The next time an AI chatbot answers in seconds, the cost may not show up on your electric bill, but it is showing up somewhere, on power grids, in water systems, across land permits, and in communities asked to host the data centers that make artificial intelligence feel instant.
At London Climate Action Week on June 23, 2026, UN Secretary-General António Guterres called on major AI companies to disclose the full environmental impact of their systems, including carbon, water, and land footprints, and to power every data center with renewable energy by 2030. His message was blunt. The cloud is not weightless.
AI’s hidden footprint
Guterres proposed what he called the AI Environmental Transparency Initiative, a push to make the industry measure and publish the resource costs behind its systems. In his words, “No more hidden costs” and “It is time to come clean.”
The call comes after a United Nations University report warned that AI’s environmental costs are still being badly mismeasured. Most public debate focuses on carbon emissions, but the report argues that every kilowatt-hour used by AI also carries a water and land footprint.
Data centers are becoming country-scale power users
The numbers are hard to shrug off. Global data centers consumed an estimated 448 terawatt-hours of electricity in 2025, or 448 billion kilowatt-hours. If they were treated like a country, they would have ranked 11th in the world for electricity consumption, behind France and ahead of Saudi Arabia.
By 2030, the report projects that global data-center power demand could reach 945 billion kilowatt-hours, almost 3% of projected global electricity use. This is the kind of demand that can shape where power plants are built, how grids are upgraded, and who ends up paying when the system gets stretched.
Water and land matter too
This is where the AI debate gets more complicated. A data center may run on cleaner electricity and still place heavy pressure on water supplies or land, especially in regions already dealing with drought, heat, or fast-growing demand.
The United Nations University report estimates that projected 2030 data-center electricity use could carry a water footprint of 9.3 trillion liters, or about 2.46 trillion gallons. Its associated land footprint could exceed 14,500 sq. km, or roughly 5,600 sq. mi. This is not a small shadow for a technology many people experience as a simple box on a screen.
The everyday AI habit adds up
For years, the public conversation focused on the power needed to train large AI models, but according to the report, the bigger burden comes after deployment, when millions or billions of people use those systems every day.
Inference, meaning the routine running of AI models to answer prompts, is estimated to account for 80% to 90% of total AI energy use. A typical chat query uses far less than a video-generation request, but at global scale even small digital habits can become a serious infrastructure issue.
That does not mean people should stop using AI. It does mean product defaults matter. A shorter answer, a lighter model, or a lower-resolution output can change the footprint, especially when repeated at massive scale.
Methane is part of the same warning
Guterres did not stop with AI. In the same speech, he launched a global call to action on methane, the powerful greenhouse gas released from energy systems, agriculture, and waste. He warned that methane is responsible for around one-third of global warming and is far more powerful than carbon dioxide over the short term.
He also pointed to one startling figure from the fossil fuel sector. In 2025, about 167 billion cubic meters of gas were flared into the sky, equal to roughly 5.9 trillion cubic ft. That is gas burned off instead of used, captured, or kept out of the atmosphere.
Fixing leaks could be surprisingly practical
The International Energy Agency says around 70% of methane emissions from fossil fuels, nearly 85 million metric tons or about 94 million U.S. tons, could be cut with existing technology. More than 35 million metric tons, or nearly 39 million U.S. tons, could be avoided at no net cost based on average 2025 energy prices.
Why no net cost? In many cases, the captured methane can be sold or used. That makes leak detection, equipment upgrades, and limits on routine flaring less like a sacrifice and more like cleaning up a leaky pipe before it floods the basement.
Communities need the full picture
For communities near data centers, the issue is not abstract. It can mean more transmission lines, more water demand, more backup power, and more pressure during that sticky summer heat we all know. When the grid strains, people feel it in bills, reliability, and local planning fights.
Guterres said communities are often left in the dark about the environmental impact of infrastructure rising around them. That is the heart of the transparency push. People cannot judge trade-offs they are not allowed to see.
AI can still help the climate
None of this makes AI the villain. Used wisely, it can help model extreme weather, improve energy efficiency, speed up scientific research, and support smarter planning. The UN’s point, however, is that climate tools also need climate accounting.
At the end of the day, what Guterres is asking for is not especially mysterious. Measure the footprint. Publish the data. Power the system with renewables. Make sure the communities hosting the infrastructure are not simply carrying the costs while the benefits flow somewhere else.
The official speech was published on United Nations Climate Change.
