Overview
- The United Nations report, published Wednesday, June 3, 2026, estimates AI data centers could produce about 400 million tonnes of CO2 by 2030, use roughly 945–1,000 terawatt-hours of electricity within four years and consume about 9.3 trillion litres of water for cooling and operations.
- Only 16% of countries host specialized AI infrastructure and the United States and China account for up to 90% of installed capacity, concentrating both the economic benefits and the environmental burdens in a few places and companies.
- The UN has also launched EnvironmentGPT this week, a UNEP-based generative tool that shows the environmental footprint tied to queries to make emissions, energy and water impacts more visible.
- Experts say the report gives useful, near-term projections but that the outcomes could change if firms adopt more efficient hardware, run models on devices, switch to renewables, deploy water-saving cooling or if regulators limit growth; they also warn greater efficiency can raise total demand through wider use.
- The report links to human impacts already seen in some regions where data centers strain local grids, risk rising e-waste and could shift water stress onto communities that do not use the services, a dynamic human-rights groups say should shape tighter governance and possible limits on commercial generative AI.