Though data centers consume far less water than industries such as agriculture and manufacturing, their growing demand can become a serious issue in water-stressed regions. Fortunately, new cooling technologies and smarter planning could significantly reduce their impact.
"Protect our water." "Water for people, not AI." "Don't mess with our water."
From Texas to New Mexico and Arizona, these messages have appeared on protest signs as communities push back against the rapid expansion of data centers. Residents are increasingly concerned about the water use, energy consumption and pollution associated with the infrastructure needed to power the AI boom.
Many data centers use water as part of their cooling systems. The principle is similar to sweating: water absorbs heat and carries it away as it evaporates. This is necessary because the powerful processors inside data centers can reach temperatures of up to 176 degrees Fahrenheit.
Yet while there is little doubt that data centers consume enormous amounts of electricity, their water footprint is more difficult to measure. Online claims range from warnings that AI chatbots consume 500 milliliters of water per query to arguments that AI's water problem barely exists. Meanwhile, water-use data published by major technology companies is often incomplete or inconsistent.
Experts emphasize that, nationally, data center water consumption remains small compared with agriculture and some forms of manufacturing. Researchers estimate that data center cooling systems consumed around 66 billion liters of water in the United States in 2023 — less than 1 percent of the country's total water consumption.
But that figure could rise sharply as AI infrastructure expands. While this may have little impact in water-rich regions, it could worsen shortages in drought-prone areas such as Arizona and New Mexico.
The good news is that there are several ways to reduce the pressure. Better cooling systems, cleaner energy and smarter choices about where to build data centers could dramatically lower AI's water footprint.
"There are many good ideas out there," says energy systems expert Fengqi You of Cornell University. Combined with efforts to restore and replenish local water resources, he says, some facilities could eventually approach net-zero water use for on-site cooling.
More renewable energy and better locations
AI technology itself is already becoming more efficient.
Electrical and computer engineer Shaolei Ren of the University of California, Riverside, says his widely cited 2024 estimate — suggesting that writing a short email with GPT-4 could consume around 500 milliliters of water — is already outdated. AI models have become more efficient since then.
In 2025, Google estimated that processing a median-length query with Gemini used roughly five drops of water.
Still, even very small amounts can become significant when multiplied by billions of AI queries. You's research group recently estimated that by 2030, US data centers could consume between 731 billion and 1.125 trillion liters of water annually. At the upper end, that is roughly equivalent to New York City's annual drinking water supply.
Importantly, these estimates include more than just the water used for cooling inside data centers. A large share of AI's water footprint comes indirectly from electricity generation.
Coal and gas power plants, for example, often use large amounts of water to cool steam after it has driven electricity-generating turbines. Solar and wind power, by comparison, require little to no water during operation.
That means shifting data centers toward renewable energy could significantly reduce their overall water footprint.
Location matters too. You argues that companies should prioritize areas with abundant renewable energy and lower water stress — such as parts of Montana, Nebraska, Texas and South Dakota — rather than building heavily in drought-prone regions like Arizona, New Mexico and Southern California.
According to his research, strategic site selection combined with other efficiency measures could reduce AI's future water footprint by as much as 86 percent.
New cooling technologies
Cooling technology is also improving rapidly, and many new AI-focused data centers are more water-efficient than older facilities.
Traditional data centers typically use fans to remove heat from racks of servers. Chilling systems then cool the surrounding air, while cooling towers release heat — often losing significant amounts of water through evaporation.
This process can consume large quantities of water.
AI processors, however, generate so much heat that air cooling alone is increasingly insufficient. As a result, companies including Amazon, Microsoft and Google are adopting liquid cooling systems.
Instead of cooling the entire room, liquid cooling directs fluid through pipes close to the processors themselves. Because these systems can operate in closed loops, the same fluid can circulate repeatedly rather than being continuously consumed.
The heated fluid can then release its heat into the outside air when temperatures allow, or through refrigeration-like systems that require electricity instead of large amounts of water.
During extremely hot or humid conditions, however, additional cooling may still be needed. Some facilities use water-based methods, such as misting systems, during the hottest periods of the year.
Mechanical engineer Vaibhav Bahadur of the University of Texas at Austin says many AI data centers in Texas only rely heavily on water-based cooling during the hottest parts of summer.
Nevertheless, water demand could grow substantially. Bahadur and his colleagues estimate that Texas data centers, which currently consume less than 1 percent of the state's water demand, could account for between 3 and 9 percent by 2040.
He expects continued improvements in efficiency to reduce that impact.
"Data centers are becoming much more efficient in their water usage," he says.
Designing processors that need less cooling
Another promising development is the ability of newer processors to operate safely at higher temperatures.
According to Josh Parker, NVIDIA's head of sustainability, the company's latest processors can be cooled using water at temperatures of around 113 degrees Fahrenheit — much warmer than traditional systems require.
This reduces the need for intensive cooling. Unless outdoor temperatures regularly exceed that level, large, efficient fans may be enough to remove heat without relying heavily on water-based systems.
Ren notes, however, that operating processors at higher temperatures may affect performance, and many data centers contain older or non-AI servers that cannot tolerate the same conditions.
Other technologies are also being explored.
Some companies are developing advanced pipe systems designed to transfer heat more efficiently away from processors. Others are experimenting with immersion cooling, in which servers are placed directly into tanks filled with specialized cooling liquids.
In China, some companies have even experimented with placing data centers underwater in the ocean, although maintaining and upgrading submerged hardware presents obvious challenges.
Beyond reducing water consumption
Improved technology will be essential if major technology companies are to meet their sustainability commitments.
Amazon Web Services, for example, has set a goal of becoming water positive by 2030 — meaning it aims to return more water to communities than its direct operations consume.
Google has similarly committed to replenishing 120 percent of the freshwater consumed by its data centers by 2030 through water stewardship and restoration projects.
Companies are also finding ways to reuse the enormous amounts of waste heat produced by data centers. Across parts of Europe, excess heat from cooling systems is being redirected into district heating networks, helping warm homes and buildings while reducing dependence on fossil fuels.
Ultimately, the future of AI's water footprint will depend less on AI itself than on the choices surrounding it.
Where data centers are built, how they are powered, the climate around them and the cooling technologies they use could make an enormous difference.
As Eric Masanet of the University of California, Santa Barbara, puts it, different choices can lead to dramatically different outcomes:
"If you choose one set of choices, your number is going to be off the charts. You choose another set of choices, it's going to be way down here."
The rapid growth of AI does not automatically have to mean an equally dramatic increase in water consumption. But reducing its environmental impact will require smarter infrastructure, cleaner energy and careful decisions about where — and how — the world's next generation of data centers is built.
