A new report from the UN University Institute for Water, Environment and Health reveals alarming projections for artificial intelligence's environmental footprint by 2030. The study, published on Wednesday, provides what researchers call the most comprehensive assessment yet of AI's environmental costs, warning that the technology could consume nearly 3 per cent of the world's electricity, produce carbon emissions comparable to the entire United Kingdom's annual output, use enough water to meet global drinking needs for over a year and a half, and generate electronic waste equivalent to discarding 250 Eiffel Towers each year.
Energy and Carbon Footprint
Data centres powering AI consumed an estimated 448 terawatt-hours of electricity in 2025, roughly equal to France's total national consumption. AI workloads accounted for about 20 per cent of that total. If that share rises to 40 per cent by 2030, AI-related electricity use could reach 374 terawatt-hours. On current trajectories, the report projects it could roughly double to 945 terawatt-hours, enough to power all 1.3 billion people in Sub-Saharan Africa for over five years. The land required to generate that electricity would exceed 14,000 square kilometres, roughly the area of Northern Ireland.
The report notes that most climate impact calculations have centred on carbon emissions, but this tells only part of the story. Cutting emissions alone may not significantly reduce AI's environmental harm. "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."
Water Consumption
Data centres used an estimated 9.3 trillion litres of water in 2025 for cooling, a figure that would meet the drinking water needs of the world's 8.1 billion people for approximately 1.6 years. 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 drawing heavily on water supplies during a drought year prompted opposition from local farmers.
Training and Daily Use
Training a single large AI model such as ChatGPT-5 requires around 100 gigawatt-hours of electricity, equal to the annual residential power consumption of 770,000 people in Sub-Saharan Africa, along with an estimated one billion litres of water and a land footprint covering roughly 215 football fields. However, the report found that the environmental cost of daily use has overtaken training costs. ChatGPT alone processes an estimated 2.5 billion prompts per day. A conventional Google search uses about 0.3 watt-hours of electricity, while an AI-enhanced generative search uses up to 3 watt-hours, a tenfold increase, applied across an estimated 5 trillion searches a year.
User choices significantly affect these numbers. Switching to a concise response mode can reduce ChatGPT's output by 30 per cent, saving 87 to 98 gigawatt-hours of electricity per year, equivalent to the annual residential electricity of nearly 760,000 people in Sub-Saharan Africa. Removing pleasantries makes prompts more concise and reduces the cumulative footprint at scale.
AI Video Generation
The rising popularity of AI-generated videos is becoming a major environmental concern. A single high-resolution AI video clip requires more than 415 watt-hours of electricity, more than generating hundreds of AI images. As resolution and frame count increase, energy requirements rise exponentially. Video generation has become embedded in mainstream social media platforms, with sites encouraging users to create and post more AI videos as part of viral trends. The report warns this is becoming an infrastructure-scale problem.
Renewable Energy Not a Panacea
The report found that powering data centres with renewable energy does not automatically make them sustainable. Switching from coal to bioenergy can reduce the carbon footprint of electricity generation by 72 per cent, but the water footprint of bioenergy is on average more than 30 times greater than that of coal, and its land footprint is 100 times greater. For example, Brazil's hydro-powered grid produces electricity 77 per cent below the global carbon average, but its water and land footprints are nearly triple the global mean.
Infrastructure Growth Outpacing Planning
In Ireland, data centres now account for 21 per cent of the country's total metered electricity, up from 5 per cent a decade ago, exceeding all urban household consumption combined. The national grid operator has paused new approvals around Dublin until 2028. Professor Alistair Knott of the Centre for Data Science and AI at Victoria University of Wellington, who was not involved in the report, said while the study calls out growing investments by AI companies, it fails to note that AI companies depend on increased market growth for survival. "The only way companies can survive is to grow the market for AI products at an ever-increasing pace, but that's not necessarily what the world needs," he said. "Governments, elected by citizens, are better placed to make the right decisions about how much AI we need, and to trade this need off against environmental impacts."
Professor Te Taka Keegan of the AI Institute at the University of Waikato, also not involved, said the concentration of infrastructure raises environmental justice concerns. "The environmental burden falls hardest on communities least likely to capture the benefits," he said. "As AI is embedded into everyday platforms and switched on by default whether users choose it or not, that footprint compounds at scale."
Recommendations
Researchers urge governments to start factoring AI infrastructure into water and energy planning. Tech companies should also include an environmental lens in planning for new features. "Technological advancement must remain environmentally manageable," the authors write. "Real progress depends on embedding sustainability at every level, from hardware and model design to deployment, governance, and public use."
