Chinese cities with more advanced AI development show stronger industrial green transformation driven by green innovation and agglomeration, but gains concentrate locally and appear to slow green progress in neighbouring cities.

Main Finding

AI development significantly promotes industrial green transformation in Chinese cities (2004–2023). The effect operates mainly through accelerating green technological progress and strengthening economic agglomeration; the expected energy-intensity reduction channel was not supported. The positive impact is stronger in coastal areas, non-resource-based cities, and urban agglomerations. Spatially, AI raises local cities’ industrial green transformation but generates negative spillovers on neighbouring cities.

Key Points

• Core result: AI development → positive and significant increase in industrial green transformation.
• Mechanisms:
  • Supported: promotion of green technological progress; enhancement of economic agglomeration.
  • Not supported: reduction in energy intensity (no robust evidence that AI has yet lowered energy intensity to drive the green transition).
• Heterogeneity:
  • Larger effects in coastal regions than inland.
  • Stronger effects in non-resource-based cities than in resource-dependent cities.
  • Greater impact inside urban agglomerations than in non‑agglomeration areas.
• Spatial effects:
  • Positive direct effects within the city.
  • Negative spatial spillovers to neighbouring cities, consistent with resource/talent siphoning and possible relocation of polluting activities.
• Policy relevance: coordinating regional policies is necessary to capture AI’s green gains while mitigating adverse cross‑regional effects.

Data & Methods

• Data
  • Panel of 283 prefecture-level (and above) Chinese cities, 2004–2023.
  • City-level raw data collection included patent data (via Python web scraping) and enterprise information from Tianyancha.
  • Contextual data point: China’s manufacturing robot density reached 470 units per 10,000 workers in 2023 (illustrating rapid AI/automation uptake).
• Key variables
  • Industrial green transformation (IGT): two-dimensional construct
    • Structural green transition: 100% minus the share of newly registered enterprises in high‑polluting industries (high pollution industries identified from SO2, COD, and solid waste shares across industry categories).
    • Efficiency green transition: measured as the reciprocal of SO2 emissions per unit of industrial value‑added.
  • AI development (intel): composite indicator constructed from multiple dimensions (infrastructure, technological application, market benefits) using city-level patent and firm data (authors compiled an AI development index).
  • Mechanism proxies: measures of green technological progress, energy intensity, and economic agglomeration.
• Econometric approach
  • Baseline: two-way fixed effects regressions (city and year fixed effects).
  • Mechanism tests: regressions of mechanism variables on AI to test mediation (following standard mediation approach).
  • Heterogeneity: subgroup analyses and interaction term specifications across geographic (coastal vs inland), resource endowment (resource-based vs non-resource-based), and spatial organization (urban agglomeration vs non).
  • Spatial analysis: spatial Durbin model with an economic–geographical weight matrix (combined reciprocal Euclidean distance and reciprocal difference in per-capita GDP; equal weights).
  • Endogeneity/robustness: two-stage least squares (2SLS) and other robustness checks reported.
  • Other methods/tools: coupling coordination degree model (used in analysis framework), Python for data collection.

Implications for AI Economics

• For policy design
  • Prioritize AI investments that explicitly target green technological innovation (R&D subsidies, support for AI-driven clean-tech startups) because green-tech diffusion is a key channel linking AI to industrial greening.
  • Leverage AI to foster beneficial agglomeration effects (cluster policies, shared digital/green infrastructure) while managing concentration risks.
  • Anticipate and mitigate negative spatial spillovers: implement regional coordination (transfer payments, talent/capital mobility policies, joint environmental standards) to prevent talent/capital siphoning and pollution relocation.
  • Since the energy‑intensity channel is not yet evident, complement AI adoption with targeted energy-efficiency policies (retrofits, standards, incentives) to realize expected energy savings.
• For firms and regions
  • Firms should combine AI adoption with process redesign and energy-management practices to capture both efficiency and emissions benefits.
  • Resource-dependent and inland cities need tailored support (skill development, diversification incentives, digital infrastructure) to realize AI’s green potential.
• For research
  • City-level, long-panel analyses and spatial econometric tools are valuable for studying technology–environment interactions—future work should continue to use (and refine) composite AI measures, address identification (instruments, natural experiments), and explore why energy-intensity gains lag.
  • Investigate mechanisms of negative spillovers more granularly (talent flows, capital allocation, industrial relocation) and evaluate policy interventions that internalize cross‑regional externalities.

Summary takeaway: AI is a promising lever for industrial green transformation in China, chiefly by spurring green innovation and agglomeration economies, but policymakers must address regional disparities and negative spatial spillovers and combine AI adoption with explicit energy‑efficiency measures to achieve broader, equitable green outcomes.