A peer-reviewed study published in the journal npj Climate Action has found that artificial intelligence-driven productivity gains across the global energy sector enable more carbon dioxide emissions from fossil fuels than they avoid through renewables, with net annual emissions rising by 0.47 to 1.8 gigatonnes across 64 modeled scenarios.
News Summary
- A peer-reviewed study in npj Climate Action found AI-driven productivity gains increase net global CO2 emissions by 0.47–1.8 gigatonnes annually across 64 scenarios.
- The increase equals 1.2–4.8 percent of 2024 global energy-related CO2 emissions.
- Researchers modeled AI as productivity shocks across both fossil fuel and renewable pathways using the GTAP-E-Power global economic model.
- Net emissions only fell in scenarios where fossil-sector productivity gains were zero. Renewables would need productivity gains four to five times greater than fossil fuels to break even.
- The study did not include direct datacentre energy demand, but found fossil fuel productivity gains produce emissions at least three times current estimates for AI datacentres.
The research, led by authors Will Alpine, N. Geldner, Holly Alpine, and M. G. Chepeliev, is the first to quantify AI’s climate impact across the full power sector using a global computable general equilibrium model. Previous studies had focused on indirect climate benefits , such as reducing renewable downtime and optimizing electricity grids , while largely ignoring the additional pollution generated by AI-assisted oil and gas extraction.
The Modeling Approach
The researchers used the GTAP-E-Power model, calibrated to a 2017 global economic database and aggregated across 20 sectors and five world regions. They introduced AI as productivity shocks rather than simple efficiency improvements, capturing how the technology influences production costs, market prices, capital allocation, and resource use across interconnected energy and industrial systems.
The analysis compared scenarios in which AI enhanced productivity in fossil fuel production, renewable energy systems, or both simultaneously. Across 64 empirically parameterized scenario combinations, net emissions reductions occurred only when fossil-sector productivity gains were set to zero. When clean and dirty energy facilities adopted AI at similar rates, renewables productivity gains would need to outpace fossil fuel gains by approximately four to five times for emissions to reach breakeven.
The authors explicitly excluded direct datacentre electricity use and operational emissions from their equilibrium calculations, using those figures only as an external point of comparison. They cautioned that the results represent a “directional and structural finding, not a precise forecast.”
Fossil Fuel Deployment Already at Scale
Holly Alpine, a co-author and former Microsoft employee who left to co-found the Enabled Emissions campaign group, said the research team used a conservative assumption that AI adoption would happen at the same rate for fossil fuels and renewables. In practice, she told The Guardian, “fossil fuel applications are already happening at scale today , real contracts, real deployment, with evidence from industry operators and financial analysts,” while “renewables applications are still largely at the pilot or academic-study stage.”
The International Energy Agency estimates that AI could boost technically recoverable oil and gas reserves by 5 percent and cut the cost of deepwater offshore projects by 10 percent. Oil and gas executives have described AI’s emerging impact as “the next fracking boom.”
Saudi Aramco said last year that it had embedded AI “in everything,” increasing productivity and the number of wells. Equinor attributed 27 discoveries on the Norwegian continental shelf to new seismic technologies and AI, including the Lofn and Langermann oil wells, which the company said was the largest discovery it operated in 2025. “AI was key, from automated data interpretation to efficient well planning,” Equinor said at its capital markets day in June.
Rystad Energy, an independent research firm based in Oslo, estimated in May that digitalisation and AI would create close to $500 billion in cumulative value for fossil fuel exploration and production companies between 2026 and 2030. The firm cited hundreds of millions of dollars in reported AI-related savings from Equinor and Abu Dhabi’s Adnoc.
A Blind Spot in Climate Research
Lynn Kaack, an assistant professor of computer science and public policy at the Hertie School who provided feedback on a draft of the research, told The Guardian that previous studies had compared datacentre energy use against AI’s emissions savings while omitting the picture of AI causing increases in emissions.
The study found that AI-enabled productivity gains in the fossil fuel sector result in emissions at least three times current estimates for datacentres. That comparison is notable given the intense scrutiny that AI datacentre energy demand has received from climate scientists and regulators. The researchers did not factor datacentre energy into their model, meaning the total climate impact of AI could be larger than the study’s headline figures suggest.
Analyst Reaction
Ketan Joshi, an independent climate analyst who was not involved in the study, told The Guardian that the AI sector was “fundamentally hungry for fossil fuels” and that the impacts extended well beyond datacentres driving fossil fuel use. “Even within some parts of the climate movement, there is still a denial that an unchecked tech industry will inherently boost fossil fuels,” he said. “Simply asking companies to throw a few scraps of cash at renewable projects is not enough to ensure the industry operates safely.”
The authors emphasize that their findings are comparative-static equilibrium scenarios rather than long-term forecasts, and they call for governance frameworks that distinguish enabled emissions from avoided emissions when evaluating climate policies.