Private vendors, not just governments, hold the pace button on frontier compute: ASML's service regimes and TSMC's qualification and scheduling jointly exert a 'double-key' chokehold that can delay or throttle usable advanced-chip output even when access to equipment is legally permitted.

Main Finding

States can wield chokepoints to restrict adversaries’ access to frontier semiconductor technology, but they do not control the private, time-sensitive operational levers that determine whether and how quickly advanced fabs actually produce chips. The article introduces the concept of the corporate chokehold — a form of private authority, enacted via a “double-key” configuration (e.g., ASML + TSMC) — that governs frontier production continuity, scale, and timing in ways that run parallel to, but distinct from, state-leveraged export controls.

Key Points

• Two distinct governance logics operate in parallel:
  • State-leveraged chokepoints (export controls, sanctions) govern access to frontier technology and inputs.
  • Corporate chokeholds govern the operational continuity and tempo of frontier production (service, qualification, scheduling).
• Corporate chokehold: operational private authority manifested as a double-key arrangement. Example:
  • ASML controls EUV lithography hardware and its service regime (installation, maintenance, tool uptime).
  • TSMC controls design qualification, advanced packaging know-how, and capacity schedules (what gets produced and when).
  • Both companies’ control is necessary for sustained, timely leading-edge outputs.
• Three-part diagnostic for identifying corporate chokeholds:
• Infrastructural embedment — how firms become indispensably embedded in the technical, organizational, and service networks of production.
• Authority relocation — the shift of operational decision-making power from states or markets to firms (private authority over critical infrastructure).
• Temporal lock-in — dependencies that make production pace and continuity sensitive to servicing, qualification cycles, and scheduling, producing path dependence and timing control.
• Empirical findings (leading-edge semiconductors and advanced packaging):
  • US-led restrictions (targeting China) change access conditions but do not substitute for corporate control over the production pipeline.
  • Corporate constraints (service regimes, qualification, capacity scheduling) limit how fast and reliably US/allied fabs can scale output even when access is permitted.
• Alternative explanations assessed:
  • Onshoring, adding OSAT vendors, or increasing market competition can expand capacity but do not, by themselves, reallocate control over servicing, qualification regimes, or production timing; thus they do not eliminate corporate chokeholds.

Data & Methods

• Empirical focus: leading-edge semiconductors and advanced packaging practices and infrastructure.
• Methodology: theory-testing through process tracing using documentary evidence (industry statements, company disclosures, regulatory documents, policy reports).
• Case evidentiary anchor: ASML (lithography tools and services) and TSMC (qualification, packaging, capacity scheduling) as exemplars of the double-key corporate chokehold in practice.
• Comparative diagnostic: apply the three-part framework (infrastructural embedment, authority relocation, temporal lock-in) to test whether corporate actors—not only states—control production tempo and continuity.

Implications for AI Economics

• Supply-side constraints for AI compute are not solely geopolitical: private firms’ operational authority can be the binding constraint on how quickly frontier compute capacity scales.
• Modeling and forecasting AI compute availability should incorporate corporate chokeholds:
  • Include institutional/operational dependencies (service regimes, qualification timelines, scheduling rules) in capacity-growth models.
  • Treat “double-key” control points as choke variables that can delay or throttle production regardless of capital investment or legal access.
• Policy implications:
  • Export controls can prevent adversaries’ access but may also indirectly slow allied compute expansion by fragmenting supply chains or incentivizing firms to restrict servicing/exports.
  • Mitigation strategies must address private operational authority: build redundant service ecosystems, invest in technician training and local servicing capacity, diversify critical tool suppliers, or negotiate binding service/access agreements with key vendors.
  • Public–private coordination is necessary: procurement and industrial policy should target not only fab construction but also the servicing, qualification, and scheduling regimes that determine usable output over time.
• Empirical research directions:
  • Quantify infrastructural embedment (e.g., share of critical tool servicing provided by single firms; geographic concentration of specialist technicians).
  • Measure temporal lock-in (mean lead times for qualification, mean downtime from service events, scheduling flexibility).
  • Map double-key relationships across the compute stack (tools, foundries, OSATs, software/qualification firms) to identify systemic choke points for AI compute.
• Strategic consequences:
  • States hoping to secure rapid, controllable access to frontier compute for AI programs must consider the private governance structures that can overdetermine timelines and availability.
  • Investments aimed at increasing capacity without addressing corporate chokeholds risk raising nominal capacity numbers while leaving real, timely, and qualified output constrained.