China’s AI governance is not monolithic: local governments and firms co-regulate alongside central authorities, producing layered, experimental regulation that fosters innovation in places but creates fragmentation and compliance costs across jurisdictions.

Main Finding

China’s AI governance is not a single, top-down, command-and-control system. Based on analysis of an original corpus of primary sources, the Article shows a polycentric, layered governance structure in which local governments and market actors play substantial, distinct roles alongside central authorities. This produces a multi-level, iterative governance process rather than a monolithic state-centered model.

Key Points

• Local governments matter: subnational jurisdictions in China adopt regulatory approaches that differ from the central government, exhibiting federalism-like features (experimentation, variation in enforcement and priorities).
• Market actors as co-regulators: firms and industry organizations actively shape rules and bargaining processes; they are strategic participants in regulation rather than passive targets.
• Iterative governance: regulation is produced through ongoing interaction among central authorities, local governments, and market players — rules evolve through negotiation and adaptation.
• Diverse objectives and tools: Chinese AI governance pursues a range of aims (economic development, social stability, technological leadership, data control, etc.) and uses varied instruments beyond command-and-control (local policy incentives, industry guidance, administrative bargaining).
• Polycentric and layered structure: these dynamics cohere into a governance architecture with multiple centers of authority and overlapping regulatory layers, producing both cooperation and tension across levels.
• Counters simplistic narratives: the findings complicate common geopolitical portrayals of China as a uniform authoritarian regulator and highlight potential points of common ground for international engagement.

Data & Methods

• Data: an original corpus of primary sources (central and local government documents, regulatory texts and notices, official statements, and contemporaneous industry materials).
• Methods: qualitative document analysis and cross-jurisdictional comparison within China to trace regulatory variation and the interactive processes linking state and market actors; emphasis on processual evidence of negotiation and policy iteration rather than solely formal statutes.

Implications for AI Economics

• Regulatory heterogeneity affects market structure and firm strategy: local variation creates fragmented regulatory landscapes that can favor local champions, enable regulatory arbitrage, and shape firm location, investment, and product choices.
• Innovation incentives and experimentation: decentralized policy experimentation can accelerate innovation and allow policy learning, but uneven rules raise compliance costs and scale-up risks for firms operating across jurisdictions.
• Bargaining power and market governance: active firm-state bargaining alters investment returns and may lead to industry-specific governance bargains (potentially regulatory capture or cooperative standard-setting), influencing market entry and competition.
• Data access and transaction costs: layered governance (central vs. local rules) can affect data governance regimes, data localization requirements, and cross-border data flows—important determinants of AI model training costs and competitive advantage.
• Systemic risk and externalities: polycentric oversight may reduce single-point failure risks but complicate management of cross-jurisdictional externalities (e.g., biased systems, privacy harms), raising the need for coordination mechanisms.
• Policy implications for international engagement: recognizing China’s internal diversity opens avenues for targeted cooperation (technical standards, mutual recognition, pilot exchanges) rather than treating China as a unitary actor; harmonization efforts might focus on common operational areas (safety standards, measurement/principles) where local variation is already present.
• Research implications: economic models of AI adoption and regulatory effects should account for multi-level governance, bargaining dynamics between firms and multiple regulators, and the effects of subnational policy experimentation on innovation and market outcomes.