Artificial intelligence could consume nearly 3% of the world's electricity by 2030, produce carbon emissions comparable to the entire UK's 2025 output, and use enough water to meet the drinking needs of every person on Earth for over a year and a half, according to a new report from the United Nations University Institute for Water, Environment and Health.
The report, published on Wednesday, provides what it calls the most comprehensive assessment yet of AI's environmental costs. It warns that focusing solely on carbon emissions tells only part of the story. “Low-carbon is not automatically low-water or low-land,” the report states, “and evaluating sustainability through a single metric can hide trade-offs and shift burdens onto places already facing water stress or land pressure.”
Data centres consumed an estimated 448 terawatt-hours of electricity in 2025, roughly equivalent to France's entire national consumption. AI workloads accounted for about 20% of that total, but could rise to 40% by 2030, pushing AI-related electricity use to 374 terawatt-hours. On current trajectories, total data centre electricity use could double to 945 terawatt-hours, requiring over 14,000 square kilometres of land – roughly the area of Northern Ireland – to generate that power.
Water consumption is also a major concern. Data centres used an estimated 9.3 trillion litres of water in 2025, enough to meet the drinking water needs of the global population for more than 18 months. Even where water is returned to the environment, large-scale withdrawals strain aquifers and river systems, particularly in water-stressed regions. In the Netherlands, a large data centre's heavy water use during a drought prompted opposition from local farmers.
Training a single large AI model like ChatGPT-5 requires around 100 gigawatt-hours of electricity, one billion litres of water, and land equivalent to roughly 215 football fields. However, the report finds that the environmental cost of daily use now exceeds that of training. ChatGPT alone processes an estimated 2.5 billion prompts per day. An AI-enhanced generative search uses up to 3 watt-hours per query, ten times more than a conventional Google search. The report notes that user choices can significantly reduce this impact: switching to concise response mode can cut ChatGPT's output by 30%, saving up to 98 gigawatt-hours per year.
The rising popularity of AI-generated videos is a growing environmental concern. A single high-resolution AI video clip requires over 415 watt-hours of electricity, more than hundreds of AI images. As video quality improves, energy requirements rise exponentially. The report warns that AI video generation has become embedded in mainstream social media, creating an infrastructure-scale problem. Professor Alistair Knott of Victoria University of Wellington, who was not involved in the report, said it highlights growing investments in AI and their environmental consequences.
