Inputs to 2026-08
Scenarios
Five stress cases over the AI-compute supply chain.
Five stress cases over the AI-compute supply chain.
A disruption to TSMC can restrict the supply of chips for new AI systems. Four cases (blockade, physical damage, loss of power and export restrictions) apply different assumed severities and propagation multipliers to TSMC. The cascade follows the recorded supplier–customer network across source classes. The connection diagram and inventory-buffer calculation use the publicly sourced subset. The underlying collection contains 18,980 supply-chain records across all sources.
Taiwan accounts for ~92% of leading-edge pure-play foundry manufacturing (CSET, 2021), a country-level measure of the wafer-fabrication stage. Each cascade starts at TSMC and follows recorded supplier–customer relationships across source classes. The diagram shows the publicly sourced subset. Recovery figures are separately authored assumptions; substitution rows cite sources and carry no replacement estimates.
The diagram shows publicly sourced supplier–customer relationships; cascade totals also use relationships from other source classes.
How concentrated is the fab layer? Country-level share for leading-edge foundry, measured HHIs for total foundry revenue and EUV lithography, plus key customer dependencies.
What manufacturer disclosures and one teardown establish about alternatives to Taiwan fabrication, packaging and EUV tool supply, and the questions each leaves open.
Programmes assessed as highly dependent on NVIDIA, ordered by reported accelerator count. The curves apply the selected disruption scenario to their chip suppliers.
How the simulated disruption develops as inventories run down and alternative supplies become available. The shaded range varies the assumed inventory buffers.
What product specifications, teardowns and shipment estimates establish about Chinese suppliers’ ability to replace disrupted production.
Export-control records and the decisions that would affect access to replacement supplies during a disruption.
Compare the assumed severity, affected fabs and timing of each disruption, with historical events for context.
The TSMC starting point, assumed severity and propagation multiplier, and the different source coverage of the cascade and inventory-buffer calculations.
Open the engine’s TSMC disruption preset to compare it with another scenario or organisation and adjust the correlation weighting. The four cases here each use their own severity and propagation settings; this link opens the preset’s defaults.