AI adoption is associated with higher urban green productivity across Chinese cities, but the benefits are uneven: some regions and city types see little or negative effects. Gains appear to operate via stronger green finance and upgraded productive forces and also spill over to nearby cities.

Main Finding

AI adoption significantly improves urban green total factor efficiency (GTFE) in Chinese cities (279 cities, 2012–2021), but the effect is nonlinear (thresholds), heterogeneous across city types/regions, operates through green finance and new-quality productive-force channels, and generates positive spatial spillovers to neighboring cities.

Key Points

• Positive aggregate effect: AI adoption is associated with higher urban GTFE overall.
• Nonlinear threshold: The AI→GTFE relationship exhibits a threshold effect (AI’s marginal impact depends on reaching certain levels or conditions).
• Robustness: Results hold under multiple robustness checks, including use of machine-learning models and alternative econometric specifications.
• Heterogeneity: The AI effect varies by geography and urban characteristics:
  • Not statistically significant in western China, mega-cities, large cities, and central cities.
  • An insignificantly negative effect observed in central-eastern China and in old industrial bases.
  • Differences also seen by urban hierarchy, city size, and transportation hub status.
• Mechanisms: Evidence indicates two main channels:
  • Upgrading of green finance development (better allocation of green capital).
  • Advancement of “new-quality productive forces” (technology-driven productivity/composition changes).
• Spatial spillovers: AI increases local GTFE and has positive spillover effects on neighboring localities’ GTFE.
• Policy implication from the study: Targeted, region- and city-type–specific policies are recommended to maximize AI’s green productivity benefits.

Data & Methods

• Data: City-level panel across 279 Chinese prefecture-level cities, 2012–2021.
• Outcome: Urban green total factor efficiency (GTFE) — measured at the city level (study uses standard GTFE metrics; details depend on paper’s computational choice).
• Main explanatory variable: City-level AI development/adoption indicator(s).
• Empirical approaches reported:
  • Nonlinear threshold modeling to detect regime-dependent effects.
  • Robustness checks using both econometric specifications and machine-learning models.
  • Heterogeneity analysis via subsample regressions (region, city size, hierarchy, transport hub status, industrial base).
  • Mechanism tests to examine mediation via green finance and new-quality productive forces.
  • Spatial econometric analysis to identify spillover effects (spatially lagged effects).
• Notes on interpretation: Threshold and heterogeneity results imply complementarities and prerequisites (e.g., institutional, financial, or infrastructure conditions) that shape AI’s green productivity impacts.

Implications for AI Economics

• Complementarities matter: AI’s green-productivity gains depend on interacting factors (green finance, productive-force upgrades). Policymaking should target these complementarities rather than treating AI adoption in isolation.
• Nonlinearity and prerequisites: Threshold effects imply returns to AI are conditional — policy should focus on enabling conditions (finance, human capital, infrastructure) to move cities into regimes where AI yields positive GTFE returns.
• Distributional and regional policy design: Heterogeneous effects across regions and city types signal potential uneven environmental and economic benefits. Targeted regional policies and coordination (especially to assist lagging regions and old industrial bases) are essential.
• Spatial spillovers and coordination: Positive spillovers mean that city-level AI investments generate neighborhood benefits; coordinated regional planning can amplify gains and avoid free-rider problems.
• Measurement and method lessons: Combining machine-learning robustness checks with spatial and nonlinear econometric methods is valuable for capturing complex AI→economic-environment relationships.
• Research priorities: Micro-level causal identification (firm or plant data), long-term dynamics, interaction with labor markets and inequality, and evaluation of specific green-finance instruments will sharpen policy prescriptions and causal understanding.