U.S. chip export controls rewire, not sever, Sino‑technology links: Beijing’s defensive substitution and regulatory countermeasures are reshaping supply chains into a durable, politically segmented decoupling rather than dissolving interdependence.

Main Finding

Weaponized interdependence in advanced-technology sectors (exemplified by semiconductors) produces an iterated coercion–adaptation dynamic: when the United States activates chokepoints via export controls and allied coordination, China responds with defensive reconfiguration (indigenization, substitution, friend-/re-shoring, regulatory measures). These interactions do not cause wholesale decoupling but instead restructure interdependence into a durable equilibrium of partial, segmented decoupling—persistent technical and commercial links governed by increasingly politicized, asymmetric rules of access.

Key Points

• Core mechanism
  • “Chokepoints” = concentrated, jurisdictionally controlled nodes in a supply chain (e.g., advanced design tools, manufacturing equipment, frontier fabs).
  • “Chokepoint activation” = deliberate state restriction (export controls, licensing, allied alignment).
  • Iterated activation produces feedback: altered expectations → firm/state adaptation → new policy pressures → further restriction.
• Sectoral conditions that make weaponization effective
  • Cumulative innovation (path dependence matters).
  • Limited substitutability (few short-run alternatives for high-end inputs).
  • Dual-use character (civilian and military applications).
  • Highly concentrated production networks (identifiable chokepoints).
• Three durable structural transformations resulting from repeated chokepoint activation
• Supply‑chain reconfiguration: geographic diversification, friend‑/re‑shoring, avoiding exposed jurisdictions.
• Substitution and redundancy: state-supported domestic alternatives and capacity-building even at higher cost/inefficiency.
• Regulatory and standards segmentation: parallel standards, differentiated export regimes, higher transaction/compliance costs.
• Political/security dynamics
  • A security-dilemma-like cycle: control measures interpreted as containment → target’s defensive buildup interpreted as latent threat → incentives for more restriction.
  • Engagement historically produced the chokepoints unintentionally; vulnerability is endogenous to long-run integration.
• Scope/limits
  • Argument is sector-specific: effective where production is concentrated, substitution costly, and dual-use stakes are high; not applicable to highly fungible commodity markets.

Data & Methods

• Approach: conceptual and historical case-study analysis drawing on the weaponized interdependence literature.
• Temporal focus: early 2010s onward (post-Snowden securitization) with special attention to U.S. export controls after the first Trump administration and allied coordination efforts.
• Methods used:
  • Theoretical refinement of weaponized interdependence: shifting from single episodes of coercion to iterated, system-level dynamics.
  • Process tracing of semiconductor sector evolution: how engagement created concentrated nodes and how policy actions produced adaptive responses.
  • Synthesis of prior empirical work and policy events (export controls, licensing regimes, allied alignment, Chinese indigenization and regulatory responses).
  • Conceptual diagrams and typology (three structural indicators) to formalize causal sequence and observable manifestations.
• Evidence types: policy documents, prior empirical studies, historical trajectory of industry structure and export-control episodes. (Paper is interpretive and qualitative rather than statistical.)

Implications for AI Economics

• Chokepoints in the AI stack
  • Key AI inputs (advanced GPUs/accelerators, frontier semiconductors, lithography tools, specialized design software, cloud compute and data flows) are chokepoint-prone—hence vulnerable to export controls and allied coordination.
• Short- and long-run effects on AI supply, costs, and capabilities
  • Restricted access to top‑end compute/hardware raises marginal cost of training/deploying frontier models for targeted actors and slows capability growth.
  • Defensive indigenization (domestic chip fabs, local cloud) increases duplication and fixed costs globally, reducing allocative efficiency but increasing strategic resilience.
  • Over time, segmented markets may produce capability divergence: different blocs with distinct hardware ecosystems, tooling, and model architectures.
• Innovation incentives and industrial policy
  • Targeted states will accelerate subsidies and industrial policy for domestic AI hardware/software, changing private profitability calculus and global competition in AI investments.
  • Firms may internalize geopolitical risk: multihoming suppliers, friend-shoring, or vertical integration to reduce exposure to foreign jurisdictional control.
• Research, talent, and standards
  • Cross-border scientific collaboration and data sharing may be curtailed by politicized access rules; parallel standards and compliance regimes will increase transaction costs for multi-jurisdiction deployment.
  • Open-source flows could be limited or channeled into trusted networks; proprietary ecosystems may deepen.
• Strategic dynamics and welfare tradeoffs
  • Security-driven controls can achieve short-term denial goals but impose long-run structural costs: higher global costs of compute, slower diffusion of beneficial AI applications, and potential inefficiencies from duplicated capability stacks.
  • A security-dilemma dynamic risks spiraling into an arms-race for compute and closed ecosystems, with negative externalities for global public-good aspects of AI (benchmarks, shared safety practices).
• For economists and modelers
  • Standard innovation-diffusion models should incorporate political segmentation and chokepoint-induced path dependence.
  • Need formal models (and empirical measures) of substitutability and jurisdictional control in the AI hardware/software stack to assess welfare impacts of export controls.
  • Scenario analyses should quantify: compute price elasticities, investment responses to export controls/subsidies, degree of capability divergence under segmented equilibria, and distributional effects across countries and firms.
• Policy-relevant prescriptions implied by the analysis
  • Map and monitor chokepoints across the AI stack (hardware, fabs, tooling, cloud/data infrastructure).
  • Design export-control policy mindful of long-run structural feedbacks: balance immediate security gains with costs of induced duplication and fragmentation.
  • Coordinate with like-minded allies but anticipate fragmentation effects; where feasible, invest in cooperative mechanisms for trusted access and standards to limit segmentation.
  • For targeted countries, transparent industrial strategies that reduce vulnerability while minimizing global welfare losses (e.g., targeted partnerships, staged indigenization, investment in substitutable layers).

Overall, the paper implies that AI economics must treat geopolitical constraint and supply‑chain chokepoints as first‑order determinants of innovation trajectories, cost structures, and global distribution of AI capabilities. Models and policy analyses that ignore iterated political adaptation and segmented interdependence will understate the real economic and strategic costs of export controls and defensive industrial policy.