US semiconductor export controls ripple through China’s economy, concentrating supply risk in communication equipment, computer manufacturers and automotive electronics while also destabilising chip-dependent digital services; boosting domestic innovation and substitution capacity is key to reducing systemic vulnerability.

Main Finding

U.S. chip export controls (EAR revisions, Entity List expansion, Foreign Direct Product Rule, allied coordination and the CHIPS Act) create a multilayered blockade across equipment, EDA/software, IP licensing and materials that produces structural dependence at critical nodes of China’s integrated circuit (IC) value chain. Using a supply‑driven input–output (Ghosh) framework on China’s 2020 national input–output table, the paper shows that constraints on chip supply propagate widely: communication equipment, computer manufacturing, and automotive electronics are most exposed, while modern service sectors (owing to reliance on digital infrastructure) show strong sensitivity. Strengthening domestic innovation and substitution capability is essential to rebuild industrial resilience.

Key Points

• Policy evolution: U.S. measures evolved from firm‑level sanctions (2018–2020) to institutional consolidation (2021–2022) and coordinated multilateral controls (2023 onward). Key instruments include EAR revisions, Entity List listings, FDPR, and export coordination with the Netherlands/Japan (e.g., ASML lithography restrictions).
• Multilayer restrictions target: advanced lithography and fabrication tools, EDA software and IP, high‑end materials (photoresists, wafers), and high‑performance AI/GPU chips.
• Structural vulnerabilities: China retains progress in many IC segments but remains externally dependent at critical upstream nodes — lithography, EDA/IP, and specialty materials — which form effective choke points under export controls.
• Transmission mechanism: Chips act as an intermediate input; supply constraints at upstream nodes cascade downstream through production networks, affecting both manufacturing sectors and digitally intensive services.
• Most exposed downstream sectors: communication equipment, computer manufacturing, automotive electronics; modern services (cloud, data centers, AI services) are also sensitive because of dependence on advanced chips (e.g., GPUs for training/inference).
• Policy implication emphasized by the paper: accelerate domestic R&D, build substitution capacity for equipment/software/materials, and design targeted industrial policies to reduce systemic risk.

Data & Methods

• Data: China’s 2020 national input–output table (single‑country, national accounting snapshot).
• Methodological framework:
  • Supply‑side input–output modeling using the Ghosh (1958) formulation (full allocation coefficient matrix), measuring how an upstream supply unit is allocated and absorbed across downstream sectors.
  • Construction of the full allocation coefficient matrix to quantify supply‑driven transmission (as distinct from Leontief demand‑side analysis).
  • Scenario simulations: impose chip export restriction shocks and trace cascading effects through the allocation matrix to identify sectoral exposure and transmission intensity.
• Contributions of the method:
  • Applies full allocation coefficients to model non‑tariff technological blockades.
  • Quantifies direction and intensity of intersectoral transmission under supply constraint scenarios.
• Limitations (noted or implied):
  • Uses a static 2020 IO table — captures structural linkages at one point in time and abstracts from dynamic firm‑level adaptation.
  • National IO framework does not model international relocation/retaliation responses or time‑path of technological substitution.
  • Aggregation may mask firm heterogeneity and granular supply‑chain variation.

Implications for AI Economics

• Chips as foundational capital for AI: restrictions on advanced GPUs and AI chips directly constrain AI training/deployment capacity in affected economies, raising costs and delaying model development and scaling.
• Short‑run effects on AI adoption and services:
  • Cloud providers, AI startups, and data‑intensive service firms may face compute shortages or higher prices, slowing adoption and degrading competitiveness.
  • Downstream sectors that integrate AI (communication equipment, autonomous vehicle electronics, smart manufacturing) experience productivity and innovation slowdowns.
• Long‑run structural consequences:
  • Incentives to onshore AI compute and accelerate domestic chip R&D — could lead to bifurcation/fragmentation of global AI hardware ecosystems and different technology standards.
  • Reallocation of investment toward substitute architectures (quantum, edge compute, domain‑specific accelerators) or software optimizations (model compression, more efficient algorithms) to mitigate hardware limits.
  • Shifts in comparative advantage: countries with secure access to advanced chips gain sustained AI productivity advantages; export controls can therefore reshape global specialization in AI‑intensive industries.
• Modeling and policy research implications:
  • AI economic models should incorporate supply‑side constraints on compute hardware (not just demand or data availability). Full allocation (Ghosh‑style) approaches are valuable for tracing upstream supply shocks into AI service layers.
  • Welfare and competition analyses must account for compute scarcity, price effects, endogenous R&D responses, and potential inefficiencies from forced substitution.
  • Policy design: targeted subsidies (R&D, fabs), investment in domestic EDA and materials, international cooperation strategies, and contingency planning for AI compute provisioning are crucial to manage systemic risk.
• Empirical priorities: quantify compute bottlenecks by firm/region, estimate price‑elasticities of AI compute demand, and simulate dynamic adoption under alternate hardware availability scenarios to inform industrial and innovation policy.

If you want, I can: (a) extract the paper’s simulated sector‑level exposure rankings into a concise table; (b) sketch how to extend the Ghosh IO approach to a dynamic multi‑country model to capture substitution and relaying of supply chains; or (c) outline policy options with likely economic costs/benefits for rebuilding chip‑related resilience. Which would be most useful?