AI could boost fossil fuel output more than data centers harm climate

A new paper published last week in the journal npj Climate Action, authored by Will and Holly Alpine, challenges the prevailing narrative that artificial intelligence will help solve climate change. The Alpines, both former Microsoft sustainability workers with years of combined experience at the company, left their positions at the start of 2024 specifically because of Microsoft's continued partnerships with the oil and gas industry. They have since begun a public campaign to highlight what they describe as a dangerous and often-ignored relationship between AI development and fossil fuel expansion.

Using a complex economic model, the Alpines introduced various factors to simulate how AI could play out across the broader economy. They drew on reports from oil and gas companies documenting demonstrated gains from AI tools, then modeled AI as a productivity enhancer across different sectors of the fossil fuel industry—from extraction to refining to electricity generation. Their calculations found that AI could increase global energy-related emissions between 1.2 to 4.8 percent. At the low end, the additional yearly emissions would equal Mexico's total; at the high end, they could match Russia's, making Russia the world's fourth-largest emitter. This range of impact significantly exceeds multiple projections around emissions from data center energy use. "The scale of this was staggering," Will Alpine told researchers.

A central finding is that the climate benefits AI brings to developing solar, wind, and other clean technologies are vastly outweighed by the emissions increases from AI-enhanced fossil fuel production. The Alpines argue that tech companies have a measurement problem: they track their own operational emissions and supply chain pollution closely, but they do not measure how much their tools help increase fossil fuel production—what they call "enabled emissions." While sustainability measures within tech companies remain focused on operational emissions, Holly Alpine explains, this ignores emissions that cause far greater climate damage. As Jon Koomey, an energy researcher not involved in the study, tells WIRED, "Machine learning can make data center cooling 30–40 percent more efficient, but [could] also make fossil fuel extraction much cheaper and faster. How that nets out nobody yet knows for sure, but this new research is a credible attempt to answer that question using a macroeconomic model."

The relationship between these two industries is already visible in real-world partnerships. Chevron and Microsoft recently confirmed that Chevron would build a large behind-the-meter gas plant in Texas to power Microsoft's data centers. During a June call with analysts, Jeff Gustavson, president of Chevron's New Energies division, indicated that Chevron would use some of the compute generated by the power plant serving Microsoft "to actually power AI inside of our company." Will Alpine describes this arrangement as "perfectly illustrative of the relationship between AI and fossil fuels." The loop is self-reinforcing: fossil fuel companies provide the power that trains AI systems, which in turn make those same companies more efficient at extraction and production. Tech companies and fossil fuel producers benefit, but the climate cost—invisible in traditional emissions accounting—grows substantially.