Artificial intelligence could produce a net annual increase of 0.47 to 1.8 gigatonnes of CO₂ if its use in fossil fuel production generates more emissions than its renewable-energy applications avoid, according to peer-reviewed research cited by Jefferies. The study, published in npj Climate Action, estimates that emissions linked solely to AI-enabled fossil fuel production could reach 3.3 to 13.3 times current data-centre emissions. Jefferies’ sustainability and transition strategy team said the findings challenge conventional assessments that weigh AI’s potential to accelerate clean energy against the electricity required to operate data centres. That approach can omit AI’s role in helping oil and gas companies reduce drilling costs, identify new deposits and improve recovery from existing fields, making additional fossil fuel supply more attractive and quicker to bring online. AI can also improve renewable power generation and energy-system efficiency, but the study found emissions avoided through those uses may be outweighed by emissions from greater fossil fuel production and consumption. Jefferies said the findings are relevant to investors evaluating AI infrastructure because the technology’s energy footprint extends beyond data-centre electricity use.