Study finds that the AI making fossil fuel more efficient is also making the planet hotter 

Artificial intelligence has long been sold as a potential climate hero. It can forecast wind and solar generation, optimise electricity grids, predict equipment failures and squeeze more efficiency out of renewable energy systems. 

But there is an uncomfortable catch: AI can make fossil fuels more efficient too. 

A new peer-reviewed study published in npj Climate Action suggests that this less-discussed side of the AI revolution could overwhelm its potential climate benefits. Across 64 scenarios, researchers found that AI-driven productivity gains in coal, oil and gas could result in a net increase of between 0.47 and 1.8 gigatonnes of carbon emissions a year. That is equivalent to roughly 1% to 5% of annual energy-sector emissions.  

The finding changes the way the climate cost of AI needs to be understood. Until now, much of the debate has centred on the electricity consumed by increasingly enormous data centres. The new research looks instead at what AI enables its users to do, including finding, drilling and extracting more fossil fuels. 

The problem is not just the data centre 

The concept at the heart of the study is what researchers call “enabled emissions”, i.e., pollution that is not generated directly by a technology, but becomes possible because of it. 

Simply put, an AI model sitting in a data centre may consume electricity and generate emissions (as well as some backlash). But if the same technology helps an oil company locate a new reservoir, optimise drilling or recover more oil from an existing field, the resulting emissions from producing and ultimately burning that additional fuel are also part of AI’s broader climate footprint. 

The researchers found that fossil-fuel applications could create 3.3 to 13.3 times more climate pollution than the emissions associated with powering AI data centres. The study’s authors cautioned that the numbers are a directional finding rather than a precise forecast, but said the underlying relationship remained consistent across their scenarios and sensitivity tests.  

That creates a particularly awkward contradiction for an industry that frequently presents AI as a tool for decarbonisation. 

Big Oil is already using AI 

This is not merely a theoretical possibility. Oil and gas companies are already using AI for seismic analysis, reservoir modelling, drilling, predictive maintenance and production optimisation. The International Energy Agency has estimated that digital technologies could increase technically recoverable oil and gas resources by around 5%, while reducing production costs. The Guardian also reports that oil and gas executives have described AI’s emerging impact as potentially resembling “the next fracking boom”.  

Saudi Aramco has said it has embedded AI across its operations, with productivity and well numbers increasing. Equinor, meanwhile, has credited seismic technology and AI with helping identify discoveries on the Norwegian continental shelf.  

There is also a powerful financial incentive. Rystad Energy estimates that digitalisation and AI could create close to $500 billion in cumulative value for oil and gas exploration and production companies between 2026 and 2030. The gains are expected to come from more efficient operations, higher production and recovery, and shorter development timelines. Rystad says the returns are already visible, pointing to hundreds of millions of dollars in AI-related savings reported by Equinor and Abu Dhabi’s Adnoc. 

In other words, fossil-fuel companies have both the motivation and the money to deploy AI at scale. 

The renewable-energy problem 

AI can undoubtedly help renewables. Better weather forecasting can improve wind and solar generation. Predictive maintenance can reduce downtime. Smarter electricity grids can balance supply and demand, while AI can improve the operation of batteries and other energy-storage systems. 

The problem is one of relative adoption. The researchers assumed AI would be adopted at comparable rates across fossil-fuel and renewable-energy industries. Under that assumption, renewable-energy productivity would have to improve by at least four times as much as fossil-fuel productivity simply for emissions to break even. In the study, emissions fell only in scenarios where AI did not increase productivity in the fossil-fuel sector.  

That is a sobering finding because AI adoption in oil and gas is already happening commercially, while many renewable applications remain at pilot or research stages. 

What AI is used for matters 

The larger lesson is that “AI” cannot be treated as inherently green or dirty. The same technology can optimise a solar farm or help an oil company drill another well. It can make electricity grids smarter while simultaneously making fossil-fuel extraction cheaper. Its climate impact therefore depends not only on how much energy AI consumes, but on the economic activity it accelerates. 

That distinction could also force a rethink of how technology companies measure their environmental footprints. Cleaning up data-centre electricity is important, but it may address only one part of AI’s climate impact. 

The question, therefore, is about what makes AI makes profitable. And if the most profitable application of intelligence is still extracting more fossil fuels, the technology may end up accelerating the climate problem it was supposed to help solve. The heat truly is on.