The AI buildout has found another resource it cannot simply order in bulk: water. Data center water use is becoming a public pressure point as cloud providers and AI companies race to build larger campuses for more powerful computing systems, while communities, investors and regulators ask a fairly basic question: how much local supply is the cloud allowed to drink?

That question used to live mostly inside engineering teams. Now it shows up in city meetings, sustainability reports and investor letters. Data centers run the systems behind AI tools, cloud software, streaming, finance, medical records and the rest of the internet’s invisible machinery. They also produce enormous heat, which means operators must choose cooling systems that often force a trade: save electricity and use more water, or save water and use more power. Conveniently, there is no magic option marked “no consequences.”

Why AI data centers are putting pressure on water supplies

The scale of the buildout is what has turned a technical issue into a political one. The International Energy Agency has projected that global electricity use from data centers could more than double to about 945 terawatt-hours by 2030. That rising power demand lands alongside water demand, especially as facilities grow to support AI workloads that require dense clusters of high-performance chips.

The most visible water use comes from evaporative cooling. In that setup, water removes heat by evaporating in cooling towers. It can be efficient from an electricity standpoint, which is why it has been so widely used. The catch is not subtle: it can consume large volumes of water, especially in hot weather, when households, farms and local businesses may also need more.

Alternatives exist. Dry cooling and air cooling can greatly reduce direct water consumption, but they can require more electricity. Closed-loop and liquid cooling systems can reduce evaporation, particularly in newer designs, but they are not equally easy to install everywhere. The result is a local calculation, not a universal slogan.

A February 2026 report from the University of California, Berkeley’s Center for Law, Energy & the Environment put it plainly: data center water and energy use are “intertwined.” More energy-efficient cooling often uses more water, while water-saving systems often draw more power.

Google’s water footprint has become a test case

Google is one of the most closely watched companies in this debate, partly because it publishes more site-level water information than many competitors and partly because its facilities have become a focal point in local politics.

In The Dalles, Oregon, Google’s water use has repeatedly drawn public scrutiny. Oregon Public Broadcasting reported in January 2026 that, by 2024, roughly one-third of the city’s water went to Google’s three local data center sites. A follow-up report said the company’s annual water use in The Dalles rose from 104 million gallons in 2012 to 434 million gallons in 2024.

Those numbers are why “water stewardship” now has to mean more than polished charts in a sustainability PDF. Google’s 2025 Environmental Report said the company replenished 4.5 billion gallons of water in 2024. It also said replenishment increased from 18 percent of freshwater consumption in 2023 to 64 percent in 2024.

Google says it evaluates watershed risk before deciding whether to use freshwater for evaporative cooling. The company also reported that 86 percent of its data center freshwater withdrawals came from sources at low or medium risk of depletion or scarcity.

On Wednesday, June 3, 2026, Google went further by releasing a water-use framework that it said should help guide the broader data center industry. The timing was not mysterious. Backlash over AI infrastructure siting has been growing, and residents in some areas are asking whether developers are moving faster than local utilities and governments can assess the long-term effects.

Axios reported that about two-thirds of Google’s data centers still use evaporative cooling. The rest use air cooling or recycled and non-conventional water sources. Google says water concerns are increasingly shaping decisions in regions including India and the American Southwest.

Microsoft, AWS and Meta are changing cooling plans

Google is hardly alone. Microsoft, Amazon Web Services and Meta are all trying to show that the next wave of AI infrastructure can grow without turning local water politics into a permanent fire alarm.

Microsoft has pledged to become water positive by 2030. The company says it has cut water intensity in its operational data centers by 18 percent compared with its 2022 baseline. Its newer cooling approach uses liquid-to-chip cooling in a closed loop, designed to eliminate evaporation and reduce reliance on fresh water for cooling.

Microsoft also points to a reuse partnership in Quincy, Washington. The company says the project reduced its potable water use in that region by 97 percent and provides 1.5 million cubic meters of water each year for community drinking water needs.

