There is no single AI policy that guarantees regional industrial competitiveness in China; provinces achieve sustainable gains through one of three distinct policy mixes that consistently include R&D support, talent development and application demonstration.

Main Finding

No single AI policy instrument is sufficient to sustainably enhance regional science and technology industrial competitiveness. Instead, sustained competitiveness emerges from multiple, distinct combinations (configurational pathways) of policy instruments. Three broad driving pathways were identified — (supply and demand)-environmental resonance, demand-driven (supply–environmental) assurance, and supply–demand complementarity — realized through five specific configurations. Technology R&D support, talent cultivation & collaboration, and application demonstration & promotion are the most recurrent core conditions across successful configurations.

Key Points

• Research question: Which combinations of AI policy instruments lead to sustained improvement in regional science & technology industrial competitiveness?
• Data scope: AI policies issued by provincial-level governments in China (policy instrument framework used to categorize instruments).
• Method: fuzzy-set qualitative comparative analysis (fsQCA) to identify combinations (configurations) of policy instruments associated with high sustained competitiveness.
• Main empirical result: No single instrument is necessary or sufficient; multiple equivalent configurations produce the outcome.
• Identified driving pathways:
• (Supply & demand) — environmental resonance: alignment of supply- and demand-side measures with enabling environmental policies.
• Demand-driven (supply–environmental) assurance: demand stimulation as the lead with supply-side and environmental supports assuring outcomes.
• Supply–demand complementarity: mutually reinforcing supply- and demand-side policies paired to drive competitiveness.
• Five specific configurations map onto these pathways, with variation reflecting regional resource endowments and development conditions.
• Recurrent core policy elements: technology R&D support, talent development & collaboration, and application demonstration & promotion.

Data & Methods

• Unit of analysis: provincial-level AI policy instruments in China (policies coded according to a policy instrument framework).
• Policy categorization: typically grouped into supply-side (e.g., R&D funding, infrastructure, talent programs), demand-side (e.g., public procurement, subsidies for application), and environmental/regulatory instruments (e.g., standards, intellectual property, market environment).
• Analytical approach: fuzzy-set QCA (fsQCA)
  • Rationale: fsQCA is suited to identifying how combinations of causal conditions jointly produce an outcome, capturing conjunctural causation, equifinality, and causal asymmetry.
  • Output: configurations (combinations of presence/absence of instruments) associated with high sustained regional science & technology industrial competitiveness.
• Limitations (inferred from method): fsQCA identifies configurations but not precise effect sizes; results are contingent on calibration and case selection and reflect configurational (not purely additive) causation.

Implications for AI Economics

• Policy design must consider complementarities and interactions among instruments rather than relying on one leverage point. Economic models of AI policy should incorporate complementarities and nonlinearity across instruments.
• Heterogeneity matters: optimal policy mixes depend on regional resource endowments and development stage. One-size-fits-all national prescriptions are likely suboptimal.
• Prioritization: investments in technology R&D, human capital (talent cultivation and collaboration), and application demonstration/early adoption are high-leverage across multiple successful configurations — these should be central to regional AI strategies.
• Demand stimulation (e.g., procurement, subsidies) can play a leading role where market pull is weak, but it is most effective when paired with supply-side supports and an enabling environment.
• For empirical AI economics research: configuration-focused methods (like fsQCA) are useful complements to regression-based approaches when studying policy mixes and institutional interactions. Future work should combine configurational analysis with causal identification and quantitative estimation of magnitudes (e.g., difference-in-differences, instrumental variables) to estimate effect sizes.
• Practical recommendations for policymakers:
  • Design coordinated AI policy packages that combine R&D, talent, application demonstration, and enabling environment measures.
  • Tailor mixes to regional characteristics (industrial base, talent pool, market demand).
  • Monitor interactions and adapt policy mixes over time; evaluate not only individual instruments but their joint effects.
  • Support pilot/demonstration projects to create demand signals and accelerate diffusion.
• Research agenda suggestions: longitudinal analysis of policy mix dynamics; firm- and industry-level outcomes of different configurations; cost–benefit analysis of alternative mixes; cross-country comparisons of configurational pathways.