China’s AI pilot zones raise urban land green-use efficiency by spurring green innovation, shifting labor and upgrading industry; gains are largest in big, digitally well-equipped cities and spill over to nearby regions.

Main Finding

The paper finds that China’s National New Generation AI Innovation and Development Pilot Zones (AIPZ) causally increase urban land green use efficiency (ULGUE). Effects are robust and operate primarily through green technology innovation, labor-structure optimization, and industrial-structure upgrading. Impacts are larger in provincial capitals/municipalities, big cities, and places with better digital infrastructure, and the policy generates positive spatial spillovers to neighboring cities.

Key Points

• Treatment: Establishment of AIPZ (started 2019; 18 pilot cities in eight batches) treated as a quasi-natural experiment.
• Core result: AIPZ significantly raise ULGUE (higher economic output per unit land while reducing undesirable environmental outputs).
• Mechanisms:
  • Green technology innovation: AIPZ spur R&D, tech diffusion, and green innovation that reduce emissions and improve resource intensity.
  • Labor structure optimization: AI increases demand for higher-skilled tasks, raising workforce quality and enabling more efficient/intelligent land-use practices.
  • Industrial structure upgrading: AIPZ promote clustering and shift toward mid-high-end, service- and tech-intensive industries, improving land productivity and lowering pollution intensity.
• Heterogeneity: Stronger effects in municipalities/provincial capitals, large cities, and cities with higher digital infrastructure levels.
• Spatial effects: Positive spillovers—AIPZ adoption improves ULGUE in both treated cities and neighboring jurisdictions.
• Contribution: First causal, policy-oriented evidence linking AI pilot policy to environmental outcomes embodied in land use efficiency.

Data & Methods

• Data: Panel of 286 prefecture-level and above Chinese cities, annual data covering 2015–2023.
• Outcome measurement: Urban land green use efficiency (ULGUE) computed using a super-efficiency SBM (slack-based measure) model that incorporates desirable outputs (economic/social/ecological benefits) and undesirable outputs (pollution).
• Identification strategy: Multi-period Difference-in-Differences (DID) exploiting staggered roll-out of AIPZ as a quasi-natural experiment.
• Mechanism tests: Mediation/mediating-variable regressions for green tech innovation (R&D, patents, tech indicators), labor-structure measures (skill composition, employment shares), and industrial upgrading (sectoral shares, value-added composition).
• Robustness checks: Battery of robustness tests reported (parallel trends tests, alternative specifications, likely placebo tests).
• Spatial analysis: Tests for spatial spillovers (spatial econometric specifications or spatial DID extensions) to capture effects on neighboring cities.
• Controls: Standard city-level controls (economic, demographic, infrastructure variables) and fixed effects to address time-invariant heterogeneity.

Implications for AI Economics

• AI policy as environmental policy instrument: National AI pilot zones can produce "positive policy externalities"—raising productivity while reducing environmental intensity—so AI-targeted industrial policy can be part of green policy mixes.
• Mechanism-rich effects: Benefits are transmitted not just via productivity gains but also by changing innovation trajectories, workforce composition, and sectoral structure; economic models of AI impacts should incorporate these channels (green innovation, skill upgrading, structural change).
• Complementarity with digital infrastructure and human capital: Local absorptive capacity matters—digital infrastructure and labor skills amplify benefits. Policy design should bundle AI investment with digital backbone and training.
• Spatial coordination: Positive spillovers imply regional coordination (e.g., zoning, infrastructure investment, cross-jurisdictional incentives) can increase aggregate returns from AI policy and avoid unequal geographic diffusion.
• Distributional and transitional considerations: While net effects on ULGUE are positive, the paper acknowledges substitution effects on some labor sectors—AI-driven restructuring requires active re-skilling and social policies to manage transitions.
• Research and policy agenda:
  • For researchers: extend causality checks to firm- and household-level outcomes, long-run dynamics, and cost–benefit comparisons with conventional environmental policies.
  • For policymakers: prioritize complementary investments (digital infra, green R&D, workforce training), tailor pilot design to city characteristics, and leverage spatial planning to maximize both productivity and environmental gains.

Limitations to note: context is China’s AIPZ program and 2015–2023 period—generalizability to other institutional environments may be limited; further work could unpack firm-level causal channels, quantify net welfare gains, and assess distributional impacts.