Artificial intelligence is often presented as a powerful weapon against climate change. It can help improve electricity grids, predict renewable energy production and make industrial systems more efficient. But new research suggests there is another side to the story.
AI fossil fuel emissions could rise significantly as oil, gas and coal companies use artificial intelligence to discover resources, reduce production costs and extract more energy from existing operations. A new study published in npj Climate Action estimates that these productivity gains could ultimately create considerably more carbon pollution than AI helps eliminate through improvements in renewable energy.
The findings shift the climate debate around AI beyond one familiar question: how much electricity do data centres consume?
Researchers argue that society also needs to consider what AI enables companies to do.
AI Fossil Fuel Emissions Could Reach a Huge Scale
The numbers presented in the study are significant.
Researchers modelled AI-driven productivity improvements across the global energy economy and found that net annual carbon dioxide emissions could increase by between 0.47 and 1.8 gigatonnes.
That represents roughly 1.2% to 4.8% of global energy-related CO₂ emissions recorded in 2024.
At the lower end, the additional yearly emissions would be comparable to the annual emissions of a sizeable industrialised economy. At the upper end, the impact could approach the emissions produced by one of the world’s largest emitting countries.
Researchers reached those estimates by examining AI not simply as an electricity consumer but as a technology capable of increasing productivity throughout the energy sector.
That distinction matters.
An AI system helping an oil producer locate deposits faster does not necessarily consume an enormous amount of electricity by itself. However, if the technology allows the company to produce significantly more oil at a lower cost, the resulting fossil fuels can eventually generate far greater emissions when they are used.
How AI Helps Fossil Fuel Companies Produce More
Artificial intelligence is already useful across large parts of the oil and gas industry.
Companies can apply machine learning to geological information, seismic surveys and enormous collections of operational data. These systems can help identify promising drilling locations, improve extraction techniques, predict equipment failures and optimise refinery operations.
AI can therefore make existing fossil fuel infrastructure more productive.
The International Energy Agency has previously estimated that widespread AI adoption could increase technically recoverable oil and gas resources while also reducing costs for some projects.
That creates an economic effect that can be easy to overlook.
If AI reduces the cost of finding and producing oil or gas, companies may be able to develop resources that were previously considered too expensive or technically difficult.
More efficient production can increase supply.
Greater supply can influence prices.
And cheaper or more accessible fossil fuels can encourage continued consumption.
The result is a chain reaction in which a relatively small improvement in industrial efficiency can eventually translate into substantial additional emissions.
AI Fossil Fuel Emissions Go Beyond Data Centres
Much of the environmental discussion surrounding artificial intelligence has concentrated on data centres.
That concern is understandable.
AI models require powerful computing infrastructure, and rapidly expanding data centres are creating enormous new electricity requirements. Global electricity generation needed to supply data centres is projected to increase from around 460 terawatt-hours in 2024 to more than 1,000 TWh by 2030 under the International Energy Agency’s base scenario.
But the latest study suggests electricity consumption may represent only part of AI’s climate impact.
The researchers introduce the idea of “enabled emissions”—pollution that occurs because AI makes carbon-intensive industries more productive.
Consider two different situations.
In the first, an AI data centre consumes electricity and produces an associated carbon footprint.
In the second, an AI system helps an energy company operate wells more efficiently, discover additional reserves or reduce production costs. The computer itself may account for relatively modest emissions, but the additional fossil fuel production it enables can eventually generate considerably more.
Looking only at data-centre electricity would miss that second effect.
The Same AI Can Also Help Renewable Energy
The research does not argue that AI is inherently bad for the climate.
Artificial intelligence can provide genuine benefits to renewable energy.
Machine learning can improve forecasts for wind and solar generation, optimise electricity networks, reduce equipment downtime and help operators balance changing electricity supply and demand.
AI can also identify inefficiencies that would otherwise waste energy.
These applications could lower emissions.
The problem identified by the researchers is one of scale.
Their economic modelling examined both sides simultaneously: emissions avoided when AI improves renewable energy productivity and emissions enabled when AI increases fossil fuel productivity.
Across the scenarios studied, the fossil fuel effect generally proved stronger.
Researchers found that AI-driven productivity gains in renewable energy would need to substantially outperform comparable improvements in fossil fuels for the overall emissions effect to become beneficial.
That challenges the assumption that simply deploying more AI throughout the energy industry will automatically help tackle climate change.
Where the technology is deployed matters.
Why Efficiency Does Not Always Reduce Pollution
At first glance, making an industry more efficient sounds environmentally positive.
But efficiency can have complicated economic effects.
Suppose AI allows an oil producer to extract the same amount of fuel using less money, fewer equipment failures and better planning.
The company could simply maintain its existing production while enjoying lower costs.
However, it could also use those savings and productivity improvements to expand production.
Economists have studied versions of this phenomenon for generations: when technology makes a resource cheaper or easier to use, consumption does not always decline. Sometimes it increases.
The latest AI research applies similar reasoning across the energy economy.
Greater productivity in renewables can make clean electricity cheaper and more available.
