AI water savings claim masks true energy costs
The claim arrives with the confidence of a company that has solved the future: artificial intelligence can now run in data centers that barely drink a drop of water. Nvidia, the American chipmaker that has become the indispensable supplier of the AI boom, announced in a June 2025 report that its newest system for designing and managing AI facilities can eliminate water consumption almost entirely. The statement is precise, technical, and seductively simple. It is also a perfect example of how AI deceives us — not by lying outright, but by presenting a narrow technical truth as a comprehensive solution, obscuring the full picture of what its operations actually consume.
The deception begins with the framing of the problem itself. Data centers are the vast warehouses of computer servers that power the internet and increasingly artificial intelligence, and they have become a flashpoint of public anger across the United States over their appetite for water and power. The computers inside them generate enormous heat, and keeping them cool has traditionally required staggering amounts of water — 222 billion liters globally in 2025, according to the consultancy Rystad Energy. Rystad Energy Without adaptive measures, that figure could nearly triple to 644 billion liters by 2030. Rystad Energy The industry’s response is not to deny the thirst but to promise a technological fix that makes the problem disappear.
Nvidia’s new method uses a closed-loop cooling system where liquid flows directly through the servers, as close as possible to the chips whose temperature can rise above 80 degrees Celsius. The innovation is subtle: by allowing the liquid to enter the servers warmer than usual, at 45 degrees Celsius, the system avoids the need for energy-hungry refrigeration. Most other closed-loop systems in 2024 ran at about 32 degrees Celsius, according to the Uptime Institute, which certifies data centers. Uptime Institute Simple fans circulating the air are often enough to cool the warmer liquid, though a mix of methods is sometimes needed, said Josh Parker, Nvidia’s head of sustainability. Nvidia
This is where the gap between appearance and reality begins to widen. The claim of eliminated water consumption applies only to sites in moderate climates, and even there, the system still requires chill airflow or water evaporation during heat waves. The company’s own documentation acknowledges these limitations, yet the public messaging omits them. In extreme climates, the water-saving promise evaporates entirely. The company has essentially solved the water problem for data centers built in temperate zones during mild weather, while presenting this conditional achievement as a general breakthrough. The public hears “we can eliminate water consumption” and does not register the unspoken qualifier: “in some places, some of the time.”
The deeper deception lies in what the closed-loop system actually does with its energy. Andy Masley, an independent researcher who covers AI and data centers, articulated the fundamental trade-off that the industry rarely mentions in its sustainability announcements: there is a pretty direct trade-off between how much water is used and how much energy is used for temperature control. Nvidia Cutting back on water use usually means more power consumption, because the liquid in those sealed pipes still has to be cooled somehow, typically by blowing air over it, and that requires electricity. The water problem is not solved; it is transferred to a different part of the equation.
This transfer has consequences that the clean narrative of efficiency conveniently ignores. Water is used to generate the electricity that powers data centers, and to manufacture their chips and servers. In the United States, that hidden water use can be twice the amount a data center consumes on its own. A facility that proudly announces it uses no water for cooling is still drinking indirectly through the power grid and the supply chain. The distinction between direct and indirect consumption is technically accurate but practically misleading, creating the impression of environmental responsibility where none fully exists.
The industry’s reporting practices compound the deception. Microsoft, Amazon Web Services, and Meta all told AFP that they use closed-loop systems involving no net water loss Meta. Amazon Web Services But there is no industry-wide consistency in how companies report on so-called environmental, social and governance efforts. The numbers that do exist are self-reported, self-defined, and self-interpreted, making meaningful comparison between companies nearly impossible.
The most telling case is Elon Musk’s SpaceX, now a major player in data centers after acquiring his artificial intelligence company xAI. SpaceX has never published an ESG report SpaceX, and in June, ratings agency MSCI gave it the lowest possible ESG score. MSCI A company that dominates the most consequential technology of the decade can simply decline to participate in the accounting of its own environmental impact, and face no meaningful consequence. The voluntary nature of sustainability reporting means the companies with the most to hide have the easiest path to hiding it.
The incentives for the industry are not aligned with actual conservation. Shaolei Ren, an engineering professor at the University of California, Riverside, noted that because water is generally much cheaper than electricity, companies have less incentive to cut water use on cost grounds alone. University of California, Riverside The incentives that do exist have more to do with public relations amid the growing backlash to data centers across the United States. The goal is not to reduce consumption but to reduce criticism, and a well-crafted sustainability report achieves that goal more efficiently than any actual change in operations.
The economics of upgrading exacerbate the problem. Retrofitting an older data center to newer, less thirsty technology is expensive, and the industry has little reason to invest in facilities that are already paid for. Minh K. Le, who leads data center and hydrogen research at Rystad, offered a counterintuitive perspective: older data centers are smaller and less powerful than the enormous new ones being built now, so they need less cooling in the first place. Rystad Energy The facilities that most need water-saving technology are the newest ones, which are also the ones where the technology is being deployed — but the older ones continue operating with their inefficient systems, unremarked and unreformed.
The public anger that data centers face is not irrational, but it is often directed at the wrong target. Communities protest the visible water consumption of cooling towers and the visible power demands of massive facilities, while the invisible consumption embedded in electricity generation and manufacturing goes unexamined. The industry exploits this asymmetry, presenting solutions to the visible problem while remaining silent on the invisible one. The result is a conversation about sustainability that is systematically distorted by the very actors who are supposed to be held accountable.
The technical achievement of Nvidia’s closed-loop system should not be dismissed. It represents genuine engineering progress, reducing direct water consumption in facilities where it can be deployed. But the gap between what the technology does and what the industry claims it does is the real story. The claim of eliminated water consumption suggests a solved problem; the reality is a complex trade-off between water and energy, direct and indirect consumption, visible and hidden impacts. The technology is real, but the narrative around it is a carefully constructed illusion.
What makes this deception particularly insidious is that it is not malicious. The engineers who designed the closed-loop system genuinely believe they have made progress, and they have. The executives who announce the results genuinely believe they are being transparent, and they are — within the narrow definition of transparency they have chosen. The deception is structural, embedded in the way the industry defines its problems and measures its solutions. Every metric is chosen to show improvement, every comparison is selected to flatter, every boundary is drawn to exclude what cannot be defended.
The water consumption of data centers is technically solvable, and the industry is making real progress toward solving it. But the social problem is not technical, and it will not yield to engineering solutions. Communities that see their water resources depleted and their power grids strained by facilities that create few local jobs and produce nothing tangible are not mollified by efficiency improvements measured in percentages. The gap between what the industry claims and what communities experience is not a data problem; it is a trust problem, and no amount of technical innovation can close it. Until the industry adopts standardized, independently verified reporting that accounts for both direct and indirect consumption, the public will have no reliable way to distinguish genuine progress from carefully framed public relations.
The future of AI depends on public acceptance of the infrastructure that powers it, and that acceptance is eroding precisely because the industry’s communications are optimized for regulatory approval rather than honest engagement. The closed-loop system is a genuine achievement, but it is presented as a solution to a problem it does not fully solve. The water that data centers no longer consume directly is still consumed somewhere, by someone, and the accounting that makes that consumption invisible is the real innovation of the AI era. The technology has advanced; the honesty has not.
Sources
1. Nvidia
5. Meta
6. SpaceX
7. MSCI