Digital innovation can cut the carbon cost of growth

How can countries continue to grow while reducing the carbon cost of that growth? Carbon productivity offers one answer because it measures whether economies are generating more value from each unit of emissions.

AI innovation is emerging as a major tool for improving carbon productivity, helping economies generate more output with fewer emissions, but its climate benefits depend on how the technology is deployed and how clean the underlying energy system is, according to new research that shows that AI can support low-carbon growth through green innovation, lower transaction costs and wider public and business attention, while also warning that energy-intensive AI infrastructure can weaken those gains if powered by fossil fuels.

The study, titled How Does AI Technology Innovation Boost Carbon Productivity? Evidence from China, was published in Sustainability. The researchers analyzed 229 prefecture-level and above cities in China from 2007 to 2023 and found that AI technology innovation, measured through AI patent stocks, significantly improves carbon productivity, defined as real GDP produced per unit of carbon dioxide emissions.

AI links economic output and emissions reduction

According to the research, AI innovation is a general-purpose technology that can raise economic output by improving production efficiency, optimizing factor allocation and creating new business models. It can simultaneously support emission reduction by improving energy management, cleaner production, supply-demand matching and industrial decision-making.

China provides a great example because it combines rapid AI development, large industrial systems, major regional differences and ambitious climate targets. The study finds that Chinese cities with stronger AI innovation capacity also showed higher carbon productivity. The result remained stable across several tests, including changes in patent measurement, lagged variables, spatial spillover controls and methods designed to address endogeneity.

The outcome is not limited to China. Economies investing heavily in AI may see similar gains if innovation moves beyond patents and pilot projects into practical low-carbon applications. AI can help factories reduce waste, cities manage transport and energy systems, firms coordinate supply chains, and consumers access more efficient services. But these gains depend on whether AI is integrated into real production and governance systems.

The researchers identify three channels that explain the link.

• AI supports green innovation by lowering technical complexity, improving research efficiency and helping firms manage uncertainty in low-carbon investment.
• AI lowers transaction costs by improving information flows, reducing market frictions, automating transactions and strengthening supply-chain coordination.
• AI raises attention to AI itself, encouraging governments, firms and consumers to adopt intelligent tools more widely.

This is crucial because carbon productivity is not shaped only by technology installed inside factories. It is also shaped by how efficiently markets work, how quickly firms adopt cleaner systems and how strongly governments and investors prioritize low-carbon transformation. AI can help connect these pieces, turning technical innovation into wider economic and environmental gains.

China shows the gains are uneven

The study finds that AI's effect on carbon productivity differs across regions, innovation types and institutional settings. In China, the benefits were stronger in eastern regions, central cities and resource-based cities. These differences point to a wider global issue: AI's climate benefits are unlikely to be evenly distributed unless policy actively supports weaker regions and slower adopters.

Eastern Chinese cities benefited more because they tend to have stronger innovation ecosystems, more advanced industries, better digital infrastructure and greater talent concentration. Central cities also gained more because they have stronger administrative capacity, deeper patent activity and more opportunities to embed AI into transport, energy and industrial systems.

Resource-based cities showed particularly strong gains, likely because they have more room to improve. Cities dependent on mining, energy extraction or heavy industry often have higher emissions baselines. In those settings, AI-driven improvements in energy efficiency, industrial optimization and resource management can deliver larger carbon productivity gains.

The findings suggest that countries should avoid a one-size-fits-all AI climate policy. Advanced urban centers may be ready to scale AI-driven green technologies quickly. Industrial and resource-dependent regions may need AI tools focused on energy efficiency, process optimization and industrial upgrading. Less-developed regions may first need stronger infrastructure, skills, financing and digital adoption capacity.

Not all AI innovation produces the same carbon productivity effect, the study notes. AI innovations tied directly to application fields had the strongest impact, followed by functional applications and then foundational techniques. This indicates that applied AI in sectors such as energy, transport, manufacturing and agriculture may deliver faster climate benefits than abstract algorithmic advances alone.

Research institutions and universities also showed stronger carbon productivity effects than enterprises, even though companies filed more AI patents. The likely reason is that academic and research bodies may be more closely aligned with long-term low-carbon goals, while firms often prioritize commercial returns. The authors therefore emphasize cooperation among industry, universities and research institutes.

Collaborative innovation performed better than non-collaborative innovation. That finding is especially relevant for climate policy because low-carbon transformation usually requires multiple fields to work together. AI must connect with energy systems, industrial engineering, environmental governance, finance and urban planning. Collaboration can help move AI from technical invention to practical deployment.

Clean power decides whether AI becomes a climate solution

AI's climate value depends on energy structure. AI systems require electricity, computing infrastructure, data centers and continuous model development. If those systems are powered mainly by fossil fuels, their energy burden can offset part of the carbon productivity gains they create elsewhere.

The study finds that clean energy plays a threshold role. AI innovation improves carbon productivity across different energy conditions, but the effect becomes much stronger as the share of clean energy generation rises. In regions with low clean-energy penetration, the benefit is weaker. Once clean energy crosses key thresholds, AI's positive effect grows sharply. This challenges a simple pro-AI climate narrative. AI is not automatically green. It can make production cleaner, but it can also raise electricity demand. Its net contribution depends on whether power systems are decarbonizing at the same time.

AI strategy and energy strategy must be planned together, the study holds. Expanding AI infrastructure without expanding clean energy can limit environmental gains. Building clean power without using AI for industrial and system optimization may also miss productivity opportunities. The strongest approach is to pair AI innovation with clean electricity, green data centers and targeted low-carbon applications.

The study's time-based results also show that AI's effect is not fixed. In China, AI innovation had no significant effect on carbon productivity before 2009, became significantly positive from 2010, peaked around 2015 and then gradually weakened. During the COVID-19 period, the effect was no longer significant, likely because production, technology application and governance systems were disrupted.

That pattern shows why policymakers should treat AI as a dynamic tool, not a permanent shortcut. Early deployment may bring large gains, but later improvements can slow as the easiest efficiency gains are exhausted. External shocks can also interrupt AI's contribution. Continuous monitoring, updated incentives and renewed application strategies are needed to keep AI aligned with low-carbon goals.

The paper also warns that patent data can show innovation capacity, but not every patent becomes a practical technology. Some AI patents may remain technical concepts or research reserves without measurable carbon reduction effects. Future research will need to connect AI patents with real-world firm behavior, energy consumption, emissions data and green performance.