AI fossil fuel emissions may be a bigger climate problem than the power draw of data centers alone. A peer-reviewed modeling study reported by WIRED finds that using AI to make oil, gas and related energy operations more productive could raise global energy-related emissions by 1.2% to 4.8% under the scenarios tested.
That is a modeled result, not a count of emissions already caused by AI. The researchers modeled productivity gains across fossil-fuel extraction, refining and electricity generation, drawing on oil-and-gas companies’ reported gains from AI tools. Their conclusion, as described by WIRED, is that the resulting increase in fossil-fuel activity outweighs AI’s modeled benefits in solar, wind and other clean technologies.
The point is uncomfortable but not mysterious. Better software can help companies identify deposits, plan drilling, improve recovery from existing fields, manage equipment and refine operational decisions. If those gains lower costs or make additional reserves economic to produce, more fuel can reach the market and create emissions when it is used.
What are AI-enabled emissions?
“Enabled emissions” are the additional greenhouse gases associated with fossil-fuel production made possible by AI and related digital systems. They differ from the direct footprint of AI infrastructure, such as emissions from manufacturing servers, constructing data centers and generating the electricity they consume.
That distinction matters because corporate climate accounting commonly emphasizes operational and supply-chain emissions. The study’s authors argue that this misses a consequential downstream effect: a technology provider can report its own facilities while its products help customers extract and sell more hydrocarbons.
Data-center emissions remain real. The International Energy Agency estimates that data centers produced about 180 million tonnes of indirect CO2 emissions from electricity consumption in 2024, about 0.5% of global fuel-combustion CO2 emissions. That estimate covers all data-center workloads, with AI only one part of the total, and excludes emissions from backup generation.
Can AI still cut emissions?
Yes, depending on what people and companies choose to do with it. The IEA says AI can help detect methane leaks in oil and gas operations, improve the efficiency of fossil-fuel power plants, optimize industrial processes, reduce transport fuel use and better manage energy consumption in buildings. Its widespread-adoption case projects that existing AI applications in end-use sectors could cut 1,400 million tonnes of CO2 in 2035.
That projection comes with a warning label. The IEA says no current momentum guarantees broad uptake of those uses, and rebound effects can erase savings. The eventual climate result depends on deployment, incentives, business cases and regulation.
So the new study does not establish that AI has one fixed carbon footprint. It does show why counting the electricity used by chips is an incomplete exercise. AI deployed to find methane leaks and cut waste has one effect; AI deployed to make new oil and gas production cheaper has another. Treating those applications as interchangeable would be convenient accounting, not serious climate analysis.
This story draws on original reporting from WIRED.