Greater productivity in fossil fuels can make oil, coal and gas cheaper and more available.
AI therefore acts less like an inherently “green” technology and more like an amplifier.
Its environmental effect depends heavily on what it is amplifying.
Oil and Tech Companies Are Becoming More Connected
The relationship between artificial intelligence and fossil fuels is also becoming increasingly complicated because energy companies and technology companies increasingly depend on one another.
AI developers need enormous amounts of electricity for data centres.
Energy companies want sophisticated AI and cloud-computing systems to improve their operations.
One recent example highlighted by researchers involves Microsoft and Chevron. Chevron is developing natural-gas-powered infrastructure intended to supply electricity for data-centre operations, while the companies already have a longstanding technology relationship involving cloud and digital tools.
This illustrates what the researchers describe as a potentially reinforcing relationship.
AI infrastructure needs energy.
Energy producers can provide that power.
Those same producers can then use increasingly powerful AI systems to improve their own businesses.
Without careful climate policy, the cycle could support expansion on both sides.
Big Tech’s Climate Challenge Is Already Growing
The study arrives as major technology companies are facing increased scrutiny over the environmental footprint of artificial intelligence.
Microsoft, Amazon and Google have all invested heavily in expanding the computing infrastructure required for AI and cloud services. Their combined reported carbon emissions increased substantially in the financial year ending in 2026, with data-centre expansion and supply-chain activity contributing to the rise.
AI demand is also influencing electricity infrastructure.
The International Energy Agency expects renewables to provide nearly half of the additional electricity needed by data centres through 2030 in its base case, but natural gas and coal are also expected to contribute to meeting that demand.
This means AI’s climate footprint can operate through several channels at once.
There are emissions associated with constructing data centres and computer hardware.
There are emissions from generating the electricity those facilities consume.
And there may now be another category to consider: emissions generated by industries whose production becomes more efficient because of AI.
AI Fossil Fuel Emissions Challenge Current Carbon Accounting
Corporate sustainability reports traditionally focus on established categories of greenhouse gas emissions.
These include pollution produced directly by a company’s operations, emissions connected to purchased energy and emissions occurring throughout its wider supply chain.
The new research suggests that these measurements may not capture the full environmental influence of technology companies.
If a cloud provider gives an oil producer AI tools that substantially increase fossil fuel extraction, the resulting emissions are generally not treated as part of the technology company’s own carbon footprint.
Yet the researchers argue that understanding such enabled emissions could become important when evaluating AI’s overall climate impact.
That does not necessarily mean every tonne of emissions produced by an AI user’s business should simply be assigned to the technology provider.
The relationship is more complicated.
Instead, the study highlights the need to understand how digital technologies can influence emissions indirectly through the economic activity they make possible.
The Answer May Depend More on Policy Than Technology
Artificial intelligence itself does not decide whether it optimises a wind farm or helps identify another oil reservoir.
People, businesses and governments make those choices.
That means AI’s eventual climate impact may depend heavily on energy and environmental policy.
If economies strongly encourage renewable energy while restricting carbon emissions, AI productivity improvements could accelerate the transition toward cleaner power.
If fossil fuel production remains highly profitable and relatively unrestricted, the same technology can make those industries more efficient too.
The research therefore complicates claims that technological innovation alone will solve the climate crisis.
AI can improve almost any industrial system placed in front of it.
Whether that produces a cleaner economy depends on which industries are allowed, encouraged and financially rewarded to use those improvements.
AI Is an Amplifier, Not an Automatic Climate Solution
The debate surrounding artificial intelligence and climate change has often been framed around two competing narratives.
One says AI is environmentally costly because data centres consume large amounts of electricity.
The other argues AI will compensate for those emissions by making energy systems dramatically more efficient.
The latest research suggests reality may be more complicated.
Both can happen simultaneously.
AI can optimise solar farms while improving oil exploration. It can make electricity grids smarter while helping refineries operate more efficiently. It can reduce renewable downtime while making fossil fuel extraction cheaper.
The important question is therefore not simply whether AI increases efficiency.
It is what AI makes more efficient.
What the Study Means for AI’s Climate Future
AI fossil fuel emissions could become one of the less visible consequences of the technology’s rapid expansion.
Data-centre electricity use remains an important environmental concern, but measuring only the computers powering AI risks overlooking what those computers enable throughout the wider economy.
The study published in npj Climate Action offers a model rather than a guaranteed prediction of future emissions. Its results depend on assumptions about how quickly industries adopt AI and how strongly productivity changes influence energy markets.
Even so, its central message is difficult to ignore.
AI is powerful precisely because it can make industries more productive.
When those industries generate renewable electricity, improved productivity can support climate goals.
When those industries extract and sell fossil fuels, the same technological progress can work in the opposite direction.
Artificial intelligence may ultimately become an important tool in the transition to cleaner energy. But simply making AI more capable will not guarantee that outcome.
How governments, technology companies and energy producers decide to deploy it could determine whether AI becomes part of the climate solution—or helps extend the fossil fuel economy it was expected to replace.
