Global data centers powering artificial intelligence are projected to consume 945 terawatt-hours of electricity in 2030, more than twice the 2025 consumption of 448 TWh, researchers at the United Nations University said Wednesday.
"Data centers are becoming country-scale consumers of electricity, water and land," according to a university analysis, which notes that the infrastructure would have been the world's 11th largest electricity consumer in 2025, behind France and ahead of Saudi Arabia.
The researchers projected that data centers will carry a water footprint of 9.3 trillion liters by 2030, mainly from cooling and power generation. Associated land footprint is estimated at over 14,500 square kilometers, from energy infrastructure and supply chains.
The report highlighted that AI's environmental cost is "systematically mismeasured," as current focus is only on carbon emissions.
"If we keep judging AI sustainability by carbon alone, we might think that renewables make AI infrastructure clean, but that is solving one problem while creating other problems, often in places that didn't ask for it," said UNU researcher Miriam Aczel.
ChatGPT alone, as noted in the report, processes around 2.5 billion prompts per day, which translate to about 383 gigawatt-hours of electricity per year.
"Per-query energy varies by orders of magnitude depending on the task," the report said, with a typical conversational chat query requiring 200 times as much energy as basic text classification. A single AI image, meanwhile, would require around 1,450 times that baseline.
The analysis also highlighted that AI computing is geographically concentrated, with only 32 countries hosting AI-specialized data centers, 90% of which are located in just two countries. More than 150 countries, on the other hand, have little to no access to AI computing.
In Ireland, the national grid operator has reportedly paused new approvals for AI infrastructure around Dublin through 2028, due to surging power demand. Data centers accounted for 21% of total metered electricity in 2023.
In Mexico and Uruguay, water supply concerns have arisen as high water use by data centers coincides with drought.
By 2030, electronic waste from AI infrastructure could total 2.5 million metric tons per year, with low-income economies processing most of it due to a lack of restrictions, according to the researchers.
Critical minerals, meanwhile, are extracted from regions with weak environmental oversight, the report said.