Broadcom's CEO has projected $100 billion or more in annual custom AI chip revenue by 2027.1 That entire production pipeline flows through one geography: Taiwan.
TSMC manufactures Broadcom's custom AI accelerators at advanced nodes no other foundry currently matches at scale. Intel Foundry and Samsung trail on yield and capability for cutting-edge AI chip designs. Switching suppliers takes years, not quarters.
Three distinct threats now pressure that supply chain. Military action in the Taiwan Strait would halt TSMC production immediately. Tightening US export controls on advanced semiconductor equipment could complicate TSMC's own tool supply. Chinese economic pressure or cyberattacks could create indirect production delays without a single shot fired.
A current risk assessment rates this scenario catastrophic in severity with medium likelihood.1 A prolonged disruption would delay customer delivery timelines across Broadcom's entire AI chip portfolio with no near-term alternative.
Broadcom's hyperscale customers — cloud providers building custom silicon for AI inference and training — have little recourse. These chips are designed specifically around TSMC's process nodes. Redesigning for another foundry, if one existed at comparable capability, would take 18 to 24 months minimum.
The $100 billion revenue target sharpens the concentration risk. A multi-quarter TSMC disruption could erase a large portion of projected AI chip revenue with no short-term mitigation path available to Broadcom or its customers.
US CHIPS Act investments are building domestic advanced node capacity, but TSMC's Arizona fabs run at lower volumes than Taiwan operations. Closing that gap takes years. The Taiwan dependency is not a 2026 problem with a 2027 solution.
For investors and enterprise customers pricing Broadcom's AI chip roadmap, the Taiwan variable remains the single largest unhedgeable risk in an otherwise strong demand story.