UN warns of AI’s soaring water and power demand, as Google pledges to give back more

The UN’s environmental institute has issued a stark warning that the hidden physical infrastructure behind artificial intelligence is consuming water and electricity at a rate that could soon outstrip entire nations’ needs. The report from the United Nations University Institute for Water, Environment and Health (UNU-INWEH) calculates that in 2024 alone, data centres gulped 4.5 trillion litres of water – enough to supply sub-Saharan Africa – and 448 terawatt-hours of electricity, exceeding Saudi Arabia’s total power use. By the end of the decade, both figures are projected to double, with AI accounting for an ever-greater share. The water required to cool the servers running generative AI could equal the basic domestic needs of 1.3 billion people.

This looming resource crunch is set against frantic industry expansion. Alphabet, Google’s parent, has just raised a record $84.75 billion in equity, partly from Warren Buffett’s Berkshire Hathaway, to fund data centre construction. Yet the spending spree is colliding with mounting local opposition in the United States, where communities are imposing moratoria and construction bans over concerns about grid strain and water scarcity. Viewed from Stockholm, however, the cold Nordic climate offers a relative advantage, with Sweden positioning itself as a less resource-intensive host for the new computing behemoths. In Africa, the calculus is different still: Nigeria’s data centre boom is powered by a young, digitally connected population, presenting investors with a pitch that marries demographic promise with rising environmental risk.

Google itself has pre-empted the UN’s findings by promising that, starting in the United States, it will replenish more water than its data centres consume by 2030. The five-point plan includes investing $17 million in water management and shifting to recycled water and air-cooling where basin conditions allow. Yet independent researchers have criticised the UNU report’s methodology, and even its authors acknowledge that AI could help narrow the development gap by strengthening health systems in poor countries. The tension between AI’s promise and its environmental costs is increasingly acute: the same technology that could make farming more efficient is, for now, training a single model like GPT-4 with enough water to meet the annual needs of 81,000 people in sub-Saharan Africa.

The scale of the challenge extends well beyond water and power. The UNU report flags the strain on land and the mountains of electronic waste that the data centre boom will generate. As AI becomes embedded in daily life – nearly two-thirds of students already use it for coursework, and in the workplace, a proliferation of poorly integrated tools is exhausting employees rather than boosting productivity – the invisible supply chain grows heavier. The UN’s message is clear: without urgent policy intervention, the global south, which often hosts the raw materials and manufacturing for this infrastructure, will bear the sharpest edge of a crisis it did little to create.