New UN Report Warns Data Centers Could Consume More Water Than the World’s Population by 2030
AI data centers are thirsty facilities. How thirsty? According to a UN report published this week, if current trends continue, by 2030 data centers will consume more drinking water than the entire global population.
The reason is that the enormous computing power generated by data centers produces significant heat emissions. That heat must be cooled to prevent performance damage or destruction of sensitive components, and water is the preferred solution. Water consumption at a typical data center can easily reach millions of liters a day, similar to a city of 50,000 residents. As a result, building a data center can significantly harm the availability and quality of drinking water in nearby communities. The arms race to build AI infrastructure has only intensified this problem over the past three years, as more and more communities have faced disruptions to water supply. Tech and AI giants are responsible for a significant share of the problem, which is beginning to threaten the pace of development of these centers. In an amendment to the prospectus it filed last week, SpaceX, owned by Elon Musk, noted that water-related issues, including shortages, water regulations and droughts, could limit the pace at which it builds data centers.
80% evaporate into the atmosphere
The main use of water is to cool servers. One common technique is to use drinking water to absorb heat, then pump it to cooling towers, where 80% of it evaporates into the atmosphere and the rest is sent to a wastewater treatment facility. Increasing the amount of water used for cooling can help reduce data center energy consumption, another sensitive issue facing AI giants, but it comes at a cost of its own. In 2024, one Google data center in Iowa used 4.5 billion liters of water. That same year, Lawrence Berkeley National Laboratory estimated that if data center operators continue to rely on water for cooling, consumption could reach 150 billion liters.
The concern is also affecting Americans’ attitudes toward data centers. According to a Gallup poll published in May, 71% oppose building a data center in their area, an unusually bipartisan objection. Water consumption, along with energy consumption, was the leading reason for opposition. The situation is especially difficult in areas already suffering from water problems, since they are often preferred locations for AI infrastructure because of easy access to available energy resources and more lenient regulation.
“Water is a very local issue,” Prof. Shaolei Ren of the Department of Electrical and Computer Engineering at the University of California, Riverside, told Wired. “It is a limited resource, and we have to manage it very carefully.” Now, as states and local authorities in the United States promote moves and decisions to stop or ban data center construction on their territory, the companies building these facilities are being forced to contend with residents’ concerns.
Info grinds water
Microsoft, one of the three largest cloud service providers alongside Amazon and Google, has begun implementing a new cooling technology that significantly reduces water consumption. The technology uses a closed-loop architecture that reuses the same water supply to cool servers. The system is filled with water during the construction of the data center. Cold water flows between the data center servers, absorbs heat, is routed to a cooling plant where it is cooled by giant fans, and then flows back to the data center. This method makes it possible to provide 90% of cooling needs. The rest is cooled by outside air, with fresh water used as a backup in hot weather.
“Once the cooling loop is filled, the data center can operate with zero water consumption,” Microsoft CEO Satya Nadella explained at the developers conference last week. “Annual water use over a full year is roughly equivalent to the use of one restaurant.” Depending on the size and type of restaurant, its water consumption can range from hundreds of thousands of liters to more than 4.5 million liters a year. Even at the upper end, that is still a significant saving in water use for a data center.
At this stage, the closed-loop technology is being used in only one Microsoft data center in Wisconsin. But the company has promised that it will become part of the standard design, and that it is already building data centers in the United States that use the method. However, its existing data center network includes more than 500 facilities in 80 regions around the world that use evaporative cooling technology, and the company has not announced a plan to upgrade them. This technology also has one major drawback, its energy consumption is significantly higher than that of closed-loop technology.
In January, OpenAI also committed to reducing water use through closed-loop systems or low-water-use technologies. “The water consumption of our facilities should be a fraction of the community’s total water use,” it said. Those reassuring words were shaken a few weeks later, when OpenAI founder and CEO Sam Altman dismissed concerns about the energy and water consumption of data centers. He waved away energy consumption by saying that “it also takes a lot of energy to train a human.” He called concerns about water consumption “fake,” and said claims that ChatGPT uses several liters of water per query are “detached from reality.”
Oracle, which is working with OpenAI on the Stargate project, unveiled a more sophisticated approach to closed-loop systems in February, a closed loop directly at the chip level. “Instead of cooling the air in the room, we remove the heat closer to where it is generated, in the servers themselves,” the company said. It said it is already incorporating the technology into data centers it is building in New Mexico, Michigan, Wisconsin and Texas.
Transparency about water consumption
Google is taking a different approach. In an announcement published last week, the company outlined five commitments to managing water resources in places where it builds and operates data centers. Google has pledged to return to communities more drinking water than the volume consumed by its facilities by 2030. The plan will include improving and upgrading water infrastructure, implementing air-cooled solutions in water-stressed areas, providing transparency about its annual water consumption, and identifying alternative solutions to protect water sources, such as using wastewater.
According to the company, it is already building 165 projects to improve the availability and accessibility of drinking water. Last year it returned 32 billion liters of drinking water to communities around the world, and when all projects are completed by 2030, it will return more than 86 billion liters to communities. That amount is twice Google’s water consumption in 2024, although the company did not say what its expected consumption will be in 2030. At the same time, Google will not completely stop using water evaporation systems.
“Data center design is much more complex than promising not to use one type of cooling,” Ben Townsend, the company’s head of infrastructure and sustainability, told Wired. “A one-size-fits-all strategy simply doesn’t work,” he stressed. In an amendment to the prospectus it filed last week, SpaceX said that water-related issues, including regulation and droughts, could limit the pace at which it builds data centers.
Townsend’s remarks are not unfounded. Closed-loop technologies may save water, but their energy consumption is higher than that of evaporative or air cooling systems. According to research by Ren, if all data centers in the United States used evaporative technology, up to 30 gigawatts of energy could be saved, the electricity consumption of about 30 million households. This is a critical figure for areas that struggle to meet their regular electricity needs, against the backdrop of the energy consumption of data centers around them.
“If you do not use water, the challenge is that you will use much more energy in the summer, and that will increase costs,” Ren said. And not just costs, but also greenhouse gas emissions, if data centers do not use renewable energy sources. The AI revolution has thrown the climate commitments of tech giants in the trash, as their carbon emissions soar. A complete avoidance of evaporative cooling would lead to continued emissions growth, unless companies can rapidly develop renewable energy sources.
One possible solution is to pressure companies to design better data centers that are efficient in both water and energy use. “We need to challenge the industry to design things that are smarter and simpler,” said Priscilla Johnson, formerly Microsoft’s director of water strategy. The main way to do this is through public pressure and regulation. But with an erratic president in the White House and an administration that promotes an anti-regulatory approach, especially toward AI, effective and sustained federal pressure does not seem likely. As a result, local governments and citizens will have to face companies with the market value of a mid-sized country on their own, while their willingness to do the right thing tends to evaporate as soon as the bottom line is threatened.
