Even without any “breakthrough discoveries,” the IEA estimates, widespread adoption of AI applications could cut emissions by 1.4 billion tons in 2035. Those reductions, “if realized,” would be as much as triple the emissions from data centers by that time, under the IEA’s most optimistic development scenario.
But that’s a very big “if.” It requires placing a lot of faith in technical advances, wide-scale deployments, and payoffs from changes in practices over the next 10 years. And there’s a big gap between how AI could be used and how it will be used, a difference that will depend a lot on economic and regulatory incentives.
Under the Trump administration, there’s little reason to believe that US companies, at least, will face much government pressure to use these tools specifically to drive down emissions. Absent the necessary policy carrots or sticks, it’s arguably more likely that the oil and gas industry will deploy AI to discover new fossil-fuel deposits than to pinpoint methane leaks.
To be clear, the IEA’s figures are a scenario, not a prediction. The authors readily acknowledged that there’s huge uncertainty on this issue, stating: “It is vital to note that there is currently no momentum that could ensure the widespread adoption of these AI applications. Therefore, their aggregate impact, even in 2035, could be marginal if the necessary enabling conditions are not created.”
In other words, we certainly can’t count on AI to drive down emissions more than it drives them up, especially within the time frame now demanded by the dangers of climate change.
As a reminder, it’s already 2025. Rising emissions have now pushed the planet perilously close to fully tipping past 1.5 ˚C of warming, the risks from heatwaves, droughts, sea-level rise and wildfires are climbing—and global climate pollution is still going up.
We are barreling toward midcentury, just 25 years shy of when climate models show that every industry in every nation needs to get pretty close to net-zero emissions to prevent warming from surging past 2 ˚C over preindustrial levels. And yet any new natural-gas plants built today, for data centers or any other purpose, could easily still be running 40 years from now.
Carbon dioxide stays in the atmosphere for hundreds of years. So even if the AI industry does eventually provide ways of cutting more emissions than it produces in a given year, those future reductions won’t cancel out the emissions the sector will pump out along the way—or the warming they produce.
It’s a trade-off we don’t need to make if AI companies, utilities, and regional regulators make wiser choices about how to power the data centers they’re building and running today.
Some tech and power companies are taking steps in this direction, by spurring the development of solar farms near their facilities, helping to get nuclear plants back online, or signing contracts to get new geothermal plants built.
But such efforts should become more the rule than the exception. We no longer have the time or carbon budget to keep cranking up emissions on the promise that we’ll take care of them later.