Last fall, when a handful of urban beavers moved into new habitat in northern Colorado’s high country, they arrived courtesy of a perhaps surprising supporter: Google.
The tech giant has invested in local nonprofit Boulder Watershed Collective’s beaver restoration work, one of the company’s more than 100 such water conservation and quality projects across the country (a Google representative declined to share the amount donated to BWC). What’s Google’s interest in busy, buck-toothed rodents? Beavers are natural engineers, building dams that hold water on the landscape. And these days, tech companies have reason to care very much about water.
The data centers that run cloud computing — and increasingly, artificial intelligence (AI) — have a collectively enormous energy and water footprint. Data centers, relied upon by companies like Google, Meta, Microsoft, and Amazon, are essentially giant warehouses storing the physical infrastructure that powers the digital world.
Today, in what’s often referred to as the AI goldrush, these tech giants and startups, are investing big money in data center development to meet AI demand — a projected $650 billion in 2026, up 59% over 2025 spending, according to an analysis by Bridgewater Associates. The computer servers in data centers use huge amounts of electricity as they perform searches and stream videos, generating lots of heat. Hot computers won’t function, so data centers must cool the equipment. Both cooling equipment and generating electricity often require water — and lots of it. Google’s investment in the Boulder beavers and their ability to conserve water is part of its goal to replenish 120% of the company’s total water use.
Exactly how much water a data center taps remains a difficult question to answer. Tech companies don’t have to publicly report their data centers’ water use, and in many cases, have proven reluctant to disclose it. That lack of transparency creates headaches for communities that are already stressed about water scarcity. If the public doesn’t know how much water these computing powerhouses will swallow up, then how can we be sure any given data center won’t use too much?
What We Know — And Don’t Know
Not all data centers are created equal. Smaller centers that have handled online searches and streaming for years don’t need as many resources as the rapidly expanding hyperscale centers doing more complex work for AI development. Speaking in broad averages, a mid-sized data center today uses some 110 million gallons of water per year for cooling, or 337 acre-feet, roughly the same as 1,000 American households, according to a 2025 report from the Environmental and Energy Study Institute. But hyperscale data centers may use around 1.8 billion gallons annually, or more than 5,520 acre-feet, according to that report.

A look inside one of Google’s data centers. CONNIE ZHOU/ GOOGLE/ ZUMAPRESS.COM
Taken together, those numbers have raised concerns, especially in drought-stricken parts of the country. A report by Lawrence Berkeley National Laboratory found that data centers in the U.S. collectively consumed more than 17.4 billion gallons, or 53,399 acre-feet in 2023, and are on track to gulp 32.76 billion gallons, or 100,537 acre-feet annually, by 2028. In the five states of Colorado, Arizona, Nevada, New Mexico and Utah alone, data centers could consume 4.5 billion gallons, or 13,810 acre-feet, annually on site by 2030, according to a Western Resource Advocates study. If no sideboards are put in place, experts expect these numbers to skyrocket as AI drives the frenzied construction of more data centers.
All those numbers, though, are basically meaningless when it comes to an individual data center.
“[Water use] can vary from zero to many million liters per day,” says Shaolei Ren, an associate engineering professor at the University of California, Riverside. “It really depends on the cooling system.” It also depends on the season — even low-water cooling systems use water during hotter months, which is when demand peaks for other water users as well, adding stress to water systems.
Some methods guzzle water for evaporative cooling, while others employ closed-loop liquid or air-cooling systems that use little or none. Air-cooling is similar to air conditioning, and closed-loop liquid cooling systems recirculate the same liquid coolant around computer equipment — the coolant doesn’t need to be refilled but evaporative cooling may still be used to cool the facility. Importantly, waterless cooling systems generally use significantly more electricity.
Cool and Dry
Innovations in cooling technologies let data centers drastically cut their water use

Courtesy of Google
Do data centers actually consume staggering amounts of water? It depends on the type of data center and the technology used to cool its servers. Methods that rely on evaporation are notorious for high water use, while other technologies get the job done with less, or even no, water.
Refrigeration cooling or dry cooling
These methods work like air conditioners, circulating cool air in a data center and removing hot air. This is done through an air-cooled chiller (known as dry cooling) or a water-cooled chiller. Air-cooled chillers are more energy-intensive but consume no water, while water-cooled chillers may use a cooling tower — those with cooling towers are generally the most water intensive cooling systems used in data centers.
Adiabatic- or direct-evaporative cooling
These systems use water to cool the air that is circulated in a data center. These systems are similar to swamp coolers found in residential and commercial buildings. They work well in dry climates, but not in humidity.
