Frontier Compute Lead +18.4 Mo US/Allied Frontier Lead
Cluster Cost Inflation +32.6% Capex per 100k GPU cluster
Stack Decoupling Index 68.2 / 100 Bilateral infrastructure separation
Substitution Horizon 4.2 Yrs Full domestic lithography replacement
US / Allied Stack
China / Sovereign Stack
Strategic Chokepoint
Severed Dependency Flow
Click any node to inspect exposure
Simulation steady. Asymmetric tech containment operational.
ENGINE: ACTIVE

Node Diagnostics & Impact

Frontier APIs
Cross-Border Frontier APIs Critical Severance

Policy stance directly rejects bilateral API access (OpenAI/Anthropic/Google). Enterprises in restricted jurisdictions lose zero-latency reasoning access, forcing migration to local open-weights (DeepSeek/Qwen).

Direct Decoupling Impact 88%
US Domestic Supply Security 94%
China Substitution Feasibility 62%
AI Stack Layer Status
Key Geopolitical Dynamic

Export controls create a bifurcated software and hardware ecosystem. While the US retains raw FLOP supremacy, China accelerates algorithmic efficiency and sovereign cluster clustering.

Strategic Architecture: The Anatomy of Bilateral AI Decoupling

1. Hardware Chokepoints

Advanced semiconductor fabrication remains concentrated in Taiwan (TSMC), lithography in the Netherlands (ASML), and EDA tooling in the US (Synopsys, Cadence). Sub-3nm nodes, HBM packaging, and GAAFET architectures represent structural chokepoints resistant to rapid domestic replication.

2. The API & Model Divide

Bilateral rejection of commercial AI integrations prompts a bifurcated standards universe. American enterprise stacks isolate proprietary closed-frontier models behind federal sovereign enclaves, while Chinese ecosystems rely on heavily optimized open weights (DeepSeek, Qwen) trained on sovereign domestic clusters.

3. Counter-Vulnerabilities

Decoupling carries mutual asymmetry: Western fabs rely on Chinese refining of critical inputs (gallium, germanium, antimony, spherical graphite, rare earth magnets). Complete decoupling spurs multi-year supply inflation across global datacenters.

Enjoy this tool? Build your own with Super