 7.
- 8.
- 9.
- 10.