Free cooling, aka ambient cooling
Uses fans to draw in “free,” colder outside air. After passing through the building, the heated air is vented outdoors. This method is best suited to colder climates, and is often paired with a backup system for when the weather gets too warm.
Liquid cooling
These methods rely on the fact that liquid conducts heat more efficiently than air. With immersion cooling, computer hardware is placed in a fluid that conducts heat but not electricity. The liquid absorbs the servers’ heat, releases it via a heat exchanger, then cools and flows back into the system. Direct-to-chip liquid cooling is even more targeted, attaching cold plates to the hottest parts of the equipment. The coolant liquid flows through the plates, removing heat and returning to the system. Liquid cooling is increasingly being used by AI data centers.
A data center’s electricity source makes a major difference in its water use, too. Fossil fuel plants consume large amounts of water, while renewables don’t use any. The greener a data center’s electricity, the less water it will use.
And that’s only the water a data center uses directly in its on-site cooling system — a fraction of its total water footprint.
A 2021 study from Virginia Tech and Lawrence Berkeley National Laboratory researchers estimates that 75% of a center’s water use is actually indirect, consumed by off-site power plants as they generate electricity to run the servers. The International Energy Agency projects that electricity consumption at U.S. data centers will more than double from 183 terawatt-hours (TWh) in 2024 to 426 TWh by 2030.
The amount of water required for that surge in power generation varies depending on the source of energy. In a statement, Xcel Energy notes that “It is difficult to predict exactly how data centers will be operating in the future and what mix of generation technologies will be supporting them. Overall, Xcel Energy is using less water for generation every year as more renewable generation like solar and wind replaces coal and gas.”

Blue pipes carry cold water into the facilities and red pipes take hot water out. Keeping ata centers cool can require large volumes of water, a demand that is rising alongside cloud computing and AI.
CONNIE ZHOU/ GOOGLE/ ZUMAPRESS.COM
Often data centers’ water and energy demands vary inversely, says Lindsay Rogers, municipal water conservation policy manager with Western Resources Advocates. Some technologies, like evaporative cooling, use less water but more energy, and other technologies, like dry cooling, will use more energy but less water.
“Unfortunately, it’s a tricky thing to answer exactly how much water and energy these facilities use,” Rogers says. “That’s partially because there’s not a lot of transparent data available at the site level on their demands.”
Local utilities know how much water data centers use, and in some cases, the companies themselves release the numbers. But many have resisted making their stats public — binding officials to nondisclosure agreements. This opacity makes it hard for officials and the public to evaluate the impacts of existing and proposed data centers.
Business Insider went to court to obtain water-use records from the Denver Water Board and Colorado Springs Utilities last year. Tech groups opposed a 2025 California bill that would have required data centers to report their water use, arguing it would expose trade secrets, arm competitors with sensitive information, and even create security risks (Governor Gavin Newsom ultimately vetoed the bill). Colorado’s proposed SB26-102 would have required large-load data centers to disclose annual electricity and water consumption but failed in the legislative session.
Plus, says Dan Diorio, the Denver-based vice president of state policy for the Data Center Coalition, mandating disclosure for data centers but not other large users, like industrial plants or golf courses, doesn’t make sense. “Other members of the system get a free ride to use as much water as they want to because they aren’t being targeted for disparate treatment,” he notes.
Despite concerns about high water use, data centers are popping up all over the Southwest, including in Colorado. The Centennial State hosts nearly 60, mostly in the Denver Metro area; other clusters have grown around Phoenix, Reno and Salt Lake City. California’s status as a tech hub has attracted even more.
Currently, Xcel Energy has received about 35 applications for data centers across its systems in Colorado. Tri-State Generation and Transmission Association says that it has received requests for 8 Gigawatts of new power, largely from data centers — more than three times the utility’s existing 2.5 GW system which spans four Western states including Colorado. Not all proposed data centers end up being built.
Companies go for sites with lots of available, affordable land to develop; access to electricity, plus electrical transmission and fiber optic infrastructure; an acceptable level of risk for natural disasters; and tax incentives. Colorado currently lacks the latter: It does not have a sales and use tax exemption for computer equipment. A proposed tax incentive for data centers failed in the state legislature in both 2025 and 2026.
Water availability matters, but it’s not necessarily a developer’s number-one concern. In some cases, such as in Goodyear, Arizona, tech companies have paid to upgrade a municipality’s water system so it could accommodate a new data center’s surge in use.
Similarly, access to energy can be a limiting factor for data center development. But some projects, such as southern New Mexico’s Project Jupiter, aim to skirt that challenge by building self-contained microgrids to generate their own electricity.
To Regulate or Incentivize?
