A new study of 272 Chinese cities reveals that while AI contributes meaningfully to carbon productivity, its benefits are being undermined by a rebound effect where efficiency gains trigger increased energy consumption overall. Researchers found that AI accounts for 7.46% of carbon productivity improvement, yet the technology's high energy demands and rapid expansion risk negating those gains through what economists call the "rebound effect." The paradox emerges because AI's efficiency benefits encourage more energy-intensive applications, ultimately driving up total carbon emissions despite technological progress.

What Is the Carbon Productivity Paradox?

The carbon productivity paradox describes a troubling disconnect between AI's theoretical promise and its real-world climate impact. While AI can optimize resource allocation, reduce waste, and enable precision decision-making across industries, the technology itself is extraordinarily energy-intensive. Training a single large-scale AI model generates more than 284 metric tons of carbon dioxide emissions, and data centers globally consume over 200 terawatts of electricity annually. As AI applications expand rapidly, the efficiency gains they deliver are being overwhelmed by surging demand for computing power and complementary products, creating a net increase in energy consumption and emissions rather than the decrease many expected.

This paradox is particularly acute in developing nations like China, where economic structures and technological capabilities may not yet be optimized to harness AI's full potential. The research suggests that without deliberate intervention, countries risk investing heavily in AI infrastructure while seeing minimal climate benefits, or even experiencing net increases in carbon emissions.

How Can Economies Escape the Carbon Productivity Paradox?

The research identifies a critical pathway forward: adaptive capability. Rather than viewing AI as a standalone technology, the study emphasizes that AI's carbon productivity benefits depend on complementary investments and structural changes across three dimensions:

• Technological Matching: Aligning AI deployment with existing industrial capabilities and infrastructure to maximize efficiency gains without creating new bottlenecks.
• Organizational Enabling: Building workforce skills, management practices, and institutional frameworks that allow organizations to fully leverage AI's potential rather than deploying it superficially.
• Environmental Supporting Capabilities: Developing digital infrastructure, data marketization systems, and market mechanisms that enable AI to operate at scale efficiently.

The study found that industrial structure upgrading, market scale expansion, and digital infrastructure investment have the most critical complementarity effects in unlocking AI's technological dividends. When these adaptive capabilities are in place, the threshold at which AI transitions from harming carbon productivity to improving it shifts significantly leftward, allowing economies to escape the paradox earlier and more decisively.

Why Does AI's Energy Consumption Keep Growing?

The rebound effect explains much of the paradox. When AI makes energy-intensive processes more efficient, the cost of those processes drops, which typically triggers increased demand. For example, if AI optimizes manufacturing to reduce energy per unit produced, manufacturers may expand production, ultimately consuming more total energy. This dynamic has played out repeatedly in technological history, from automobiles to air conditioning, and it's now occurring with AI at a global scale.

Additionally, the AI industry itself is expanding explosively. The global AI market reached $136.55 billion in 2022 and is projected to grow at a compound annual rate of 37.3%, meaning the sheer volume of AI systems being deployed is outpacing efficiency improvements in individual systems. Experts project that AI applications alone could consume up to 21% of global electricity by 2030, a staggering figure that underscores the urgency of addressing the carbon productivity paradox.

What Role Does Data Marketization Play?

The research reveals that data marketization allocation actually outpaces AI itself in driving carbon productivity improvements, contributing 16.27% compared to AI's 7.46%. This suggests that creating efficient markets for data, enabling its optimal allocation across industries, and breaking down information silos may be more impactful than raw AI deployment. When combined with labor structure improvements, which contributed 13.27%, the findings point toward a holistic approach: AI works best when embedded within broader economic reforms that improve how societies allocate resources, develop talent, and share information.

The implication is clear: simply deploying more AI without addressing underlying economic structures and market inefficiencies will not solve the carbon productivity paradox. Instead, policymakers and industry leaders must view AI as one tool within a larger ecosystem of complementary capabilities.

What Does This Mean for Global Climate Goals?

The stakes are extraordinarily high. The global carbon budget for limiting warming to 1.5 degrees Celsius has shrunk from approximately 500 billion tons in 2020 to roughly 250 billion tons in 2023. At current annual emissions of about 54 billion tons, there is a 50% probability the budget will be exhausted before 2030, risking warming of 1.7 degrees Celsius within 15 years. In this context, the carbon productivity paradox is not merely an academic curiosity; it represents a critical barrier to achieving dual-carbon goals that balance economic growth with emission reductions.

The research offers theoretical support for a path forward: unlocking AI's potential to resolve the carbon productivity paradox requires deliberate investment in adaptive capability. Economies that simultaneously upgrade industrial structures, expand market scale, and build digital infrastructure will be positioned to harness AI's efficiency benefits while avoiding the rebound effect trap. Those that treat AI as a standalone technology, without these complementary investments, risk accelerating emissions growth even as they deploy more sophisticated AI systems.