AI May Help Clean Energy, but Its Oil Gains Could Add 1.8 Billion Tonnes of CO₂
A global economic model estimates that AI-driven gains in fossil-fuel production could outweigh its benefits for renewable energy, adding 0.47 to 1.8 gigatonnes of annual CO₂. The result covers 64 modeled scenarios, not observed future emissions, and excludes possible clean-technology breakthroughs.
Artificial intelligence can help a wind farm predict its output, but the same computing techniques can also locate oil deposits, lower drilling costs and extract more fuel from existing fields. A new global modeling study finds that the second effect could dominate, increasing annual carbon dioxide emissions even when AI also improves renewable energy.
The 30-second summary
- What happened? Researchers modeled how AI-driven productivity changes in fossil fuels, renewable power, grids and selected industries would move through the global economy.
- Why does it matter? Across 64 scenarios, parallel AI adoption increased annual CO₂ emissions by 0.47 to 1.8 gigatonnes.
- What is the catch? This is a comparative economic model, not a forecast of measured emissions. Its output depends on assumptions about adoption, productivity and market responses.
KEY NUMBER
Renewable-energy productivity would need to improve roughly four to five times as much as fossil-fuel productivity for the modeled emissions balance to break even.
The climate question extends beyond data centres
Most public arguments about AI and climate focus on the electricity and water consumed by data centres. Those direct costs matter, and NewTqnia has examined how local communities are becoming a constraint on AI infrastructure. The new paper asks a different question: what happens when AI changes how much energy companies can produce?
The peer-reviewed study in npj Climate Action treats AI as a productivity amplifier operating on both sides of the energy system. Better forecasting and maintenance can increase renewable output, while seismic interpretation, drilling optimization and reservoir management can make additional fossil extraction economical.
How the researchers reached the estimate
The team used a computable general equilibrium model, a mathematical representation of interconnected industries, prices, trade and demand. It applied productivity changes to fossil extraction and generation, renewable generation, electricity grids, maritime shipping and selected energy-intensive industries, then calculated how markets would respond.
Across 64 combinations of adoption assumptions, equal progress in fossil and renewable sectors produced a net increase of 0.47 to 1.8 gigatonnes of CO₂ a year, equivalent to 1.2% to 4.8% of global energy-related emissions in 2024. Upstream extraction was the decisive mechanism: reducing the resources needed to find and produce coal, oil or gas made more supply economically viable.
The result does not mean AI offers no climate benefit. NewTqnia has covered a system that gave cyclone forecasters about one additional day of useful forecasting skill, and a chemistry pipeline that screened 100,000 candidate battery solvents. The paper’s narrower conclusion is that helpful applications do not automatically cancel the emissions created when AI expands fossil supply.
Why fossil gains carry more weight
Renewable optimization usually extracts more electricity from equipment already built. Fossil-sector productivity can affect both operating efficiency and the amount of fuel considered profitable to recover. Once additional fuel reaches the market, lower costs and greater supply can stimulate consumption elsewhere in the economy.
This explains the model’s uneven break-even point. Each 1% fossil-fuel productivity gain required roughly 4% to 5% renewable gains to neutralize the resulting emissions. Even modeled improvements to grids and selected end uses reduced the total without reversing its direction under parallel adoption.
Independent reports published on August 11 brought the result into wider public scrutiny. Axios documented the oil industry’s expanding use of AI, while the Guardian obtained outside expert assessments and examples of reported deployments.
Before we overstate the result
- The paper was published on August 4, 2026. The new event is the substantial independent reporting and scrutiny published on August 11, not a later experiment.
- The 0.47 to 1.8 gigatonne range is a modeled annual effect under specified scenarios, not a measured increase or a precise prediction.
- The framework holds many structural conditions constant and does not model the timing of adoption or every possible market and policy response.
- It excludes speculative AI-enabled breakthroughs such as commercial fusion or long-duration storage, because their future effects cannot yet be quantified reliably.
- Several authors are affiliated with the advocacy group Enabled Emissions Campaign, although the article passed peer review and publishes its assumptions and sensitivity tests.
What would turn the model into an observable test?
The immediate need is disclosure. Energy and technology companies would have to report where AI is deployed, what productivity change it produces and whether that change raises total extraction or merely reduces operational emissions. Without those data, policymakers can measure data-centre electricity more easily than the larger market effects of AI applications.
The International Energy Agency’s Energy and AI assessment provides a broader baseline for electricity demand, energy security and potential efficiency gains. The new study argues that future assessments should add a separate ledger for emissions enabled by cheaper or more productive fossil extraction.
The takeaway
AI is neither inherently clean nor inherently polluting. In this model, the decisive variable is where its productivity gains land: better renewable forecasting avoids some emissions, but cheaper upstream extraction increases the quantity of fossil fuel the economy can profitably burn. The estimated 0.47 to 1.8 gigatonnes remains a scenario result until transparent deployment data allow researchers to test that mechanism against real production.
Verified topics and entities
Sources and citations4 sources
External references used to support the reporting in this article.
Published by
NewTqnia Climate Technology Desk
An institutional editorial team within NewTqnia
