UN report warns AI could use water equal to 1.3 billion people by 2030

A United Nations University Institute (UNU‑INWEH) report projects that by 2030 the water used by artificial‑intelligence (AI) data centres will match the annual consumption of about 1.3 billion people in sub‑Saharan Africa. Electricity demand of AI‑powered computing is expected to reach 945 terawatt‑hours per year – roughly three times the combined annual consumption of Pakistan, Bangladesh and Nigeria, and comparable to the total electricity use of Japan. The study notes that 90 % of AI’s energy use comes from inference, not model training, and that a single ChatGPT query requires about 0.15 kWh, while generating an AI image consumes 1 400‑times more energy.

The report estimates AI‑related CO₂ emissions could hit 400 million tonnes, similar to the United Kingdom’s total output, and predicts 2.5 million tonnes of e‑waste annually, largely ending up in low‑income countries. More than 90 % of global AI compute capacity is concentrated in the United States and China, creating a geographic imbalance where the environmental burdens fall on nations with little AI benefit. Kaveh Madani, director of UNU‑INWEH, stressed that the findings are “not an accusation against AI, but a call to use it responsibly and to account for its unintended impacts.” The report urges governments to require standardized environmental‑footprint reporting and for the industry to design AI systems for efficiency by design.