Overview
- The United Nations University released a report on Monday projecting that AI and data centers could require about 945 TWh of electricity by 2030, produce up to 400 million tonnes of CO2 annually, use roughly 9.3 trillion litres of water each year, and occupy more than 14,500 square kilometres of land.
- The study warns the gains from more efficient models may be erased by higher overall use, citing a Jevons‑paradox effect where lower costs drive expanded demand for AI services.
- Researchers highlight a stark geographic split: most AI infrastructure and economic gains sit in a few countries while energy, water and land burdens are likely to fall on parts of Asia and lower‑income regions.
- Policy and market responses are multiplying, with regulators reviewing permitting and grid rules and tech firms piloting demand‑flexibility tools such as a reported Google‑Voltus virtual power plant and other load‑shifting programs to avoid building new generation.
- The report calls for lifecycle environmental disclosures, integrated energy‑water‑land planning, and international cooperation because communities face concrete risks to local water supplies, land use and emissions unless policy changes guide where and how capacity grows.