Hidden carbon cost of AI is testing the green growth narrative

New research suggests that AI's environmental impact may be far more complicated than the dominant green-technology narrative implies. The study published in Sustainability finds that AI innovation can raise carbon emissions in highly innovative economies when it expands energy demand faster than efficiency gains can offset it.

Titled The Green Illusion? How Artificial Intelligence and Finance Shape Environmental Futures, the study examines the world's 15 most innovation-driven economies and uses Method of Moments Quantile Regression to measure how AI innovation, financial development and energy consumption affect carbon emissions across different emission levels.

AI innovation is not automatically green

The study challenges the assumptions that digital innovation will naturally help countries cut emissions. AI can improve energy efficiency, optimize industrial processes, support renewable energy systems, reduce material waste and improve environmental monitoring. However, the research finds that these gains do not always translate into lower carbon emissions, especially in economies where AI deployment depends on energy-intensive infrastructure.

The analysis focuses on 15 highly innovative economies: Canada, China, Denmark, Finland, France, Germany, Israel, Japan, the Netherlands, Singapore, South Korea, Sweden, Switzerland, the United Kingdom and the United States. These countries are important because they sit at the center of global innovation, finance and digital infrastructure. They are also among the economies most likely to shape how AI is used in energy systems, industrial production, finance, trade and climate policy.

The findings show a clear asymmetry. In countries or conditions with lower carbon emissions, AI innovation has a limited and statistically weak effect on emissions. However, at higher emission levels, the effect becomes stronger and statistically significant. In the upper emission quantiles, AI innovation is associated with rising carbon emissions, suggesting that AI can deepen environmental pressure when deployed in energy-heavy economies.

The finding points to what the authors call as a green illusion. AI may appear clean because it is digital, intangible and often tied to efficiency. But behind many AI systems are data centers, large-scale computation, cloud infrastructure, model training, data storage and cooling systems. These systems consume large amounts of electricity. If that electricity comes from fossil-fuel-heavy grids, AI expansion can raise emissions even while individual processes become more efficient.

The study also connects this problem to the rebound effect. When AI improves efficiency, production costs can fall. Lower costs can stimulate more output, more consumption and greater energy demand. In that case, the efficiency gain may be partly or fully offset by expanded activity. A technology that helps one factory, building or logistics system use energy more efficiently may still raise total emissions if it supports broader economic expansion powered by carbon-intensive energy.

This finding is notable because governments and corporations increasingly present AI as a key part of climate strategy. AI is used in smart grids, power forecasting, energy management, industrial automation, renewable integration, transport optimization and emissions tracking. The study does not dismiss those benefits. Instead, it warns that the environmental outcome depends on the energy system behind the technology and the sectors where AI is deployed.

In high-emission economies, AI may be used heavily in manufacturing, logistics, finance, data services and energy-intensive production. If those systems are not powered by low-carbon electricity, the net environmental effect can become negative. The study therefore argues that AI policy cannot be separated from energy policy. Digital transformation must be evaluated not only by productivity gains or innovation rankings, but also by its carbon footprint.

The research also shows that energy consumption serves as a central transmission channel. AI affects carbon emissions partly through its impact on energy demand. In the model examining energy consumption, AI has a positive and statistically significant effect in lower energy-consumption quantiles, but that effect weakens as energy use rises. This suggests that AI can create new energy demand when digital infrastructure is expanding, while its added effect may become less visible in countries where energy use is already high.

AI innovation must be powered and governed differently if it is to contribute to environmental sustainability. Countries cannot assume that more AI will automatically mean fewer emissions. They need cleaner grids, energy-efficient data centers, stronger reporting of AI-related electricity demand and incentives that push AI toward low-carbon applications rather than unchecked computational expansion.

Finance can cut emissions, but its green impact weakens in high-emission economies

The study also examines financial development, a factor often linked to sustainable growth because deeper financial systems can channel capital into cleaner technologies, renewable energy, energy efficiency and green infrastructure. The findings show that financial development can support environmental improvement, but its benefits are uneven and weaken as emission levels rise.

In lower and middle emission quantiles, financial development is associated with lower carbon emissions. This suggests that access to stronger financial systems can help countries fund cleaner investments, support sustainable business models and expand environmental technologies. Finance can play a useful role when capital flows into renewable energy, low-carbon transport, energy-efficient buildings, clean industry and green innovation.

However, this effect fades in higher emission quantiles. In countries with higher carbon emissions, the negative relationship between financial development and emissions becomes weaker and loses statistical strength. In other words, finance may help reduce emissions in cleaner or moderately emitting economies, but it is not powerful enough on its own to overcome high-carbon industrial structures.

The result complicates the green finance narrative by showing that finance can grow without delivering environmental benefits. If credit and investment flow into fossil-fuel-based production, energy-intensive industries, large infrastructure expansion or high-consumption sectors, financial deepening can reinforce emissions. If capital is guided by strict sustainability criteria, it can support emissions reduction. The difference depends on regulation, market incentives and the direction of investment.

