The metaphor does the hiding. “The cloud” suggests something weightless; the machines behind it are anything but. A new report from the UN University's Institute for Water, Environment and Health, Environmental Cost of AI's Energy Use, sets out to measure the footprint we usually wave away — and insists that carbon is only part of it.
By 2030, the report projects, the data centres powering AI could draw around 945 terawatt-hours of electricity a year — close to three per cent of the world's supply, and nearly triple the combined annual use of Pakistan, Bangladesh and Nigeria, home to more than 650 million people. The hidden costs travel with the power: water use to match the basic annual needs of some 1.3 billion people, and a land footprint that could exceed 14,500 square kilometres. The authors' sharpest warning is against the comfort of efficiency. Under what economists call the Jevons paradox, making a resource cheaper to use tends to increase, not reduce, how much of it we consume — so a more efficient model may simply mean more models, more queries, more draw. And most of that draw, they note, comes not from training the systems but from running them.
For readers of this publication the report rhymes with a story we have already told — the scramble for the minerals and power that feed the machine, fought out in places like Tanzania. The case is not that AI should not be built. It is that the bill is physical, it is unevenly distributed, and it is being paid in carbon, water and ground we rarely think to count.
Sources
The report and its projections. The 945-terawatt-hour and ~3 per cent figures, the comparison to Pakistan, Bangladesh and Nigeria, the water and land footprints, the Jevons-paradox argument, and the finding that inference rather than training dominates consumption are drawn from the UN University (UNU-INWEH) report Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints (Aczel, Chamanara, Matin et al., 2026) and its reporting by UN News and The Conversation.
Location: Dublin, Ireland — 53.35° N, 6.26° W
About these sources
UN University (UNU-INWEH) — Environmental Cost of AI's Energy Use (report)
UN News — AI's environmental costs threaten water, land and climate
The Conversation — UN report warns AI could soon use 3% of the world's electricity
Further reading — including dissenting views
For the counter-argument that AI's environmental cost should be weighed against the resources it can save — precision irrigation, crop mapping and yield gains that reduce agriculture's own land and water demand — see the industry response summarised alongside the report (IndexBox, linked above).