A new United Nations University report warns that AI data centers could consume 945 terawatt-hours of electricity by 2030. The study highlights a critical trade-off where switching to low-carbon energy sources can significantly increase water and land usage, shifting environmental burdens to stressed regions. While most discussions focus on the energy used to train models, the report finds that inference, or the daily running of AI, accounts for up to 90 percent of total energy use. This growth risks outpacing grid capacity and depleting local freshwater reserves in areas like Mexico and Uruguay.