Amazon Web Services is emphasizing efficiency and reclaimed water. Amazon says AWS’s average water use effectiveness in 2024 was 0.15 liters per kilowatt-hour, a 40 percent improvement since 2021. The company also said AWS was 53 percent of the way toward its goal of becoming water positive by 2030.

AWS says 24 of its facilities use 100 percent reclaimed water, with 130 more facilities globally contracted to use reclaimed water. Its argument is that treated wastewater can reduce pressure on potable supplies, assuming the needed treatment systems and local agreements are in place.

Meta is also leaning into closed-loop liquid cooling and dry coolers for some next-generation AI facilities. The company says its typical one-gigawatt AI-optimized data center design, expected to begin operating later in 2026, uses a closed-loop liquid cooling system with dry coolers. In that setup, Meta says there is no operational water use for cooling, with site water use mainly limited to domestic, janitorial, cleaning and fire-protection needs.

Meta has set a goal to become water positive by 2030 by restoring 200 percent of consumption in high-water-stress regions and 100 percent in medium-water-stress regions.

The cooling toolkit is growing, but every option has limits

The industry is not ignoring the problem. Operators are experimenting with and deploying a wider mix of water-saving methods, including:

  • Recycled wastewater for cooling
  • Rainwater harvesting
  • Closed-loop cooling systems
  • Dry cooling and air cooling
  • Direct-to-chip liquid cooling
  • AI-driven cooling optimization
  • Watershed restoration projects

That list sounds reassuring until the fine print arrives, as it usually does.

Recycled water depends on local treatment infrastructure. If a city does not have the pipes, processing capacity or agreements in place, the idea may remain more elegant than practical. Dry cooling can reduce water use but increase electricity demand, which matters in regions already struggling to connect new data centers to the grid.

Closed-loop cooling can sharply reduce water consumption, but it may be easier to design into new facilities than retrofit into older campuses. Direct-to-chip systems can be powerful, especially for dense AI hardware, but they require careful engineering and capital investment. Watershed restoration can improve long-term water health, but it may not replace water in the same place, at the same time or under the same drought conditions as a data center’s withdrawals.

That last point is central. Water is local in a way that carbon accounting often is not. A company can claim progress across a global portfolio while one town still feels the strain from one very thirsty facility.

Why “water positive” claims are under scrutiny

The phrase “water positive” has become a standard promise across the sector. In theory, it means a company replenishes or restores more water than it consumes. In practice, the claim can raise harder questions than it answers.

If a facility draws heavily from a stressed municipal system or aquifer, a restoration project elsewhere may not reassure residents living near the site. Timing also matters. Replenishing water during one season does not necessarily solve peak demand during a heat wave or drought. Geography matters too. A benefit in one watershed does not automatically offset harm in another.

The UC Berkeley report warned that current reporting requirements do not give the public a clear picture of how much water data centers use, where impacts occur or how different water sources compare. It recommended more transparency, including site-specific water data in corporate sustainability reports, so local governments and residents can evaluate projects before approvals become difficult to reverse.

Investors are starting to press the same point. In April 2026, more than a dozen investors urged Amazon, Microsoft and Alphabet’s Google to provide more detailed information on water and power use at U.S. data centers, according to Reuters reporting republished by MarketScreener. Their concern was not only environmental. They argued that water stress and community opposition are becoming material business risks as AI infrastructure expands.

The next fight over AI infrastructure will be local

The next phase of the data center water debate will probably not be decided by a single global standard. It will be decided town by town, watershed by watershed, permit by permit.

A cooling system that makes sense in a wet, cool region may be unacceptable in a drought-prone basin. A facility using reclaimed wastewater may face less resistance than one drawing potable water from a municipal supply. A company that publishes annual global water figures may still meet distrust if residents cannot see the numbers for the site near their homes.

For operators, the message is getting difficult to ignore. Water efficiency is no longer just an engineering metric buried in a facilities spreadsheet. It is part of the industry’s permission to build.

The companies best positioned for the AI boom will be the ones that can pair rapid expansion with credible local planning, transparent disclosure and cooling systems matched to regional conditions. Those that cannot may discover that the next AI bottleneck is not chips, land or electricity. It may be public confidence in how much water the cloud gets to consume.