TSMC 2024: The Geopolitics of Silicon

Situation

It is April 2024. TSMC, Taiwan Semiconductor Manufacturing Company, has just received a $6.6 billion direct grant and $5 billion in federal loans from the US CHIPS and Science Act for its three planned fabs in Phoenix, Arizona. The first (4nm) is scheduled to start production in 2025; the second (2nm) by 2028; the third (2nm+) by 2030.

TSMC's position in the global technology supply chain is singular:

• It manufactures chips designed by Apple, NVIDIA, AMD, Qualcomm, MediaTek, and essentially every major semiconductor design company
• It controls ~90% of the market for chips below 7nm, the advanced chips that power AI training, smartphones, and high-performance computing
• Its 2nm and 3nm processes are 2-3 years ahead of Intel's best manufacturing technology and 3-4 years ahead of Samsung Foundry
• The H100, H200, and B200 GPUs that power the AI boom (NVIDIA) are manufactured exclusively at TSMC
• TSMC's Hsinchu Science Park campus and surrounding ecosystem employ approximately 65, 000 people in a cluster that took 35 years to build

The US government's concern: Taiwan is 110 miles from mainland China. China views Taiwan as a breakaway province; the PLA has conducted exercises simulating a Taiwan Strait blockade. US defense planners assess an attack on Taiwan as the primary military scenario for US-China conflict in the 2025-2035 timeframe.

If Taiwan is blockaded, sanctioned, or subject to military conflict, access to leading-edge semiconductors disappears, along with AI development, smartphone production, and military electronics for the US and its allies.

The CHIPS Act is, at its core, a bet: $52 billion in subsidies to rebuild domestically what 35 years of globalization had concentrated in Taiwan.

The decision moment

You are a senior strategy consultant advising TSMC's board on its geographic diversification strategy. It is May 2024, two months after the CHIPS Act grant announcement, and one month after TSMC publicly announced delays to the Arizona fab timeline (from 2024 to 2025 for first production). The delays were attributed to a shortage of specialized technicians; TSMC flew over 500 Taiwanese engineers to Arizona to compensate.

Three strategic questions frame your advice:

1. The ecosystem problem. TSMC's competitive advantage does not reside solely in its equipment or its processes, it resides in the 35-year accumulated knowledge of its workforce and the ecosystem of Taiwanese suppliers, equipment manufacturers, and process engineers that surrounds its Hsinchu campus. When TSMC builds a fab in Arizona, it is moving the physical building and some of the equipment. It is not moving the ecosystem. Morris Chang estimates Arizona wafer costs at 50% above Taiwan. How long does the ecosystem gap persist, and is there a point at which the Arizona fabs become genuinely cost-competitive, or will they always be a national security infrastructure investment that operates above market price?

2. The silicon shield paradox. Taiwan's geopolitical protection has historically derived from its economic indispensability: destroying Taiwan would destroy the global semiconductor supply chain, which no rational actor would do. This "silicon shield" theory implies that TSMC's concentration in Taiwan is strategically beneficial to Taiwan's security. If the US CHIPS Act succeeds in diversifying leading-edge manufacturing to Arizona, Japan (Kumamoto fab), and Germany (Dresden fab), does Taiwan become more or less secure? The silicon shield requires TSMC to remain concentrated in Taiwan to be effective.

3. The customer concentration risk. Apple represents ~25% of TSMC's revenue. NVIDIA is growing rapidly as its AI chip volumes expand, and may represent 15-20% of revenue by 2025. If Apple moves even a portion of its production to Samsung Foundry or Intel (which it did briefly with some Mac chips), or if NVIDIA builds its own manufacturing capacity (unlikely near-term), TSMC's revenue would be materially impacted. How does TSMC manage the tension between its dominant position (which requires being the best option for all customers) and customer concentration risk (which creates dependency on specific customers' technology roadmaps)?

Key financial datapoints

The competitive moat mechanics

TSMC's technological lead is the result of compounding advantages across four dimensions:

Process technology: TSMC has been first to production at every leading-edge node for 10+ years. Each new node requires 2-5 years of R&D and $10-15B in capital investment. The knowledge embedded in each node's development accumulates across subsequent generations.

Customer ecosystem lock-in: Designing chips for TSMC's process is a 2-3 year engineering investment. When Apple designs the A18 chip for TSMC's 3nm, those design rules, simulation models, and IP blocks are specific to TSMC's process. Switching to Samsung Foundry requires a full redesign, at enormous cost and with risk of performance reduction.

Equipment relationships: TSMC has exclusive or priority access to critical equipment from ASML (EUV lithography), Applied Materials, and Tokyo Electron. Its scale (~$40B in annual capex at peak) gives it leverage over equipment roadmaps that no competitor can match.

Yield expertise: Semiconductor manufacturing yield (the % of chips on a wafer that work correctly) is the primary source of economic advantage. TSMC's yields at 3nm are reportedly 80%+; competitors launching at the same node typically start at 40-50%. This yield advantage translates directly to cost per chip.

