Ten Across | Environment
Are data centers depleting the Southwest’s water and energy resources?
Iron Mountain’s AZP-1 data center in Phoenix, Arizona. Photo taken on Feb. 19, 2025 by Maya Chari.
Metropolitan Phoenix is the second-largest hub for data centers in the country, but the industry’s growth risks straining the city’s energy grid and depleting its limited water supply
Editor’s note: This article is part of a collaboration between APM Research Lab and the Ten Across initiative, housed at Arizona State University.
by MAYA CHARI | February 27, 2025
Iron Mountain’s AZP-1 data center is tucked under the curve of Highway 202, just before it meets Highway 143. It’s next to a post office, a laundromat, and a pizza-and-wings joint. Roughly the size of a city block, the data center’s campus appears to consist of a chrome-and-glass office building, several other low, concrete buildings with no windows, a half-full parking lot, rows of electrical boxes, and cooling towers. Signs in English and Spanish warn of high voltage. Security is intense: the property is surrounded by, at various places, a wrought-iron fence, a concrete wall, and barbed wire, punctuated by security booths with guards who will, if you step off the sidewalk, remind you that you are on private property.
This data center is one of three that Iron Mountain operates in metropolitan Phoenix. In 2023, Google broke ground on a data center in Mesa, Arizona. Microsoft also operates data centers in El Mirage and Goodyear. All together, Phoenix hosts about 707 megawatts of IT capacity, more than any major city besides Dallas.
The Phoenix area is attractive to companies seeking to build data centers for several reasons: environmental and labor policies are relatively lax, and industrial operations are lured to the state with tax incentives. Electricity is cheaper than in neighboring California. There are no major natural disasters other than extreme heat, which is considered easier to deal with than other natural forces due to its predictability.
Crucially, the Phoenix area is also located on a major fiber-optic pipeline that runs from Texas to Southern California, connecting several metropolitan areas across the southwestern part of the country.
But data centers are resource-intensive. Especially in hot climates, they require cooling with either air or water. The growth of data centers risks straining the Southwest’s energy grid and depleting its limited water supply.
The rise of data centers
A data center is an offsite computing structure that houses servers which store and process information in digital form. Enterprise data centers are owned by large companies like Google or Amazon for their own use. Colocation data centers, like those operated by Iron Mountain, are owned by third parties and rented out to smaller companies for a rate based on square footage.
Before the turn of the millennium, data centers weren’t integral to digital life. The bulk of internet traffic traveled between on-site locations like homes or businesses. The shift towards offsite, shareable data storage offered as a service, or “cloud computing,” accelerated in the early 2000s with the arrival of commercial providers like Amazon Web Services and Microsoft Azure. Cloud computing gained favor with corporate information technology (IT) departments because it is easier and cheaper to update and scale software in the cloud than it is on many different physical machines.
Recent years have also seen a shift towards compute- and data-intensive workflows: processes that involve numerous calculations and significant data storage and thus increase the demand for data centers. Video conferencing and streaming fall into this category.
Between 2010 and 2018, the amount of data generated globally increased almost thirtyfold, from 1.2 zettabytes to 35 zettabytes. In 2025, data generation is projected to jump further to 175 zettabytes, with artificial intelligence (AI) and cryptocurrency driving much of the increase.
All of that data has to live somewhere. The demand for third-party data centers quadrupled between 2020 and 2024 and is still growing. President Donald Trump recently announced the Stargate Initiative, a partnership between the federal government and several AI companies. The companies, which include Softbank, OpenAI, and Oracle, intend to invest $500 billion in new data infrastructure. That follows President Joe Biden signing an executive order in the last month of his presidency securing energy resources for data centers.
Cooling data centers and the water-energy tradeoffs
It’s ideal for data centers to be kept at an ambient temperature of 68 to 82 degrees Fahrenheit. Cooling methods in data centers broadly fall into three categories: refrigeration cycles, which use outdoor air, water, or chemical refrigerants to cool air; adiabatic cooling, which cools air by blowing it through water-moistened pads; and free cooling, which uses colder ambient air to reduce indoor temperatures.
Adiabatic coolers are more effective in arid climates, like Phoenix, where water easily evaporates. The technique uses relatively little energy but quite a bit of water. The same is true for water-cooled refrigeration systems. Air-cooled refrigeration systems use less water, but more energy. This is known as the water-energy tradeoff: cooling techniques that use less water tend to use more energy, and vice versa.
