“The factory that makes the brains of the modern world.” A semiconductor foundry — also called a fab or contract manufacturer — is a facility that fabricates integrated circuits designed by other companies (fabless chip designers), using photolithographic processes to etch transistors onto silicon wafers at nanometer-scale precision.
Executive Summary
The foundry model, pioneered by TSMC (Taiwan Semiconductor Manufacturing Company) in 1987, decoupled chip design from chip manufacturing — enabling fabless design companies like NVIDIA, Qualcomm, and Apple to innovate in chip architecture without building billion-dollar fabs. The result is extreme geographic concentration: TSMC alone produces roughly 90% of the world’s most advanced logic chips (below 7nm), and its Taiwan fabs represent one of the most strategically consequential industrial facilities in human history. This concentration has made semiconductor foundry access a central variable in U.S.-China tech competition, the Taiwan geopolitical risk calculus, and allied industrial policy.
The Strategic Mechanism
The foundry ecosystem has several distinct tiers:
- Leading-edge fabs (below 5nm): Only TSMC (Taiwan) and Samsung (South Korea) currently produce at this level. Intel is attempting to reenter through its Intel Foundry Services (IFS) business. These nodes power the most advanced AI chips, mobile processors, and high-performance computing
- Mature-node fabs (28nm and above): Produced globally by TSMC, Samsung, SMIC, GlobalFoundries, UMC, and others. Critical for automotive, industrial, and defense chips. China’s SMIC is competitive at this level despite export controls
- Equipment dependence: Advanced fabs depend on ASML’s extreme ultraviolet (EUV) lithography machines — manufactured exclusively in the Netherlands — to print sub-7nm features. ASML is itself a chokepoint
- Packaging and advanced packaging: Chiplet architectures and 3D stacking (CoWoS, SoIC) have made advanced packaging a new frontier; TSMC dominates here too
- CHIPS Act fabs: The U.S. CHIPS and Science Act (2022) allocated $52B+ to incentivize leading-edge foundry construction in the United States, with TSMC, Samsung, and Intel all receiving awards for Arizona, Texas, and Ohio facilities
Market & Policy Impact
- A Chinese military action against Taiwan that disrupted TSMC’s operations would remove roughly 90% of advanced logic chip capacity from global supply within months — a catastrophic scenario for every technology-dependent industry globally
- TSMC’s Arizona fab (Fab 21) began production of 4nm chips in 2024, representing the first leading-edge U.S. foundry capacity in decades, but cost, yield, and workforce challenges underlined the difficulty of geographic diversification
- Export controls preventing ASML from shipping EUV machines to China are the primary barrier to Chinese leading-edge foundry advancement — SMIC is working at 7nm with older DUV equipment but cannot progress further without EUV
- The foundry concentration problem has driven Japan (Rapidus), EU (Intel Ohio and TSMC Dresden), and India (Tata) to launch subsidized fab programs, representing hundreds of billions in global semiconductor industrial policy
- AI chip demand — driven by large language model training and inference infrastructure — has made leading-edge foundry capacity the scarcest and most strategically valuable industrial resource of the 2020s
Modern Case Study: TSMC Arizona and the Limits of Geographic Diversification, 2022–2025
TSMC’s decision to build in Arizona, announced in 2020 and accelerated by U.S. CHIPS Act funding, was presented as a flagship success for U.S. semiconductor industrial policy. But the project encountered significant difficulties: construction costs ran substantially above Taiwan equivalents, TSMC struggled to recruit sufficient U.S. engineers and skilled technicians, and cultural and operational differences between TSMC’s Taiwan workforce and U.S. construction contractors caused delays. Volume production of 4nm chips began in 2024 — later than originally projected — and TSMC executives publicly acknowledged that Arizona production costs were materially higher than Taiwan. The episode illustrated the structural challenge of replicating decades of semiconductor manufacturing ecosystem density in a new geography within a policy-driven timeline.