China’s AI patent network has become multi-tiered and Beijing-dominated, and a province’s position matters: both centrality and brokerage boost collaborative innovation up to a point but then diminish. Government R&D subsidies and stronger IP protection blunt the benefits of centrality while amplifying gains from structural holes during rapid expansion.

Main Finding

China’s provincial AI innovation network (2010–2024, built from patent-application ties) has evolved into a multi-tiered, polycentric system centered on Beijing. Both network centrality and structural-hole positions affect regional collaborative innovation performance nonlinearly: each exhibits an inverted U-shaped relationship with performance across development stages. The external institutional environment—notably government R&D subsidies and intellectual property (IP) protection—moderates these effects: it generally weakens the benefit of centrality while strengthening the benefit of structural-hole (brokerage) positions during rapid expansion. Regional heterogeneity exists (east/central/west follow the main pattern; northeast shows only a centrality effect). The study integrates network-location theory and institutional perspectives to explain how network governance and policy shape sustainable innovation ecosystems.

Key Points

• Network structure
  • China’s AI innovation system is polycentric and multi-tiered, with Beijing as the primary hub.
• Nonlinear effects of network position
  • Centrality → inverted U-shaped effect on regional collaborative innovation performance (initial positive effect that declines and reverses at high centrality).
  • Structural holes (brokerage) → inverted U-shaped effect as well (benefit up to a point, then diminishing/negative returns).
• Institutional moderation
  • Government R&D subsidies and stronger IP protection change the slope/magnitude of the relationships:
    • Reduce the positive returns to high centrality (attenuation).
    • Increase the returns to structural-hole positions, especially during rapid expansion periods.
• Regional heterogeneity
  • Eastern, central, and western provinces display the inverted U-patterns.
  • Northeast provinces show a significant association only for centrality (no clear structural-hole effect).
• Theoretical framing
  • Combines network-location theory (position matters) with institutional economics (formal environment alters returns to network positions).
• Policy relevance
  • Evidence-based guidance for targeting subsidies, IP policy, and network governance to boost collaborative innovation and reduce regional disparities.

Data & Methods

• Data
  • Patent application data (2010–2024) mapped to provincial-level actors; used to construct inter-provincial collaborative ties in AI-related patents.
• Network construction & measures
  • Provincial-level innovation network formed from co-application or co-inventor patent links.
  • Centrality measures (likely degree and/or betweenness centrality) to capture hub status.
  • Structural holes / brokerage measured via Burt-style metrics (constraint, effective size, or similar).
• Outcome variable
  • Regional collaborative innovation performance proxied by patent-based metrics (volume, possibly quality-adjusted metrics), measured across stages (e.g., early, rapid expansion, mature).
• Empirical strategy
  • Social network analysis to characterize topology and evolution.
  • Econometric models (panel regressions) testing the (nonlinear) effects of centrality and structural holes on innovation performance, including inverted U specifications (e.g., linear and squared terms).
  • Interaction terms and subgroup analyses to test moderation by institutional variables (government R&D subsidies, IP protection) and regional heterogeneity (east/central/west/northeast).
• Robustness checks
  • Temporal evolution analysis across phases (growth stages) and regional subsamples; unspecified but implied robustness procedures for network- and institution-interaction effects.

Implications for AI Economics

• Nonlinear returns to network position
  • Being a hub or broker is beneficial up to a point; excessive centralization or excessive brokerage can reduce collaborative innovation performance. Models of spatial/industry dynamics should incorporate diminishing returns to centrality and brokerage.
• Role of institutions in shaping network payoffs
  • Policy instruments (R&D subsidies, IP regime) materially alter the payoffs of network positions. Econometric and theoretical models should include institutional interaction terms to predict innovation outcomes.
• Regional inequality and policy targeting
  • Central hubs (e.g., Beijing) will remain pivotal, but overconcentration risks crowding-out benefits for collaborators. Policies should promote productive brokerage ties and support peripheral regions to avoid widening disparities.
• Design of innovation policy
  • Subsidy and IP policy should be calibrated by region and network structure:
    • Use subsidies and IP strengthening to encourage brokerage activity that connects otherwise disconnected regions (enhancing spread of knowledge).
    • Avoid policies that simply amplify hub dominance without enabling spillovers.
• Firm and cluster strategy
  • Firms and regional planners should seek optimal network positions rather than maximal centrality/brokerage; strategic placement depends on stage of regional development and the institutional environment.
• Modeling and measurement recommendations
  • Empirical AI-economics work should use network metrics (centrality, structural holes) and allow for nonlinearity and institution × network interactions when estimating innovation production functions.
• Sustainability of innovation ecosystems
  • Network governance and tailored institutional interventions can foster sustainable, resilient AI innovation ecosystems—balancing hub strength with broker-mediated connectivity to diffuse innovation and reduce fragility from overconcentration.