UN Report: AI Data Centers' Environmental Footprint Rivals Nations, Doubling by 2030

A new report from the United Nations University reveals that the environmental footprint of global data centers already rivals that of some of the world's largest countries, with energy and water use expected to double by 2030 as artificial intelligence expands. The report, released Wednesday, found that data centers consumed 448 trillion watt-hours of electricity last year—more than all but 10 countries—producing about 208 million tons of carbon dioxide, comparable to Argentina's emissions, and using approximately 1.2 trillion gallons of water.

By 2030, data centers are projected to account for nearly 3% of global electricity use, reaching 935 trillion watt-hours, and if they were a country, they would rank sixth-highest in power consumption. That would generate nearly 440 million tons of carbon dioxide.

The study focused on energy use and did not examine water used for cooling. Co-author Kaveh Madani, a water scientist at the United Nations University Institute for Water, Environment and Health in Canada, said, "If you look at these numbers, we're seeing scales comparable to nations.

The demand is enormous." About 20% of data center energy is currently driven by AI, but that share is expected to grow to 40% by 2030. The report is the first global assessment of AI's environmental harms, according to Jean Su of the Center for Biological Diversity.

Industry representatives, including Caleb Max of the National Artificial Intelligence Association and Josh Levi of the Data Center Coalition, emphasized efficiency improvements and commitment to responsible growth. The report also noted that users can reduce AI's energy appetite by being less polite and more concise in queries—cutting word use by 30% can reduce energy by 25%, saving electricity equivalent to what 700,000 people in Africa use annually.

Madani said, "If you're too polite, then that extra 'please' you put there can make a huge difference." AI-generated images and video require much more energy, and while training models like GPT-3 used 1.3 billion watt-hours, newer versions use 50 to 70 billion watt-hours. However, about 90% of AI's power use comes from operational requests, with GPT alone handling 2.5 billion prompts daily.

The report highlighted a paradox: as machines become more efficient, they are used more, increasing total energy use. It also criticized a lack of transparency from companies about data center consumption.

Cornell University professor Fengqi You said, "We cannot manage what companies do not disclose."