Frameworks invoked

• Resource-Based View. TSMC's competitive advantage comes from resources that are valuable (leading-edge chips), rare (only TSMC and partially Samsung can make them), inimitable (the accumulated manufacturing knowledge cannot be copied by writing a check), and non-substitutable (no alternative technology can replace advanced logic chips in the near term). The RBV predicts that TSMC's moat will persist as long as these four conditions hold, and the US CHIPS Act is an experiment in whether massive capital subsidies can overcome the "inimitable" criterion.
• Geopolitical Risk in Supply Chain Strategy. Supply chain decisions that were optimized for cost efficiency during the 1990-2020 globalization era are now being re-optimized for resilience under geopolitical fragmentation. The CHIPS Act represents a recognition that "just-in-time" global supply chains are a national security vulnerability when the concentration point is a geopolitical flashpoint. The strategic question is not whether to diversify, the US government has decided that, but at what cost and on what timeline.
• International Expansion with Ecosystem Dependency. TSMC's Arizona challenge illustrates a general principle: when competitive advantage is ecosystem-dependent (not just asset-dependent), geographic expansion requires moving the ecosystem, not just the factory. TSMC's attempts to accelerate the ecosystem by flying Taiwanese engineers to Arizona and offering cultural training programs for American workers is the right direction; the question is whether a 35-year ecosystem can be replicated in 5-10 years at any price.
• Customer Concentration and Platform Power. TSMC's 90% market share at leading-edge nodes creates a platform dynamic: customers need TSMC more than TSMC needs any single customer. This gives TSMC pricing power (advanced node prices have increased 5-10% annually) and supply allocation authority (TSMC decides who gets capacity during shortages). Customer concentration risk is real but secondary to the structural leverage of platform position.

Discussion questions

1. Morris Chang publicly stated that manufacturing chips in Arizona will cost approximately 50% more than in Taiwan, and that this cost premium cannot be eliminated by subsidies alone, since it reflects permanent differences in labor costs, supply chain density, and regulatory environment. If he is correct, does the CHIPS Act create sustainable US chip manufacturing, or permanently subsidized US chip manufacturing that ends when subsidies end?
2. The "silicon shield" theory holds that Taiwan's chip manufacturing indispensability protects it from Chinese military action. If the US CHIPS Act succeeds in diversifying leading-edge manufacturing globally, Taiwan loses some of its silicon shield protection, meaning US industrial policy may inadvertently reduce Taiwan's security. How should TSMC's board weigh the US government's CHIPS Act requirements against Taiwan's national security interests?
3. NVIDIA's H100/H200/B200 GPUs, which power the AI training market, are manufactured exclusively at TSMC. NVIDIA's revenue grew from $27B (FY2023) to $61B (FY2024) to a forecast $130B+ (FY2025), driven by AI chip demand. TSMC's capacity is the primary constraint on this growth. How should TSMC price its leading-edge capacity during a period of extreme demand, and what are the risks of capacity allocation decisions that favor some customers over others?
4. Intel's Pat Gelsinger pursued an aggressive strategy to recapture TSMC's leading-edge customers through Intel Foundry Services (IFS). Gelsinger was removed as CEO in December 2024. Intel's foundry ambitions have been significantly scaled back. What does Intel's failure to challenge TSMC tell you about the nature of TSMC's competitive moat, and which of the four RBV criteria (valuable, rare, inimitable, non-substitutable) proved most durable?
5. Japan (Kumamoto), Germany (Dresden), and the US (Arizona) have each provided massive government subsidies to attract TSMC fabs. Each fab produces chips at less advanced nodes than Taiwan's leading-edge fabs. From TSMC's perspective, these fabs provide geographic diversification and government relations benefits; from the host governments' perspective, they provide national semiconductor capacity. Is this an alignment of interests, or does TSMC get more value from the relationship than the host governments do?

The real outcome (revealed at session end)

February 2024: TSMC's Kumamoto fab (Japan) starts production, on schedule. The fastest TSMC geographic expansion in history, attributed to Japan's construction efficiency and existing semiconductor supply chain.

April 2024: TSMC Arizona delays first production from 2024 to 2025; blames US construction workforce shortage and regulatory complexity.

2024: TSMC revenue surges 34% to $90B+ driven by AI chip demand (NVIDIA, AMD, Apple). Wafer capacity remains the binding constraint on AI hardware availability.

December 2024: US government approves $6.6B CHIPS grant to TSMC Arizona, conditional on meeting production milestones.

The lesson: Geopolitical supply chain risk cannot be resolved on a commercial timeline. It took 35 years of market-driven concentration to create TSMC's Hsinchu cluster; it will take 20+ years and hundreds of billions in subsidies to create comparable alternatives. The CHIPS Act buys insurance, not parity. The most important strategic question for any company dependent on advanced chips is: what does your supply chain resilience plan look like for the decade before US manufacturing catches up?

Sources

• TSMC Annual Report 2023 and 2024 investor presentations.
• US CHIPS Program Office, TSMC grant announcement (April 2024).
• Morris Chang, TSMC Foundation speeches on Arizona manufacturing costs (2022–2023).
• Semiconductor Industry Association (SIA), "State of the US Semiconductor Industry" (2023).
• Bloomberg, "TSMC's Arizona Chip Factory Faces a Skills Shortage" (2023).
• The Economist, "The Taiwan chip crisis" (special report, 2023).
• Chris Miller, "Chip War: The Fight for the World's Most Critical Technology" (2022).