China’s AI pilot-zone program raised city-level green productivity by accelerating industrial upgrading and green innovation; benefits were concentrated in eastern, non-resource-rich cities with stronger digital infrastructure.

Main Finding

Zhou, Gu & Yang (2026) find that China’s National New Generation Artificial Intelligence Innovation and Development Pilot Zones (AIIDPZ) causally increase urban green total factor productivity (GTFP). Using a staggered rollout of pilot-zone designations as a quasi‑natural experiment for 282 Chinese cities (2011–2023) and a multi‑period difference‑in‑differences estimator, the authors show a significant positive effect on city‑level GTFP. The primary mechanisms are industrial structure upgrading and green technological innovation, and effects are stronger in eastern cities, non‑resource‑based cities, and cities with stronger digital infrastructure. (Front. Environ. Sci. 14:1808785. doi: 10.3389/fenvs.2026.1808785)

Key Points

• Policy evaluated: China’s AI Innovation and Development Pilot Zones (AIIDPZ), rolled out in multiple batches 2019–2023, with AI‑specific support (computing/data infrastructure, talent programs, innovation platforms, application scenarios, regulatory sandboxes, and fiscal support).
• Outcome: Urban green total factor productivity (GTFP) — a productivity metric that incorporates conventional inputs, desirable outputs and undesirable outputs (pollution) — used to assess whether growth is both more efficient and cleaner.
• Identification: Staggered difference‑in‑differences (multi‑period DiD) exploiting timing variation in pilot‑zone designation across cities; control cities did not receive the coordinated AI policy package during the sample.
• Main result: AIIDPZ designation leads to a statistically significant improvement in city‑level GTFP.
• Mechanisms:
  • Industrial structure upgrading: AIIDPZ promotes expansion of AI‑related and higher value‑added activities, accelerates intelligent transformation of traditional sectors, and facilitates factor reallocation toward cleaner, more productive firms/sectors.
  • Green technological innovation: AIIDPZ stimulates green R&D and innovation that shift the production frontier outward in a lower‑emission direction.
• Heterogeneity: Larger positive effects in (i) eastern (coastal) cities, (ii) non‑resource‑based cities, and (iii) cities with better digital infrastructure — indicating strong complementarities with local absorptive capacity.
• Caveats noted by authors: Short‑run increases in electricity demand and emissions from data centers/computing facilities and transitional adjustment costs; AI’s net environmental impact is conditional on local energy mix, infrastructure, and human capital.

Data & Methods

• Data: Panel of 282 Chinese prefecture‑level cities, 2011–2023.
• Treatment: AIIDPZ designation (staggered across cities and years; first batches in 2019, further expansions through 2023).
• Empirical strategy: Staggered/multi‑period difference‑in‑differences to estimate the causal impact of AIIDPZ on urban GTFP; mechanism tests for industrial upgrading and green innovation; heterogeneity analyses by region, resource dependence, and digital infrastructure.
• Outcome construction: GTFP measured to include both desirable outputs and undesirable outputs (pollution emissions), providing a combined productivity‑and‑environment metric (paper emphasizes this as more comprehensive than single indicators like green patents or energy intensity).
• Robustness: Authors report baseline results and robustness checks (placebo and other standard checks referenced in paper structure).

Implications for AI Economics

• Place‑based AI policy can produce measurable environmental returns: The study provides causal evidence that AI‑specific, place‑based policy (beyond generic digitalization) can raise green productivity at the city level, underlining AI’s potential as a green development lever when supported by coordinated public intervention.
• Complementarities matter: Returns to AI policy are conditional on local absorptive capacity (digital infrastructure, industrial structure, human capital). For economists and policymakers, this highlights complementarities between AI investments and pre‑existing regional capabilities.
• Mechanisms link AI to structural change and innovation: AI influences green outcomes not only via operational efficiency (process optimization) but also by shifting industrial composition and accelerating green technological innovation—important for modeling long‑run structural effects of general‑purpose technologies.
• Policy design lessons:
  • Integrate AI policy with energy and infrastructure planning (to manage data‑center energy demands and avoid offsetting efficiency gains).
  • Support complementary investments (digital infrastructure, talent, green R&D) to maximize green dividends from AI.
  • Targeted, phased pilot approaches allow causal evaluation and regional tailoring.
• Research implications:
  • Need for micro‑level (firm and plant) studies to unpack how AI adoption translates into emissions and productivity changes across sectors.
  • Investigate the net lifecycle energy and emissions footprint of AI infrastructure versus operational savings from diffusion.
  • Explore distributional effects and longer‑run dynamics (e.g., labor market impacts, exit of polluting firms, regional divergence).
• Methodological note for AI economists: The staggered DiD design here is a useful template for evaluating place‑based AI policies; combining productivity measures that internalize undesirable outputs (GTFP) is critical when analyzing environmental consequences.

Limitations and future directions (authors’ emphasis): short‑run carbon costs of infrastructure, conditionality on local conditions, and the need to assess long‑term sustainability tradeoffs — suggesting policymakers should pair AI promotion with decarbonization of power and strengthened digital foundations to realize robust green gains.