AI Could Drive Global Emissions Up by 1.8 Billion Tonnes a Year, Study Finds
Artificial intelligence could ultimately increase global greenhouse gas emissions rather than reduce them if its use in the fossil fuel industry continues to outpace applications in renewable energy, according to a new peer-reviewed study.
The study estimates that AI-driven productivity gains across the global economy could result in an additional 0.47 to 1.8 gigatonnes of carbon dioxide emissions annually when AI adoption expands at similar rates across fossil fuel and renewable energy sectors.
That would amount to between 1.2% and 4.8% of global energy-related emissions recorded in 2024, the researchers say.
The additional emissions, which the researchers call “enabled emissions,” come not from the electricity consumed by AI data centres, but from the economic activity that AI can make more productive.
According to the study, AI can lower the cost of extracting and producing fossil fuels, making previously less commercially viable resources more attractive to producers. Increased production can then affect energy prices and stimulate additional demand, ultimately generating more emissions.
The researchers estimate that fossil fuel applications of AI could enable emissions 3.3 to 13.3 times greater than the International Energy Agency’s estimate of current data-centre emissions.
The study was authored by Will Alpine and Holly Alpine of the Enabled Emissions Campaign, independent researcher Nathan Geldner, and Maksym Chepeliev, a Research Associate Professor at Purdue University’s Center for Global Trade Analysis.
The researchers describe AI as a “bidirectional productivity amplifier”, a technology capable of accelerating both the transition toward cleaner energy and the continued expansion of fossil fuels.
Fossil fuel gains could outweigh clean-energy benefits
The analysis examined 64 different AI adoption scenarios, looking at productivity improvements across fossil fuels, renewable energy, electricity grids and demand-side efficiency.
The researchers found that net emissions reductions occurred only in scenarios where AI generated no productivity gains for the fossil fuel sector.
They argue that such a scenario is unlikely because AI applications are already being deployed in fossil fuel operations, while many AI applications aimed at accelerating renewable energy remain at pilot or research stages.
Under scenarios in which AI productivity improvements occur across both fossil fuel and renewable energy sectors, renewable-energy productivity gains would need to exceed fossil fuel gains by four to five times simply for the overall emissions impact to reach a break-even point.
The researchers say this imbalance remained evident even when substantial carbon pricing was included in the model.
“I spent years building AI platform tools and have seen firsthand how they’re used,” said Will Alpine, lead author of the study.
“Like any tool, AI can accelerate whatever it’s applied to. Yes, it can advance renewable energy, strengthen the grid, and improve efficiency. But it has also been boosting the productivity of the fossil fuel industry for years, and our research shows that effect is asymmetric: it acts as an economic lever that reinforces the viability and dominance of fossil fuels.”
A different measure of AI’s climate impact
The researchers stress that their findings address a different issue from the growing debate over the amount of electricity consumed by AI data centres.
Data-centre emissions are associated with the energy required to run AI systems. Enabled emissions, by contrast, refer to emissions generated by the economic activity AI makes possible or more profitable, particularly increased fossil fuel production.
The two impacts are not directly additive, the researchers note, but they can reinforce each other.
As AI data centres increase electricity demand, they place additional pressure on an energy system that remains heavily dependent on fossil fuels. At the same time, AI can improve productivity within the fossil fuel industry, potentially encouraging further production.
“Most assessments of AI’s climate impact are framed as a tradeoff between datacenter energy use and the emissions AI might help avoid,” said Holly Alpine, co-founder of the Enabled Emissions Campaign.
“What’s missing entirely is the other side of the ledger for AI’s applications: the emissions enabled from the additional fossil fuel production being made commercially viable.”
She said the findings show the need to account for emissions linked to AI-enabled economic activity when assessing the technology’s overall climate impact.
Global economic modeling
The study uses a global computable general equilibrium (CGE) model, known as GTAP-E-Power, to examine how AI-driven productivity changes could affect different parts of the energy system.
The model was calibrated using the GTAP-Power Data Base and additional parameters based on real-world data. The researchers tested their findings under different assumptions about economic responses, baseline conditions and carbon pricing.
They modeled AI productivity improvements across fossil fuels, renewable energy, electricity-grid infrastructure and demand-side energy efficiency.
The authors caution that their estimates may actually understate AI’s potential net emissions impact.
According to the study, the renewable-energy productivity gains used in the modeling were calibrated toward the upper end of their estimated technical potential, while fossil fuel productivity gains were based on figures disclosed or projected by industry operators and financial analysts.
The findings come as governments and companies increasingly look to AI to improve energy efficiency, expand renewable energy, optimize electricity grids and accelerate climate-related innovation.
But the researchers argue that the same technology could also strengthen fossil fuel production unless its deployment is deliberately directed toward accelerating the energy transition.
About The Author
SUBSCRIBE TO OUR NEWSLETTER
