AI productivity could supercharge fossil emissions faster than clean energy can cancel them

The model counts the pathway most climate claims omit

The study uses a global computable general equilibrium model to compare AI-driven productivity gains in fossil fuels and renewable energy. Fossil applications can reduce extraction costs, expand economically viable supply, and extend the productive life of incumbent assets. Renewable applications can improve forecasting, maintenance, generation, and grid integration.

Many public assessments emphasize the electricity consumed by AI infrastructure and the emissions avoided by efficiency tools. The authors argue that this boundary misses economy-wide rebound and induction effects, including new production that becomes profitable because AI lowers costs.

The modeled balance is structurally tilted

Across parallel adoption scenarios, modeled net annual carbon dioxide emissions rise by 0.47 to 1.8 gigatonnes. Enabled emissions exceed avoided emissions whenever fossil-sector productivity gains are nonzero, and a net reduction requires renewable gains four to five times larger than fossil gains.

The result is conditional on the model, parameters, adoption pathways, and 2024 economy used by the researchers. It should not be presented as a prediction that AI will necessarily cause the upper estimate. It is evidence that equal technical improvement across unequal energy systems can reinforce the incumbent.

Efficiency needs direction

A general-purpose technology does not choose the sector in which its productivity is socially beneficial. Companies deploy it where returns are available, including fossil extraction, logistics, and demand expansion. Governance determines whether clean-energy gains actually displace carbon-intensive production.

Policymakers should require full-system emissions accounting, distinguish efficiency from absolute reductions, and pair AI deployment with fossil constraints, clean-energy expansion, and transparent assumptions. A green use case does not cancel a high-carbon use case elsewhere in the economy.

• Report enabled emissions and avoided emissions within the same analytical boundary.
• Test rebound, induction, and fossil-supply pathways instead of assuming efficiency becomes decarbonization.
• Tie public incentives to measured absolute reductions and displacement of fossil production.
• Publish model assumptions and sensitivity ranges for independent scrutiny.