The challenges of measuring data center water use
While Arizona presents many advantages for data centers, the state is already highly water-stressed. In addition to the water used to cool data centers directly, the electricity plants that the centers draw from, whether thermoelectric, nuclear, or coal-fired, also consume water for cooling. (Solar facilities use the least amount of water.) A full picture of a data center’s water use must take both on- and off-site use into account.
Companies that operate data centers aren’t always required to report the amount of water they withdraw. Last year, a journalist covering Microsoft’s data center in Goodyear, for example, found that its water use records were considered proprietary. An analysis Microsoft shared with the city council, however, estimated the data center’s water use at around 56 million gallons of potable water annually, equivalent to 670 Goodyear households or 1.9% of the town’s population.
Water to cool data centers doesn't need to be potable —- though it often is — but it does require filtration to avoid contamination by metals or salts. In Arizona, data centers typically draw from the Central Arizona Project or the Salt River Project, though some also pump groundwater. Any water that data centers in Arizona draw from the Colorado river comes out of Arizona’s allocation, even if the company operating the data center is based elsewhere.
Data centers can also recycle their water, though it’s not universal. Recycling water requires space and manpower, and doesn’t eliminate demand entirely. In water-cooled systems that use cooling towers, for instance, about 50% to 80% percent of the water can be recovered.
Google and Meta have published sustainability reports that document their water usage. In 2021, Google’s 15 U.S. data centers withdrew around 4.2 billion gallons of water. Over 3.3 billion of those gallons were consumed, meaning that they weren’t returned to their original source. That’s about as much water consumption as 35,486 U.S. households. The amount of water consumed by each data center varied widely, ranging from 100,000 gallons annually to over 845 million gallons.
Meta’s data centers, meanwhile, withdrew 1.3 billion gallons of water in 2021, 367 million of which were from areas with high or extremely high water stress. Total global water consumption from Meta’s data centers was over 635 million gallons, equivalent to about 6,697 U.S. households. It’s not clear how much of this water withdrawal occurs in the United States, although that’s where most of Meta’s data centers are located. Neither report reveals the specific water use of the company’s Arizona data center.
Data centers and electricity demand
Data centers typically pull electricity from the city grid, for which they are charged industrial rates. As of November 2024, the average industrial rate for electricity in the state of Arizona was 7.89 cents per kilowatt hour, up 17% from 2014. That’s considerably lower than the residential electricity rate, which was 17.01 cents per kilowatt hour, a 36% increase from 2014.
Because data centers use so much electricity, there’s potential for them to strain the grid, particularly during peak summertime hours, when electricity demand in Arizona cities is at its highest. Kerri Hickenbottom, a researcher at the University of Arizona who has studied the water footprint of data centers in a collaboration with the Salt River Project, speculated that in the future, data centers might adopt a tiered pricing structure, charging consumers more during hours of peak demand.
The cryptocurrency Bitcoin, for example, is considering charging more for coin mined during peak hours. “Moving forward, these data centers might really resemble our energy utilities and their price structures, with tiered usage,” said Hickenbottom.
How data centers can use less energy and water
According to Leila Karimi, a chemical engineer and colleague of Hickenbottom, it’s very possible for data centers to operate more efficiently. They can capture and store rainwater, reducing reliance on municipal water sources and groundwater — even in Arizona. More efficient recycling techniques are also being developed.
Directly connecting data centers to energy sources can reduce both their carbon footprint and grid strain. Google and Amazon are investing in using small nuclear reactors to power their data centers.
Data centers that use air or water coolers can reduce energy use with devices called economizers. In an economized chilling system, when temperatures outside are cold, the air or water bypasses the chilling mechanism and is simply filtered and used to cool infrastructure directly. Unfortunately, Karimi said, retrofitting technologies is costly and complex, though companies often make up the costs in the long term due to energy and water savings.
There’s also potential for artificial intelligence to reduce its data demand, which is especially important since much of the projected increase in data center demand comes from AI. The Chinese company DeepSeek, for example, has created highly sophisticated AI models that appear to be compute- and data-efficient, meaning that they train on a relatively small amount of high-quality data and therefore require fewer resources.
Due to the relatively low price of water, however, the financial incentives for data centers to conserve water don’t reflect the issue’s importance. Karimi emphasized that gains in water efficiency, while technically possible, are unlikely to come without significant pressure from communities.
But without that pressure, and in the absence of limitations on the amount of water they can withdraw, companies might not see the need to make their data centers less thirsty.
“Water is still really cheap,” said Hickenbottom. “We really undervalue our water resources.”
The author would like to thank Robert Norwood of the University of Arizona and Oswald Chong of Arizona State University for valuable background information that informed this article.