AI data centers guzzle as much electricity as entire cities, igniting demand for any energy source imaginable. This surge is straining the grid and increasing utility bills. Yet, a new paper suggests that this gargantuan power consumption is merely a sideshow in the larger energy conversation.

The paper seeks to quantify how artificial intelligence will change the energy system, concluding that the most profound impact won’t come from how the technology is powered, but what it is used for. The central prediction is that AI is on course to extend the dominance of fossil fuels.

By modeling scenarios where AI boosts productivity for both renewable energy and fossil fuels, the research finds that the improved economics of oil-and-gas production would have a significant environmental cost. Specifically, it estimates that these efficiencies would push up global carbon-dioxide emissions by between 0.47 billion tons and 1.8 billion tons a year. While clean-tech exports and lithium-free batteries emerge, the Strategic Petroleum Reserve has already shrunk below 300 million barrels, highlighting the volatile landscape of global energy resources as AI integrates into the oil patch.