UN Report: AI data center environmental impact goes beyond carbon emissions to water and land use by 2030

A new report from the United Nations University argues that measuring AI's environmental impact only by carbon emissions misses the full picture. By 2030, global data centers could consume 945 terawatt-hours of electricity annually, use 9.3 trillion liters of water, and require a land footprint of over 14,500 square kilometers. The report highlights trade-offs where lower-carbon energy sources like bioenergy can increase water use thirtyfold and land use one hundredfold compared to coal. The report also notes that 80% to 90% of AI energy consumption comes from inference (serving user requests) rather than training models. ChatGPT alone processes about 2.5 billion prompts per day, using roughly 383 gigawatt-hours per year. The authors argue that sustainability discussions need to account for water, land, and lifecycle effects alongside emissions, especially as regulatory frameworks in Europe begin to require broader reporting under the Energy Efficiency Directive.