Financial systems need environmental rules, not just expansion. More lending, deeper capital markets and broader financial access may not improve sustainability unless finance is aligned with low-carbon goals. Green bonds, sustainable banking rules, climate-risk disclosure, green investment funds and restrictions on high-carbon lending can determine whether financial development becomes a climate tool or a source of added ecological pressure.

The relationship between finance and energy consumption adds another layer. The study finds that financial development has a negative but statistically weak effect on energy consumption in lower quantiles, while its direction becomes positive as energy consumption rises. This suggests that finance may support efficiency or low-carbon investments in some contexts, but can also fund energy-intensive growth when countries cross certain thresholds.

This pattern reinforces calls for targeted policy. In lower-emission economies, financial development may already be more capable of supporting cleaner investment. In high-emission economies, green finance must be much more deliberate. Financial systems in carbon-heavy countries need stronger screening of projects, tighter climate-risk rules and incentives that shift capital away from polluting sectors.

Additionally, a strong unidirectional causality was found between financial development and carbon emissions and from financial development to energy consumption. This means finance is not just a background condition, it is a driver of environmental outcomes. The direction of capital allocation shapes how much energy economies consume and how much carbon they emit.

Economic growth, meanwhile, does not show a statistically significant direct effect on carbon emissions across the model. This does not mean growth is irrelevant. Rather, the study suggests that the environmental impact of growth depends on structure. Growth based on clean technologies, services and efficient infrastructure can have a different environmental profile from growth based on fossil fuels, heavy industry and carbon-intensive consumption.

Trade openness also shows no statistically significant direct effect on emissions in the carbon model. The result points to a similar structural issue. Trade can improve environmental outcomes when it spreads green technologies and cleaner production practices. It can worsen them when it expands energy-intensive exports, carbon-heavy supply chains or pollution-intensive production. Trade volume alone does not determine the environmental result.

Population is the strongest and most consistent pressure factor. It has a positive and statistically significant effect on carbon emissions across all quantiles. The outcome is that population growth, urbanization and rising consumption continue to increase pressure on energy, transport, housing, agriculture and infrastructure. For densely populated and rapidly urbanizing economies, emissions policy must therefore include sustainable cities, public transport, energy-efficient buildings and resource-efficient infrastructure.

Green digitalization needs cleaner energy and stricter policy design

Environmental sustainability cannot be achieved through AI innovation or financial development alone. Both can help, but both can also worsen environmental outcomes if they are not governed through energy-aware and climate-aligned policies.

The study finds that energy is not merely another variable in the emissions equation. It is the channel through which AI and finance shape environmental outcomes. AI raises computing and infrastructure demand. Finance funds production, consumption, green investment or fossil expansion. Energy systems then determine whether those activities produce high or low emissions.

Countries need to integrate digital policy, financial policy and energy policy. AI strategies should be designed with electricity demand, grid composition and carbon intensity in mind. Financial development should be evaluated not only by credit growth or market depth, but by whether it shifts capital toward low-carbon systems. Energy transition policy should account for the digital economy's growing power needs.

The findings are crucial because the sample consists of highly innovative economies. These are not countries lacking technological capacity. They have advanced research systems, strong financial markets and sophisticated policy institutions. If AI can still raise emissions in this group, the risk may be greater in countries where electricity grids are more carbon-intensive and regulatory systems are weaker.

The authors call for green digitalization, meaning AI systems should be built and deployed with environmental performance as a core requirement. That includes powering data centers with renewable energy, improving hardware efficiency, reducing wasteful computation, measuring the carbon footprint of AI services and prioritizing AI applications that directly support emissions reduction.

The study also points to the need for more careful governance of AI infrastructure. AI-related electricity use is rising as models become larger, data processing expands and cloud services grow. Policymakers may need reporting standards for AI energy consumption, energy-efficiency requirements for data centers and incentives for locating digital infrastructure in regions with cleaner power grids.

On the finance side, the study supports stronger green macroprudential policy and sustainability-oriented central banking. These approaches can guide credit allocation, require climate-risk assessment and discourage lending to projects that lock in high emissions. Financial institutions may also need stricter rules for defining green investments, because weak standards can allow capital to flow into projects that claim sustainability benefits without meaningful emissions reductions.

The paper also stresses the value of country-specific policy. A one-size-fits-all approach is unlikely to work because AI, finance and energy systems interact differently across emission levels. Countries with low emissions may need to ensure that digital expansion does not create new energy burdens. Countries with high emissions need stronger controls to prevent AI and finance from reinforcing carbon-intensive structures.

Limitations

The study measures AI innovation through AI-related patent publications, which captures inventive activity but not the full range of AI deployment. It does not distinguish whether AI is used in industry, transport, agriculture, finance, consumer services or other sectors. Different applications may have very different energy and emissions effects. Future research, the authors argue, should separate AI use by sector and connect it with sector-specific energy intensity.

The study is also limited to 15 highly innovative countries, meaning the findings should not be directly generalized to low- and middle-income economies. In less developed economies, AI infrastructure, finance, energy systems and regulatory capacity may differ substantially.