Concept

AI chokepoint dilution

AI chokepoint dilution is the proposition that controls over scarce computing inputs can gradually weaken the leverage of the original chokepoint by encouraging efficiency, substitution, diversion or new domestic supply. It is an analytical hypothesis, not a demonstrated consequence of every AI-chip and compute export control.

Mechanisms

Dilution can follow four distinct paths. Software or model efficiency can reduce the compute needed for a task. Designers can substitute lower-performance hardware or redesign systems around available components. Intermediaries can divert controlled items. Producers can develop indigenous equipment, chips, memory or packaging. Each path has different evidence and policy implications.

The Chokepoint effect remains strong if substitutes impose large cost, delay or performance penalties. It weakens when a target can obtain enough usable capability through alternatives within the relevant time horizon. The relevant measure is delivered system capability, not a product announcement or benchmark in isolation.

DeepSeek's technical report provides primary-party evidence about its model architecture and training claims. It does not independently establish total cost, hardware provenance, comparative performance or the failure of United States controls. Those conclusions require reproducible evaluation and supply-chain evidence.

Current control setting

As at 30 July 2026, current EAR Parts 740, 742 and 744, the May 2026 guidance and later amendments must be read together. The January 2025 diffusion framework was subject to a May 2025 non-enforcement announcement, but pre-existing headquarters and end-use controls continued.

Dilution is one possible Self-undermining arsenal dynamic within the wider Semiconductor chokepoint. A control can impose meaningful delay while also accelerating adaptation. Assessment should state the controlled item, target capability, countermeasure, cost, delay and counterfactual before claiming erosion.

Sources

  1. US Export Administration Regulations, Part 740 (accessed 30 July 2026).
  2. US Export Administration Regulations, Part 742 (accessed 30 July 2026).
  3. US Export Administration Regulations, Part 744 (accessed 30 July 2026).
  4. Bureau of Industry and Security, advanced-computing guidance, May 2026.
  5. Bureau of Industry and Security, AI Diffusion Rule non-enforcement announcement, May 2025.
  6. DeepSeek-AI, *DeepSeek-V3 Technical Report*.

Recommended citation

Cite this entry

Tennant, James J., ed. 'AI chokepoint dilution.' The Encyclopedia of Economic Statecraft, version 2.0, last reviewed 30 July 2026. https://jamesjtennant.com/entries/ai-chokepoint-dilution/.

Suggest an edit