Artificial intelligence could require enough water to meet the annual household needs of 1.3 billion people and consume electricity comparable to that used by more than 650 million people by 2030, according to a new United Nations University report.
The study by the United Nations University Institute for Water, Environment and Health (UNU-INWEH) estimates that AI-powered data centers will consume 945 terawatt-hours (TWh) of electricity each year by the end of the decade. That is nearly three times the combined annual electricity consumption of Pakistan, Bangladesh and Nigeria.
Researchers said the environmental impact of AI should not be measured by carbon emissions alone. Electricity used by AI systems also carries significant water and land costs through cooling systems, electricity generation and the infrastructure needed to support data centers.
The report projects that AI’s electricity demand will require a land footprint of more than 14,500 square kilometers by 2030, roughly double the size of the Jakarta metropolitan area.
It also found that once AI models are deployed, inference—the process of responding to users’ requests—accounts for 80 to 90 percent of total AI energy consumption, making day-to-day use far more resource-intensive than training the models.
According to the report, ChatGPT processes an estimated 2.5 billion prompts every day, using around 383 gigawatt-hours (GWh) of electricity annually.
The researchers said the environmental impact varies sharply by task. AI image generation requires far more electricity than basic text-processing tasks, while generating short AI videos consumes significantly more energy than text or image queries.
The report also highlighted the rebound effect, where improvements in AI efficiency lower costs and encourage wider adoption, ultimately increasing total electricity, water and land consumption instead of reducing it.
UNU-INWEH said policymakers and technology companies should assess AI using multiple environmental indicators rather than focusing only on greenhouse gas emissions, warning that the rapid expansion of AI infrastructure could place growing pressure on natural resources worldwide.
