AI development strengthens China's manufacturing supply‑chains, primarily by boosting regional economic growth; the gains are larger in more urbanized, data-rich provinces and concentrate in the economically advanced east, with benefits rising once data‑infrastructure crosses a threshold.

Main Finding

AI development significantly and robustly increases the resilience of manufacturing industrial chains in China (2012–2023), both directly via technological empowerment and indirectly by raising regional economic development. The positive effect is moderated upward by urbanization, shows nonlinear threshold behavior with respect to data-element development, and exhibits spatial heterogeneity (strongest in the east, then west, weakest in the central region). Stronger industrial chains gain more from AI (a “Matthew effect”).

Key Points

• Direct effects: AI enhances situational awareness, enables proactive early warning, supports dynamic optimization and autonomous decision-making, and promotes flexible reconfiguration (digital twins, robotics, computer vision), improving buffering, recovery, adaptation and evolution capacities of manufacturing chains (Hypothesis H1 supported).
• Indirect effects: AI fosters regional economic development (TFP, new business models, structural upgrading), which in turn strengthens industrial-chain resilience — regional economic development mediates the AI → resilience link (Hypothesis H2 supported).
• Moderation: Higher urbanization rates amplify AI’s positive impact on chain resilience by concentrating talent, capital and infrastructure and improving local linkages and spillovers.
• Nonlinearity / thresholds: The AI → resilience effect is nonlinear with respect to the development level of data elements; the relationship changes beyond a data-element development threshold (authors report a change in marginal effect — described in the paper as a “marginal increase” beyond the threshold).
• Heterogeneity: Regional variation in effect size — east > west > central. Quantile/heterogeneity analysis indicates benefits grow with higher base-level resilience (stronger chains reap larger gains).
• Robustness & endogeneity: Findings held under multiple checks — subsample analysis, alternative variable measures, simultaneous-equation modelling, instrumental variables, and quantile regressions.

Data & Methods

• Data: Provincial panel (30 Chinese provinces), 2012–2023.
• Measurement:
  • AI index constructed as a multi-level indicator covering AI infrastructure, industrial inputs and outputs (authors’ composite measure).
  • Manufacturing industrial-chain resilience measured via a multi-dimensional indicator system (scale, efficiency, innovation, sustainability) aggregated using principal component analysis (PCA) / entropy-weighted methods.
  • Mediator: regional economic development (proxied in the paper by standard GDP-related measures and development indicators).
  • Moderator: urbanization rate.
  • Threshold variable: development level of data elements (a constructed measure capturing data resource/element development).
• Empirical strategy:
  • Baseline panel regressions (fixed effects framework implied) to estimate direct AI → resilience effects.
  • Mediation analysis to test economic development as transmission channel.
  • Moderation analysis to test urbanization rate interactions.
  • Panel threshold models to detect nonlinearities with respect to data-element development.
  • Heterogeneity explored via regional subsamples and quantile regressions.
  • Endogeneity addressed using simultaneous-equation specifications and instrumental variable methods.
  • Robustness checks: subsample splits, alternative variable constructions and measurement substitutions.
• Note: The manuscript is an accepted pre-publication version; detailed variable lists, exact instruments and estimation statistics should be consulted in the published version for replication.

Implications for AI Economics

• Policy design: Promote differentiated, region-sensitive AI policies. Invest in data-element development and data infrastructure to move regions beyond threshold levels where AI’s marginal impact on resilience increases. Use urbanization and agglomeration policies to magnify AI benefits.
• Equity and regional strategy: Because benefits are heterogenous and display a Matthew-type effect, policy should include targeted support for central/lagging regions (subsidies for data infrastructure, talent programs, coordinated industry clusters) to avoid widening resilience gaps.
• Industrial strategy: Firms and industrial parks should prioritize AI deployments that improve supply-chain visibility, dynamic scheduling and modular production; however, expected returns depend on local data resources and existing chain strength.
• Research agenda: Further work should unpack longer-term dynamic/evolutionary effects of AI on chain resilience, model cross-chain contagion and simulation dynamics, and integrate institutional and organizational coordination mechanisms (policy, finance, standards) into empirical analyses.
• Measurement & evaluation: The paper highlights the importance of composite indices for AI and resilience; future economic work should refine indicators for data elements, AI capability (use vs capability), and firm–chain level disentanglement to better estimate heterogeneous returns.