Key Takeaways
- Economic modeling suggests that if productivity in the fossil fuel and renewable energy industries improves by similar amounts, annual global carbon dioxide emissions would increase by approximately 470 million to 1.8 billion tonnes.
- To offset the additional carbon emissions caused by AI, productivity gains in the renewable energy sector would need to be roughly four to five times those in the fossil fuel sector.
- AI’s climate impact can vary depending on policy. Carbon pricing alone may not fully offset the additional emissions, and policies that limit the expansion of fossil fuel production also need to be considered.
October 24, 2026 | AI · Climate Change · Energy Transition
Artificial intelligence (AI) is attracting attention as a powerful technology for addressing the climate crisis. AI can be used to forecast solar and wind power generation, optimize power grid operations, and reduce energy waste in buildings and factories. Yet concerns are also growing, because the data centers that run AI consume enormous amounts of electricity. When considering AI’s environmental impact, we often ask:
Does AI save more energy than it consumes?
But perhaps this question alone is not enough. In August 2026, a study published in the international journal npj Climate Action raised an issue that had received relatively little attention: just as AI helps generate renewable energy, it is also being used to make oil and gas production easier and cheaper. The researchers analyzed these two effects from an economy-wide perspective. Across several scenarios assumed in the study, they found that AI generated more additional carbon emissions than it avoided.
AI Does Not Help Only Green Industries
One characteristic of AI technology is its ability to improve productivity across many industries. Here, higher productivity means producing more output with the same resources, or producing the same goods at a lower cost. Consider the renewable energy industry. AI can help predict wind speeds and sunlight levels more accurately, detect faults in generation equipment before they occur, and manage electricity production efficiently. These improvements can make renewable energy more competitive, helping it replace fossil fuel power generation. But the same technology also applies to the oil and gas industry. For example, AI can be used to locate underground oil and gas reserves more precisely, optimize drilling, and reduce the operating costs of production facilities. As production costs fall, oil and gas fields previously considered economically unviable may become worth developing. If fossil fuel supplies increase and prices fall, consumption may also rise. In other words, making individual industries more efficient through AI does not necessarily mean reducing carbon emissions for the planet as a whole.
What an Economy-Wide Simulation Revealed
To analyze these opposing effects, the researchers used a computable general equilibrium (CGE) model. The name may sound complicated, but the basic idea is relatively simple. It assumes that the industries, consumers, and producers that make up an economy are interconnected, and uses a computer to calculate how other industries and the economy as a whole respond when technology or prices change in one industry. The researchers assumed situations in which AI improves productivity in the fossil fuel and renewable energy industries, respectively. They also analyzed the effects of energy efficiency improvements in power grids and certain industries. The results were striking. According to the researchers’ model, productivity gains in the renewable energy sector would need to be roughly four to five times those in the fossil fuel sector to offset the additional carbon emissions caused by AI. In scenarios where AI improved productivity by similar amounts in the fossil fuel and renewable energy sectors, annual global carbon dioxide emissions were calculated to increase by 470 million to 1.8 billion tonnes. That is equivalent to approximately 1.2–4.8% of global energy-related carbon dioxide emissions in 2024. These results, however, arise from a particular model and set of assumptions; they are not a universal rule that applies to every real-world situation. Even so, they illustrate an important point: in a fossil fuel-centered economic structure like today’s, carbon emissions could actually rise if AI improves productivity by similar amounts in the fossil fuel and renewable energy industries.
Does Greater Technological Efficiency Benefit the Environment?
Underlying this is an issue long discussed in economics: the “rebound effect.” For example, when a car becomes more fuel-efficient, it needs less fuel to travel the same distance. But what if lower driving costs encourage people to use their cars more often? Some of the fuel savings may be offset by additional driving. The same applies to AI. If adopting AI lowers production costs in an industry, product prices may fall, and production and consumption may increase. Particularly if that industry relies on fossil fuels, higher productivity could actually increase total carbon emissions. AI can also change the economics of production itself—for example, by enabling the development of fossil fuel resources that were previously economically unviable. What matters about this study is that it looked not simply at how much energy AI saves, but also at how those efficiency gains change prices, production volumes, and consumption, and how those changes in turn affect carbon emissions.
So Is AI Harmful to the Climate?
We cannot draw that conclusion outright. This study does not demonstrate that AI has already caused that amount of additional carbon emissions in the real world. It is a scenario-based study that feeds assumptions about AI spreading across industries into an economic model and calculates the resulting outcomes. The research also has clear limitations. To begin with, the actual pace of AI adoption and the productivity gains in each industry remain uncertain. Nor is the model a dynamic one that directly tracks how technology and energy systems change over time. Emissions from data centers’ own electricity consumption were not included in the core simulations, and greenhouse gases other than carbon dioxide, including methane, were excluded from the analysis. The study also did not encompass every potential energy-saving effect of AI. Its findings should therefore not be treated as a final verdict on AI’s overall climate impact. Even so, the issue it raises is important: when assessing AI’s environmental impact, looking only at data centers’ electricity consumption or efficiency improvements in green technologies may not be enough.
Ultimately, Policy Matters More Than Technology
How, then, should we use AI? The researchers argue that encouraging advances in renewable energy technology alone is not enough to improve AI’s climate impact. We also need to account for AI’s role in expanding fossil fuel production and consider measures such as carbon pricing and policies that constrain fossil fuel supply. Among the measures the researchers compared in their model, carbon pricing had the greatest effect on reducing additional emissions. Applying a carbon price reduced the additional emissions caused by AI. Nevertheless, in scenarios where AI improved productivity by similar amounts in the fossil fuel and renewable energy sectors, even a high carbon price did not fully eliminate the increase in emissions. This suggests that carbon pricing alone may not be enough, and that policies limiting AI-enabled expansion of fossil fuel production also need to be considered. The study leaves us with important policy questions. Where, and toward what ends, should we steer the use of advances in AI technology? And what economic and institutional conditions are needed for gains in technological efficiency to translate into environmental benefits? When a new technology emerges, we often ask how it will change society. But perhaps the more important question is this: In what direction will we choose to let new technology reshape society? AI’s future will not be determined by its technical capabilities alone. It will also depend on which industries use it, which economic incentives are at work, and what rules governments and societies put in place. Technological progress does not automatically mean sustainable development. Ensuring that it leads to sustainable outcomes remains our responsibility.