Study says AI could raise global emissions by up to 1.8 gigatonnes a year

A new peer-reviewed model finds AI's fossil fuel uses could drive more emissions than its clean-energy uses avoid, creating a net annual increase of up to 1.8 gigatonnes of CO₂. The study argues policymakers are undercounting AI's climate impact by focusing on datacenter power while missing emissions enabled by higher fossil fuel output.

Why it matters: - AI is increasingly being treated as a climate tool, but the study says its fossil fuel applications may amplify emissions far more than its renewable-energy uses reduce them. - The modeled net increase of 0.47 to 1.8 gigatonnes of CO₂ a year equals 1.2% to 4.8% of 2024 global energy-related emissions. - The authors say the missing metric is “enabled emissions,” or emissions caused when AI makes fossil fuel extraction and production more profitable. - The finding could reshape how governments, investors and companies measure AI’s climate footprint.

What happened: - A peer-reviewed study used energy-economic modeling to assess AI's impact across fossil fuel and renewable energy sectors. - The research was led by Will Alpine and Holly Alpine of the Enabled Emissions Campaign, Nathan Geldner, an independent researcher, and Maksym Chepeliev of Purdue University. - The authors modeled AI as a “bidirectional productivity amplifier” in an economy that remains about 80% powered by fossil fuels. - They found that when AI adoption happens at similar rates in fossil and renewable sectors, fossil fuel applications create more emissions than renewable applications avoid. - The study estimates 0.47 to 1.8 gigatonnes of additional CO₂ annually from those effects.

The details: - The paper says lower production costs from AI can expand what is commercially viable to extract and produce. - That shift can increase supply, reduce prices and trigger additional demand. - The authors say these emissions are 3.3 to 13.3 times the International Energy Agency’s estimate of current datacenter emissions. - Across 64 modeled adoption scenarios, net emissions fell only when fossil-sector productivity gains were zero. - The study says that outcome is unlikely because fossil fuel applications are already deployed at scale while renewable applications remain mostly in pilots and academic studies. - Under parallel adoption, renewable productivity gains had to exceed fossil fuel gains by 4 to 5 times just to break even on emissions. - That asymmetry remained directionally true even with substantial carbon pricing. - The authors say the findings likely understate the effect because renewable gains were set near the top of technical potential while fossil gains were based on industry and analyst disclosures.

Between the lines: - The research argues AI should not be judged only by how much electricity datacenters use. - Datacenter demand and enabled emissions are separate categories, but they reinforce each other because both draw on a fossil-heavy energy system. - The study frames AI as a tool that can accelerate decarbonization or extend fossil fuel dominance, depending on where it is deployed. - That means the climate debate may be missing a bigger, harder-to-measure effect: the economic lift AI gives to fossil fuel production. - Will Alpine said AI can help renewable energy and the grid, but has also been boosting fossil fuel productivity for years. - Holly Alpine said current assessments capture only part of AI’s climate impact because they ignore emissions enabled by additional fossil fuel production.

What's next: - The authors are calling for “enabled emissions” to be recognized, measured and governed. - The study uses a global computable general equilibrium model, GTAP-E-Power, calibrated with the GTAP-Power Data Base and real-world data. - The framework tested assumptions around elasticity, baseline conditions and carbon pricing. - The Enabled Emissions Campaign says its broader mission is to close a disclosure and governance gap around AI’s use in fossil fuel expansion. - The group says the full study and a plain-language summary are available for further reading.

The bottom line: - The study argues AI's climate cost may be much larger than datacenter emissions alone, because its fossil fuel uses can unlock more emissions than its clean-energy uses avoid.