AI May Add Up to 1.8 Gigatonnes of Emissions a Year, Study Finds

The big number in a new peer-reviewed study, reported by Wired and covered by Axios, is that AI could add between 0.47 and 1.8 gigatonnes of CO2 to global emissions every year. Not because of the electricity to train the models. Because of what the models help oil and gas companies do faster and cheaper.

The paper, published in npj Climate Action by Will Alpine and Holly Alpine of the Enabled Emissions Campaign, independent researcher Nathan Geldner, and Maksym Chepeliev of Purdue University's Center for Global Trade Analysis, models what happens when AI adoption grows at similar rates across fossil fuel and renewable energy sectors. The fossil-fuel productivity gains win. In the modeled scenarios, the extra emissions come to between 1.2% and 4.8% of the global energy-related emissions recorded in 2024, and the top of that range is roughly three times Germany's annual emissions.

The framing is the piece that shifts the conversation. Most of the AI-and-climate discussion has been about data-centre power draw. The authors call the far larger category 'enabled emissions': AI lowers the cost of finding, extracting and refining fossil fuels, so resources that were not commercially viable yesterday become viable tomorrow. They put the fossil-fuel enabled-emissions figure at 3.3 to 13.3 times the International Energy Agency's estimate of current data-centre emissions.

That is a wide band for a reason, and the retrieved coverage doesn't pin down which vendor contracts drive the modeled uplift, how the productivity assumption is calibrated for renewables versus oil, or what would happen if AI adoption in clean energy outran the base case. Read it as a modeling exercise, not a forecast.

Still, the shape of the finding is what matters for anyone making decisions today. If you sell cloud or upstream software into oil majors, your customer's biggest AI ROI story may be one that lands on someone else's balance sheet as emissions. If you regulate tech, the emissions worth counting may sit outside the data-centre fence. And if you fund climate-tech, the paper is a citable argument for pushing AI harder into the renewables side of the ledger before the enabled-emissions gap widens.