AI's Productivity Gains in Fossil Fuels Outweigh Its Climate Benefits for Renewables, Study Finds

For every ton of carbon emissions saved by artificial intelligence optimizing a wind farm or solar grid, the same technology helps extract enough additional oil and gas to emit up to four tons more. That asymmetry sits at the center of a new energy-economic model published this week in the Nature Portfolio journal npj Climate Action, which attempts to quantify a previously unmeasured variable in the climate debate. While public and regulatory attention has focused intensely on the massive electricity demands of new AI data centers, the researchers argue that the technology's primary climate impact actually happens downstream, in the physical economy.[1][2][5]

By acting as a "bidirectional productivity amplifier," artificial intelligence accelerates whatever industry adopts it. When applied to an energy sector that is still roughly 80 percent powered by fossil fuels, the technology's ability to locate and extract hydrocarbons currently outpaces its ability to integrate clean energy into the grid. The research introduces a metric termed "enabled emissions"—the additional carbon pollution generated when AI makes fossil fuel extraction cheaper, faster, and more efficient, thereby expanding the total viable supply of oil and gas.[4][5]

The study's authors, which include researchers from Purdue University and the Enabled Emissions Campaign, modeled 64 different adoption scenarios across the global energy system. They found that if clean and dirty energy facilities adopt artificial intelligence at similar rates, the net global carbon output rises by 0.47 to 1.8 gigatonnes annually. That represents an increase of roughly 1.2 to 4.8 percent of all global energy-related emissions in 2024.[1][2][4]

To put that scale into perspective, the physical electricity consumption of the tech industry looks relatively small. The International Energy Agency estimates that all global data centers combined emitted approximately 0.18 gigatonnes of carbon last year. The enabled emissions generated by AI's application in the fossil fuel sector are therefore estimated to be between 3.3 and 13.3 times larger than the emissions generated by the computers running the models. This shifts the center of gravity in the AI climate debate from the power grid to the oil field.[4][5][6][7]

To understand how software generates physical emissions, the mechanism must be broken down into how different energy sectors deploy machine learning. In the renewable energy sector, artificial intelligence is primarily used for grid optimization and forecasting. Solar and wind operators deploy algorithms to predict weather patterns, schedule predictive maintenance for turbines, and determine the exact optimal moments to dispatch stored battery power to the grid.[6]

These applications are highly effective at reducing waste and maximizing the utility of existing clean energy infrastructure. However, the study notes that software cannot solve the primary bottlenecks holding back renewable energy deployment. Artificial intelligence cannot clear physical interconnection queues, it cannot speed up local land-use permitting delays, and it cannot manufacture the physical transformers needed to upgrade transmission lines. The technology optimizes what exists, but struggles to accelerate the physical build-out of new clean infrastructure.[1][6]

In the fossil fuel sector, the mechanism of AI adoption looks fundamentally different. Upstream oil and gas operators are using advanced machine learning models to process seismic data, locate previously hidden hydrocarbon deposits, and optimize the drilling process itself. By analyzing vast datasets of geological information, AI reduces the rate of "dry holes" and lowers the overall capital expenditure required to bring a new well online.[6]

The International Energy Agency has previously estimated that artificial intelligence could boost technically recoverable oil and gas reserves by 5 percent and cut the cost of complex deepwater offshore projects by 10 percent. When the cost of extraction falls, deposits that were previously considered too expensive or geologically complex to drill suddenly become commercially viable. This expands the total global supply of fossil fuels, which in turn lowers prices and induces additional global demand.[1][5]

This dynamic creates a steep exchange rate between clean and dirty energy optimization. According to the researchers' models, because fossil fuels still dominate the global energy mix, a 1 percent productivity gain in oil and gas extraction requires a corresponding 4 to 5 percent productivity gain in renewables just to keep global emissions flat. In almost every scenario modeled by the team, the net emissions only fell if artificial intelligence failed to increase productivity in the fossil fuel sector entirely—a scenario the authors deem highly unlikely given current market incentives.[1][5][6]

"Like any tool, AI can accelerate whatever it's applied to," noted Will Alpine, the study's lead author and a former Microsoft employee who left to found the Enabled Emissions Campaign. "Yes, it can advance renewable energy, strengthen the grid, and improve efficiency. But it can also optimize oil and gas extraction." The adoption curves for these technologies further compound the asymmetry, as fossil fuel applications are already being deployed at commercial scale, backed by massive capital investments from legacy energy companies.[2]

An IBM industry study cited by the researchers found that 44 percent of upstream extractors are already using artificial intelligence when prospecting for new deposits, with another 45 percent intending to deploy the technology within the next three years. Tech giants are actively competing for these lucrative enterprise contracts, selling proprietary, custom-built AI tools specifically designed to accelerate oil and gas production. In contrast, many of the most promising applications for artificial intelligence in the renewable sector remain at the pilot or academic-study stage.[1][4][6]

The researchers are transparent about the limitations of their modeling, dedicating significant space to what remains unknown. They caution that the 0.47 to 1.8 gigatonne range is a "directional and structural finding, not a precise forecast." The exact future emissions will depend heavily on macroeconomic variables, including the future price of carbon, the speed of regulatory permitting for renewable infrastructure, and the exact adoption curves of AI tools across different geographies.[1]

The model also assumes a parallel adoption rate between fossil fuels and renewables, which may shift if governments heavily subsidize clean energy AI applications or restrict tech companies from contracting with fossil fuel providers. However, the authors note that the broad relationship—where fossil fuel optimization outpaces renewable optimization—held true across every sensitivity test they ran, even under scenarios with substantial carbon pricing.[1][5]

The findings are already reshaping how climate advocates and policymakers view the technology sector's environmental commitments. While major tech companies frequently highlight their investments in renewable energy to power their data centers, critics argue this accounting ignores the downstream impact of their enterprise software sales. "This research exposes Big Tech as a lead accomplice to the fossil fuel industry in ways we never imagined," said Clara Vondrich, senior policy counsel for Public Citizen's Climate Program. As the artificial intelligence boom continues, the study suggests that the technology's ultimate climate legacy will be determined not by how efficiently its servers run, but by which industries use it to build the physical world.[4]