China is betting on diffusion, not just frontier models: a state-led push channels finance and coordinated governance to put AI into factories and robots at scale, shifting the strategic contest from who reaches frontier capabilities first to who most rapidly integrates AI into productive activity.

Main Finding

China pursues a “diffusion‑forward” AI strategy: a state‑directed push to embed AI as a general‑purpose technology throughout the physical economy (especially manufacturing, industrial robotics, and embodied AI). This contrasts with U.S. emphasis on frontier models and LLM competition. China’s institutional setup—decentralized but hierarchical governance, an investor‑state model, and campaign‑style industrial policy—facilitates rapid commercial diffusion, shifting the core competition from who attains AGI first to who more quickly integrates AI into productive activity.

Key Points

• Diffusion‑forward strategy: policy and industrial practice prioritize spreading AI capabilities into factories, robotics, and embodied systems rather than focusing solely on frontier model capabilities.
• Sectoral focus: strong emphasis on manufacturing, industrial robotics, and embodied AI applications that affect physical production and services.
• Institutional enablers:
  • Decentralized yet hierarchical governance lets local authorities implement central directives while tailoring support.
  • Investor‑state model channels substantial capital into firms and industrial projects aligned with national priorities.
  • Campaign‑style industrial policy mobilizes resources and coordinates actors across public and private sectors for targeted diffusion.
• Observable outcomes: original registration data for generative AI services and a patent‑based case study (UBTECH) show measurable commercial activity consistent with the diffusion objective.
• Reframing competition: the strategic question becomes which political economy can most rapidly, widely, and effectively integrate AI into productive use, not only which reaches frontier AI milestones first.

Data & Methods

• Policy analysis: systematic review of official Chinese policy documents to characterize strategic objectives and instruments.
• Administrative/registration data: original dataset of generative AI service registrations in China to track commercial deployment and sectoral uptake.
• Case study / patents: patent analysis for UBTECH (humanoid robotics firm) to trace innovation trajectories, firm strategy, and commercialization in embodied AI and robotics.
• Analytical approach: combine qualitative institutional analysis with quantitative indicators (service registrations, patent activity) to link policy instruments to measurable diffusion outcomes.

Implications for AI Economics

• Measurement: economists should broaden metrics beyond model parameters and compute to include adoption rates, robotics deployments, embodied AI patents, and firm‑level commercialization indicators.
• Comparative analysis: cross‑country comparisons must account for institutional differences (state capacity, governance style, financing models) that shape diffusion speed and direction.
• Forecasting competitiveness: scenarios of AI geopolitical competition should weight diffusion into the physical economy (productivity, industrial upgrading) as a central axis, not just frontier model superiority.
• Industrial policy and firms: the effectiveness of campaign‑style coordination and investor‑state financing suggests different pathways for scaling AI commercialization; private firms and investors should evaluate exposure and opportunities in embodied AI sectors.
• Labor and welfare: rapid diffusion into manufacturing and services will have distinct labor market impacts (automation of embodied tasks), requiring targeted retraining and social policies.
• Trade, standards, and geopolitics: faster diffusion in production technologies can alter comparative advantage, reshape supply chains, and drive divergence in technical standards and regulatory regimes.
• Research agenda: prioritize empirical work on embodied AI adoption, sectoral productivity effects, the role of local governance in technology diffusion, and the long‑run consequences of state‑led scaling strategies.
• Policy caution: while diffusion can accelerate productivity, it raises questions about competition, concentration, and strategic dependencies that require complementary governance (competition policy, export controls, data and safety standards).

Caveats: the paper’s findings are based on Chinese policy documents, registration data, and a firm‑level patent case study; generalizing to other contexts requires careful institutional and sectoral adjustment.