UN study raises alarm over AI’s growing water and energy footprint
The United Nations University Institute for Water, Environment and Health is warning that the rapid expansion of artificial intelligence infrastructure could create major new pressures on global water, energy and land resources — with data center electricity demand projected to nearly double France’s current annual power consumption by 2030.
In a new report examining the environmental cost of AI-driven data center growth, researchers estimated global data centers could consume 945 terawatt-hours of electricity annually by 2030, up from an estimated 448 TWh in 2025. The report also projects associated water use could reach 9.3 trillion liters annually — roughly equivalent to the basic domestic water needs of 1.3 billion people in Sub-Saharan Africa.
The report argues that current discussions around AI sustainability focus too heavily on carbon emissions while overlooking the sector’s growing water and land footprint. Researchers said every unit of electricity used by AI systems also carries indirect water demands tied to cooling systems, power generation and supporting infrastructure.
“Low-carbon is not automatically low-water or low-land,” the report states, warning that some renewable energy pathways can significantly increase water and land intensity even while reducing greenhouse gas emissions.
The study also emphasizes that AI “inference” — the ongoing operation of deployed models — now accounts for an estimated 80% to 90% of total AI energy consumption, far exceeding the energy required to initially train large models. Researchers estimated ChatGPT alone processes approximately 2.5 billion prompts daily, consuming roughly 383 gigawatt-hours of electricity annually.
Water demands vary dramatically depending on the type of AI application being used. The report estimates generating a single AI image requires roughly 1,450 times more energy than basic text classification, while high-complexity AI video generation can consume enough electricity to power a 10-watt LED bulb for 42 hours.
Researchers also warned that efficiency improvements alone may not reduce AI’s overall environmental footprint because lower costs and faster systems tend to drive higher usage — a phenomenon known as the “rebound effect” or Jevons Paradox.
The report calls for governments, utilities, data center operators and technology companies to incorporate water availability, land impacts and infrastructure constraints into AI planning and permitting decisions. Recommendations include standardized reporting for carbon, water and land footprints, expanded community engagement during siting decisions and greater use of water-efficient cooling technologies.
The findings come as water utilities and regulators increasingly evaluate the long-term implications of large-scale data center development on municipal water supplies, wastewater infrastructure and regional resource planning.