Colorado lawmakers have tried to lead when it comes to AI and data centers, but differing approaches have left them stymied. Here’s what’s been happening with state legislation:
SB24-205, Consumer Protections for Artificial Intelligence
Aimed to govern high-risk AI systems in order to protect consumers from risks of AI discrimination. The bill was signed into law in 2024 and lauded by consumer groups as landmark legislation. Meanwhile, many in the tech industry warned that it could create heavy administrative burdens for startups, deter innovation, and slow economic growth. Lawmakers pushed the policy’s start date back so legislators could make changes during the 2026 session. While SB24-205 was set to take effect in June, Colorado’s Attorney General agreed to suspend the state’s enforcement of the law when a federal judge granted a stay in a lawsuit brought against the state by Elon Musk’s xAI.
SB26-189, Automated Decision-Making Technology
Replaces SB24-205, narrowing the scope of the previous law. This new law simply requires disclosure when automated decision-making technology is being used and requires that more info is provided to consumers if requested.
SB26-102, Large-Load Data Centers
Failed in session, but would have set guardrails on large-load data centers with requirements around utility contracts, environmental impacts, operations and location. Among other things, the bill called for data centers to implement water-efficient technology and report annual electricity and water consumption.
SB26-1030, Data Center and Utility Modernization
Failed in session, but would have incentivized data center development in Colorado by allowing a 100% state sales and use tax exemption on qualified purchases to the operator of a certified data center.
Planning for the Boom
With interest and investment in data centers exploding, a few communities have attempted to get in front of tricky water-use questions. In 2024, Aurora became a national leader when it developed a large water users guide — the guide was codified in 2025 — capping how much water a new development is allowed to consume. Any company anticipating water use above a certain daily threshold simply won’t be served. As the city saw increasing interest from data centers and other industrial operations in recent years, officials at Aurora Water began to get nervous.
“We don’t have an unlimited supply of water,” says Chong Woo, deputy director of engineering services at Aurora Water. “This is a way to ensure that we have a water supply that will allow us to supply our community
well into the future.”
Water rights are becoming harder to find in Colorado, says Aurora Water’s deputy director of internal and external affairs, Shonnie Cline, adding to the urgency. “This is protective to both industry and residents — looking at the long view, as opposed to just jumping on an economic opportunity that sounds really good for today, but that could potentially be damaging to the municipality if we don’t have the resources to serve them into perpetuity,” she says.
Since the policy went into effect, the city has negotiated with proposed data centers to get their water use down to acceptable levels.
“We’ve asked data centers to be more creative in terms of how their processes consume water,” says Vern Adam, Aurora Water’s recently retired deputy director of engineering services. “We’ve had users that no longer use wet evaporative cooling, and are using other means to provide cooling for their servers.”
Also last year, Tucson passed an ordinance regulating “large-quantity water users” after public outcry sank a proposed data center called Project Blue. The new rules require any prospective water-intensive development to apply to the city council (making its water use public record), submit a water conservation plan, and agree to use some recycled water.
In 2024, the town of Marana, Arizona, went even further by passing a city ordinance stating it wouldn’t supply potable water to a data center at all, and that a company must disclose its water use publicly.
“A couple of years ago, we were excited about the prospect of having data centers move into our community,” says Marana Mayor Jon Post.
“Then we started to hear some discontent from some of the communities in metro Phoenix about what data centers were taking from their resources. We wanted to make sure that they did not take from the resources that we have available to our residents. We have done a really good job with our water portfolio. We feel we are positioned well for future growth in our community. But that did not include data centers.”
The regulations haven’t stopped data centers from approaching Marana. As of last winter, Post says the city has been in talks about at least three proposed developments since passing the ordinance, all of which would use a closed-loop cooling system.
“Having the ordinance in place is going to make [the application] process much simpler,” he says. “They already know what they have to do to get approvals from the town.”
In Colorado, more communities are taking a step back to set data center policies. In Spring 2026, Larimer County, Jefferson County, Saguache County, and the Denver City Council adopted a moratorium on applications for new data centers, to allow them time to gather public input and develop land use regulations. Similarly, in April, Weld County Commissioners added data centers to their land use code, limiting data center construction to industrial zones, and requring proof of water.
With so many unknowns about water use within such a rapidly evolving industry, perhaps the best water policies will be customized to every data center — and its neighbors. “The question is, how do you use the least amount of water as possible, and how do you do that in conjunction with the stakeholders in the community?” says Howard Neukrug, executive director of the University of Pennsylvania’s Water Center, which recently launched the Water-AI Nexus Center of Excellence. “Water is a global issue. The solutions are all very local.”
Elisabeth Kwak-Hefferan is a writer who specializes in the environment. She writes for The New York Times, Sierra, National Geographic, and more.
Print