WASHINGTON (AP) – The environmental footprint of data centers already rivals that of some countries around the world, and water and energy use and pollution are expected to double in just four years as the use of artificial intelligence increases, according to a United Nations University report.
Last year, the world’s data centers used 448 trillion watt-hours of electricity, more than all but 10 countries in the world, according to a report released Wednesday. A report on the environmental impact of AI’s energy use found that its electricity use produced about 208 million tons (189 million tons) of carbon dioxide, about the same amount as Argentina, and that producing the same amount of energy consumed about 1.2 trillion gallons (4.5 trillion liters) of water.
By 2030, data centers will account for nearly 3% of the world’s projected electricity usage, amounting to 935 trillion watt-hours. If a data center were a country, it is predicted that by 2030, that country would be the 6th largest electricity user in the world. In that case, about 440 million tonnes (399 million tonnes) of carbon dioxide would be produced, the report said. The study focused on energy use and did not examine the large amounts of water used to cool data centers.
“When you look at these numbers, you’re looking at a scale comparable to that of a nation,” said study co-author Kaveh Madani, a water scientist and director of the United Nations University Institute for Water, Environment and Health in Canada. “The demand is huge.”
Much of the growth in data centers is being driven by AI. Currently, around 20% of data center energy comes from AI, which should rise to 40% by 2030, according to the report.
First global study of ecological impacts
Fengqi Yu, a professor of energy engineering at Cornell University who directs the university’s AI sustainability initiative, said the report is important not just because of its eye-popping numbers, but also because of the credibility and authority of the United Nations.
“The value is that the U.N. agency brings together carbon, water, land, life cycle impacts and environmental justice in one framework,” said Yu, who was not involved in the report, about issues that are often shrouded in secrecy and partial disclosure.
“The public should be concerned but not panic,” he added.
Jean Hsu, director of the Energy Justice Program at the Center for Biological Diversity, said the report is important because it is the first UN, or even global, report to “shine a light on the environmental harm caused by AI.”
Caleb Max, president of the National Association for Artificial Intelligence, highlighted how his industry is becoming more efficient and how it is benefiting the nation, saying, “AI is rapidly becoming a part of our daily lives, delivering benefits such as increased safety, longer lives, more efficient work, enhanced food production, and reduced poverty.Evidence is growing every day that the energy return on investment in AI development is transformative for our world and therefore worth it.”
Josh Levi, chairman of the Data Center Coalition, said the industry is taking its environmental impact seriously.
“We remain committed to working with policymakers, communities, and industry partners to ensure that as data centers grow, they do so in ways that are responsible, transparent, and reflect available best practices,” he said in a statement.
How much energy does a query use and how to reduce it?
Madani, who is also the latest recipient of the Stockholm Water Prize, said the figure shows the environmental cost of AI, which at first glance may seem cleaner than other mechanical equipment such as cars or furnaces that have visible pollution.
“AI is not just a virtual thing. We’re talking about something physical, something that actually has an impact. There’s infrastructure there, there’s energy being used,” Madani said. “There’s a lot of hardware behind all these operations, and it looks very clean to us because we don’t see any smoke coming from our devices. On our phones, we don’t see any visible smoke or smoke coming from computers or anything like that. But someone somewhere is suffering.”
Madani said AI could reduce its huge energy consumption if people asked questions less politely and more concisely. The report found that reducing the use of words in requests by 30% can reduce the energy used by AI by 25%. This would save approximately the same amount of electricity that approximately 700,000 people in Africa use in a year, the report said.
“If you’re too polite, that extra ‘ask’ can make a big difference,” Madani says. “It has to be very precise and short.”
A typical ChatGPT-style query consumes approximately 200 times more energy than the type of basic text classification used in email spam filters, for example. Images and videos generated by AI require more energy.
And the more complex the AI, the more energy it requires to train and learn. GPT-3 used about 1.3 billion watt-hours for training, while the next version used between 50 billion and 70 billion watt-hours, according to the report.
But study co-author Miriam Akzel, an environmental policy researcher at the United Nations National University, said it’s not the training that actually enjoys power. About 90% of AI’s power usage comes from operational requests, she said. GPT alone sends 2.5 billion prompts per day, she said.
Increased efficiency still requires increased power usage
Technology advocates can argue that machines are becoming more efficient, Madani said, but a common paradox is that when things become more efficient, machines are used more frequently and total energy use spikes, even as individual usage efficiency increases. Some companies are touting the use of renewable energy in data centers, Madani said, but that means the supply of clean electricity is depleted and the dirty energy is used elsewhere.
One problem with conducting this study is that many companies and locations are not transparent about what their data centers and AI are consuming, or even its location and scale, Akzel and Madani said.
“You can’t control what companies don’t disclose,” said Cornell University’s You.
(Copyright 2026 The Associated Press. All rights reserved. This material may not be published, broadcast, rewritten or redistributed.)
