Ask the people building artificial intelligence what it is for, and sooner or later the answer turns to the planet. Sam Altman has written that AI would bring astounding triumphs, among them fixing the climate. A Google sustainability executive argued back in 2023 that the technology could cut global emissions by 5 to 10 percent by 2030. It is a reassuring story, and it does a lot of quiet work justifying the enormous energy the industry is now drawing. The trouble is that the evidence for it is thin.
That is the argument of a new analysis in MIT Technology Review by Mary Nichols, who once chaired the California Air Resources Board, and Jeremy Tamanini, who runs a sustainability advisory firm. Their verdict is blunt. The case that AI will help solve climate change has not been demonstrated, and so far the evidence mostly points the other way.
The optimistic case, and where it thins out
There is a real version of the hopeful argument. A 2025 study from the London School of Economics and Systemiq estimated that AI could avoid between 3.2 and 5.4 billion tons of carbon dioxide a year by 2035, counting its own energy use against the ledger, with most of the savings in power, in meat and dairy, and in transport. On paper that is a serious contribution.
On the ground it gets harder to find. Of the grid operators the authors reviewed across China, the United States and Europe, only one had a clearly working program: the UK's National Grid ESO, which used forecasting from Open Climate Fix and halved its solar prediction error. Elsewhere the wins are scattered and small. A battery maker reportedly cut material-screening time sharply. Independent researchers found that AI routing trimmed per-trip emissions from one ride-hailing fleet by about 16 percent. Meanwhile the funding behind cultivated meat, one of the hoped-for AI-assisted breakthroughs, fell from nearly a billion dollars in 2021 to around 83 million last year. And the same tools that help a grid can help an oil and gas firm pump more efficiently.
The cost side of the ledger is not hypothetical
The reason this matters is that the other half of the equation is already here and growing. The build-out of data centres to train and run these models has driven a steep rise in energy use, and several of the companies making the climate promises, Microsoft and Google among them, have drifted off their own emissions targets. The International Energy Agency's base case has global data-centre emissions passing 300 million tons a year by 2035.
So the honest answer to the question is neither the marketing line nor its mirror image. AI is not a climate savior, and the worry about its footprint is not overblown fear, a tension this site has traced through claims that data centres will break the local grid, questions about how much water and power a single prompt uses, and even the idea of moving the whole problem into orbit. The benefits are possible and, in a few places, real. The cost is certain and rising now. Treating the first as settled because the second is uncomfortable is how a convenient story hardens into a myth.
Commentarii · 0