Artificial intelligence could increase global energy emissions by making the oil and gas industry more productive, adding between 1.2 to 4.8 percent to worldwide pollution levels, according to new research published last week in npj Climate Action. The study, led by former Microsoft sustainability workers Will and Holly Alpine, warns that AI's role in boosting fossil fuel production may harm the climate far more than the emissions from powering data centers themselves. The researchers found that this increase in pollution outweighs the benefits AI brings to clean energy development.
At the low end of the researchers' projections, the additional annual emissions could match Mexico's total output; at the high end, AI-enhanced fossil fuel production could add as much greenhouse gas pollution as Russia, currently the world's fourth-largest emitter. The study used a complex economic model to estimate how AI serves as a productivity booster across multiple segments of the fossil fuel sector, from extraction to refining to electricity generation. These projections significantly exceed multiple forecasts around emissions from data center energy consumption. Oil and gas companies have deployed various forms of AI for decades to locate and develop underground resources more efficiently, the report notes.
"One of the key insights of our paper is that you cannot treat them independently," Will Alpine told WIRED about the relationship between technology companies and fossil fuels. "They are two sides of the same coin." The researchers refer to AI-supported greenhouse gas pollution as "enabled emissions," which tech companies don't tend to measure even as they track their own operational emissions and supply chain impact. According to Holly Alpine, sustainability measures within tech companies focus heavily on operational emissions rather than the pollution their tools enable. "The scale of this was staggering," Will Alpine said of the modeling results.
The research argues that this AI-driven productivity boost increases global reliance on fossil fuels and makes it harder for the world to meet its climate targets. While machine learning can make data center cooling 30 to 40 percent more efficient, energy researcher Jon Koomey noted, it can also make fossil fuel extraction much cheaper and faster. The relationship between AI developers and oil giants is growing closer: Chevron and Microsoft recently confirmed plans for the oil company to build a large behind-the-meter gas plant in Texas to power Microsoft's data centers, with Chevron's president of New Energies hinting the company will "use some of that compute" to power AI inside Chevron itself. The Alpines, who quit Microsoft at the start of 2024 over the company's continued work with the oil and gas industry, began publicly campaigning to draw attention to the AI-fossil fuel connection.
The paper contends that while accounting for operational emissions remains important, it ignores pollution that does substantially more damage to the climate. Energy researcher Jon Koomey, who wasn't involved in the analysis, said the researchers' conclusions appear solid and represent "a credible attempt" to answer how AI's effects across all industries net out using a macroeconomic model. The study challenges claims from AI proponents who argue the technology will solve the climate crisis and should therefore be developed as quickly as possible, with Koomey noting such arguments ignore AI's impact on every sector, not just renewable energy and efficiency. Tech companies and fossil fuel producers appear locked in what the Alpines describe as "a self-reinforcing effect between supply and demand," where each side strengthens the other's climate impact. The question now is whether policymakers and corporate leaders will begin measuring and limiting the emissions their AI tools enable, not just the emissions from running them.