China's AI pilot zones lift cities' green economic efficiency by spurring green innovation and reshaping industry; gains are concentrated in inland and non-resource cities and grow where government and public environmental attention are high.

Main Finding

China’s National New-Generation Artificial Intelligence Innovation Development Pilot Zone (NAIDPZ) policy causally increased urban green economic efficiency (GEE). The positive effect operates mainly through boosting green technological innovation and optimizing industrial structure, and is stronger in inland, central, and non–resource-based cities; it is amplified where government attention and public environmental concern are higher.

Key Points

• Research design: NAIDPZ treated as a quasi‑natural experiment and analyzed with a staggered difference‑in‑differences (DiD) model on city panel data.
• Sample: 267 Chinese prefecture‑level cities observed from 2007 to 2023.
• Primary result: NAIDPZ implementation led to a statistically significant increase in city‑level GEE.
• Mechanisms identified:
  • Green technological innovation (e.g., more green patents / innovation activity) is an important mediator.
  • Industrial structure optimization (movement toward less carbon‑intensive industries) is another key channel.
• Moderation / boundary conditions:
  • Stronger policy effects where local governments pay greater attention to AI/green issues.
  • Stronger effects when public environmental awareness is higher.
  • Larger impacts in inland and central region cities and in non–resource‑based cities.
• Robustness: Authors report multiple checks (parallel‑trend tests, alternative samples and specifications, placebo and sensitivity analyses) supporting the causal interpretation.

Data & Methods

• Data: City‑level panel (267 prefecture‑level cities) spanning 2007–2023. Outcome: green economic efficiency (GEE) at the city level.
• Treatment: Cities designated as NAIDPZ (timing varies across cities → staggered rollout).
• Identification strategy: Staggered difference‑in‑differences exploiting the phased introduction of NAIDPZ to estimate causal effects on GEE.
• Mechanism tests: Mediation/stepwise analyses linking NAIDPZ to green technological innovation indicators and measures of industrial structure change.
• Heterogeneity and moderation: Interaction analyses to test effects by region (inland vs. coastal; central), resource endowment (resource‑based vs. non‑resource), and by measures of government/public environmental attention.
• Robustness checks: Multiple specifications and sensitivity tests (authors report checks such as parallel‑trend testing, alternative outcome definitions/samples, placebo tests, etc.) to validate findings.

Implications for AI Economics

• Policy effect: Well‑designed AI industrial policies can raise environmental efficiency, not only output — AI adoption can be a driver of low‑carbon urban transformation when linked to innovation and sectoral upgrading.
• Mechanisms matter: Models of AI’s economic and environmental impacts should endogenize technological change (green R&D) and structural shifts in industry to capture the full benefit of AI policies.
• Governance and demand side: The effectiveness of AI policy depends on institutional and social context — government priorities and public environmental awareness amplify benefits. Evaluations that ignore these moderators may misstate returns.
• Targeting and equity: Heterogeneous effects imply that location and resource endowment shape payoff to AI investments; policymakers should tailor AI/green policies to local conditions (e.g., prioritize capacity building in resource‑rich or coastal cities where effects are smaller).
• Caution and generalizability: Results are empirically robust for China’s institutional setting and phased pilot program; extrapolation to other countries requires attention to governance, market structure, and regulatory environments. Future work should investigate long‑run effects, possible rebound effects, distributional outcomes, and interactions with energy and climate policy instruments.