AI to double data centre power and water use by 2030 - UN

TL;DR:

  • UN researchers project data centres will consume twice as much power and water by 2030 as AI demand surges.
  • Last year data centres used 448 TWh of electricity — more than Saudi Arabia — and 4.5 trillion litres of water.
  • By 2030, power use could hit 945 TWh, with AI accounting for 40% of the total.

The physical cost of the AI boom is coming into focus. A report from the United Nations University Institute for Water, Environment and Health warns that data centre power and water consumption will roughly double by 2030, and that governments ignoring the environmental bill risk strained grids, water stress and mounting electronic waste.

The numbers behind the build-out

The figures are stark. In the past year, data centres drew 448 terawatt-hours of electricity globally — exceeding Saudi Arabia’s entire consumption — with AI responsible for a fifth. They also used 4.5 trillion litres of water and generated 189 million tonnes of CO2. By 2030, annual power use is forecast to reach 945 TWh (comparable to Japan), water consumption 9.3 trillion litres, and the land footprint to more than double to over 14,500 square kilometres.

“The public debate still often treats AI as software, but AI is also physical infrastructure,” said the institute’s director and lead author, Kaveh Madani. His warning is pointed: AI will not “run out” of water or power globally, but poorly planned expansion could collide with existing pressures in specific places.

Looking forward

The report sharpens questions the UK is already confronting as capital pours into compute. Backers are committing sums such as SoftBank’s €45bn for French data centres, and the scale of AI debt issuance is reshaping bond markets — yet the resource side of that equation rarely makes the headlines. For UK planners weighing new data centre applications against water-stressed regions and grid constraints, Madani’s message is that responsible siting decisions matter “now, before infrastructure and dependencies become locked in”.

Share this article