AI-Driven Fossil Fuel Efficiency Could Boost Global Emissions by 4.8%: Study
Former Microsoft sustainability workers Will and Holly Alpine argue that “enabled emissions” from AI tools used by energy giants like Chevron and Microsoft pose a greater climate risk than data centre expansion.
New research published in the journal npj Climate Action suggests that artificial intelligence could increase global energy-related emissions by between 1.2 and 4.8 percent by enhancing productivity within the oil and gas sector. The study, authored by former Microsoft sustainability workers Will and Holly Alpine, argues that the efficiency gains AI brings to fossil fuel extraction significantly outweigh its benefits for clean energy technologies and exceed projected emissions from data centre expansion.
The authors, who left Microsoft in early 2024 due to the company’s continued partnerships with the fossil fuel industry, describe a self-reinforcing cycle where technology firms provide power and tools to energy companies. This relationship increases reliance on carbon-intensive sources, creating what the Alpines term “enabled emissions.” Unlike operational emissions, which tech companies typically measure, enabled emissions refer to the greenhouse gas pollution resulting from AI-supported fossil fuel production.
Using a complex macroeconomic model, the researchers estimated that AI could add yearly emissions equivalent to Mexico’s at the low end, and up to Russia’s—the world’s fourth-largest emitter—at the high end. The study highlights that while AI is being used to improve efficiency in renewable sectors, such as making data centre cooling 30 to 40 percent more efficient, its impact on making fossil fuel extraction cheaper and faster is substantial.
A key example of this dynamic is the confirmed partnership between Chevron and Microsoft. The oil giant is building a large behind-the-meter gas plant in Texas to power Microsoft’s data centres. Chevron executives have indicated that the company plans to use some of the compute generated by this power for its own AI capabilities, illustrating the intertwined nature of tech and fossil fuel interests.
Energy researcher Jon Koomey, who was not involved in the study, described the research as a credible attempt to answer the net climate impact question. While noting that the final net outcome remains uncertain, Koomey acknowledged that arguments claiming AI will simply solve the climate problem often ignore the effects it has on all industries, including those that increase carbon output.
