Posts
The Foundry Doctrine
How a Four-Day Business Plan in 1987 Became the Hardware of Geopolitical Order
The Redundancy Tax
TSMC is voluntarily accepting lower margins to build fabs in Arizona, Japan, and Germany that it does not need for efficiency and did not want for profit. That is not a corporate concession. It is a structural signal — the architecture paying the price of its own survival in a world it was not originally designed for.
For the first thirty years of its existence, TSMC operated from a single geography. All of its leading-edge fabs were in Taiwan — clustered in and around Hsinchu Science Park, where the density of suppliers, engineers, equipment specialists, and institutional knowledge made the production ecosystem essentially irreproducible elsewhere. The concentration was not carelessness. It was deliberate optimization. Manufacturing at the frontier of semiconductor process technology requires the kind of deep local ecosystem that takes decades to build and cannot be meaningfully replicated by writing a check, however large.
TSMC understood this better than anyone. Which is why the decision to build outside Taiwan — announced with increasing commitment from 2020 onward, accelerating through the CHIPS Act era, now fully underway in Arizona, Kumamoto, and Dresden — is not a reversal of the founding logic. It is the founding logic responding to a changed threat environment. The architecture is not being abandoned. It is being adapted, at significant cost, to survive conditions that 1987 could not have anticipated.
That cost has a name. This series has been calling it the redundancy tax. Post 6 is where we measure it.
What the Tax Actually Is
The redundancy tax is not a line item on TSMC's income statement. It is the aggregate margin dilution produced by operating fabs in geographies where the cost structure is fundamentally higher than Taiwan and where the production culture that drives TSMC's yield advantage — the accumulated institutional knowledge of tens of thousands of engineers trained in Hsinchu over three decades — cannot be fully transplanted on any near-term timeline.
TSMC's gross margins in recent years have run in the 53–56% range for Taiwan operations at steady state. Overseas fab operations are estimated to dilute consolidated gross margins by approximately 2–4 percentage points during the ramp phase — a drag that reflects higher construction costs, higher labor costs, lower initial yields as new teams learn processes that their Hsinchu counterparts have optimized over years, and the inefficiency of managing a geographically dispersed operation from a headquarters culture built around geographic concentration.
At TSMC's current revenue scale — approximately US$35.71 billion in Q1 2026 alone — 2–4 percentage points of gross margin represents billions of dollars annually. This is not a rounding error. It is a deliberate, ongoing structural cost that TSMC's board has chosen to absorb because the alternative — remaining exclusively in Taiwan — carries a different kind of cost that does not show up in the income statement until it does catastrophically.
The Three Sites and What Each One Buys
Fabs 21 & 22
Japan Advanced Semiconductor Manufacturing
European Semiconductor Manufacturing Company
The Actual Frontier
The fab map above contains the most important structural observation in this post: every overseas site is buying a specific political relationship. Arizona buys Washington. Kumamoto buys Tokyo. Dresden buys Brussels. None of them buy operational independence from Taiwan — the leading edge remains in Hsinchu and Tainan, as it always has, as it will continue to for the foreseeable future. The overseas fabs produce mature or near-frontier nodes at higher cost. They are not substitutes for Taiwan. They are insurance policies written in concrete and silicon.
The Culture Problem That Money Cannot Solve
The margin dilution is the visible cost of the redundancy build. The less visible cost — and arguably the more durable one — is the culture problem: the difficulty of transplanting the production culture that drives TSMC's yield advantage to geographies where that culture does not exist and cannot be rapidly manufactured.
TSMC's yield advantage over its nearest competitors is not primarily a technology advantage, though the technology is genuinely superior. It is an execution advantage — the product of decades of accumulated process knowledge, quality discipline, and workforce culture that makes TSMC's fabs systematically more reliable than alternatives. That culture was built in Taiwan, by Taiwanese engineers trained in the TSMC system, operating in an industrial ecosystem where the norms of semiconductor manufacturing are embedded in the workforce at every level.
The Arizona experience has been instructive. TSMC initially planned to staff its Arizona fabs primarily with locally recruited American engineers. The yield and quality results were not meeting TSMC's standards on the original timeline, and the company responded by flying in hundreds — eventually thousands — of Taiwanese engineers to stabilize operations and transfer process knowledge. The cultural and logistical friction this produced became public in 2022 and 2023, surfacing in worker accounts of communication breakdowns, differing workplace norms, and the fundamental difficulty of replicating in eighteen months what took thirty years to build in Taiwan.
TSMC has not abandoned the Arizona project — the scale of committed investment and the CHIPS Act obligations make that inconceivable. But the experience has made visible a constraint that pure capital expenditure analysis obscures: the redundancy tax is not paid once, at construction. It is paid continuously, in the form of lower yields, higher defect rates, and the ongoing cost of knowledge transfer that closes — slowly — over years of operation.
The Anatomy of the Tax
| Cost Driver | Taiwan Baseline | Overseas Premium | Duration |
|---|---|---|---|
| Construction cost per sqft | Optimized over decades; deep local supply chain | Estimated 2–4× higher in U.S.; 1.5–2× in Japan/Europe | One-time; partially offset by government subsidy |
| Labor cost | Highly trained workforce at Taiwan wage levels | U.S. engineering salaries 2–3× Taiwan equivalent roles | Permanent; structural feature of each geography |
| Yield ramp timeline | Overseas fabs estimated 24–36+ months to equivalent yield | Temporary; closes as workforce matures — over years | |
| Supplier ecosystem | Dense local network; just-in-time parts and expertise | Thin local base; long supply lines; higher logistics cost | Partially permanent; ecosystem takes 10–15 years to develop |
| Knowledge transfer | Embedded; engineers trained in the system from entry | Active; requires Taiwanese engineer deployment at scale | Diminishing; 5–10 year transfer horizon at current pace |
| Management overhead | Single-timezone, single-culture coordination | Multi-timezone, multi-culture; significant coordination cost | Permanent feature of geographic dispersion |
The table makes visible what the aggregate 2–4 point margin dilution estimate conceals: the tax has multiple components with different durations and different trajectories. Some of it — construction cost, initial yield drag — is temporary and will diminish as the overseas fabs mature. Some of it — labor cost differentials, supplier ecosystem thinness, management overhead — is structural and permanent. The aggregate dilution will narrow over time, but it will not reach zero. Overseas fabs will always be more expensive to operate than Taiwan fabs, because the conditions that make Taiwan operations cost-efficient are properties of Taiwan — its workforce, its ecosystem, its three-decade accumulation of semiconductor industrial culture — not of TSMC alone.
Why the Architecture Is Paying It Anyway
The redundancy tax is being paid because the alternative assessment — that Taiwan concentration is an acceptable risk — has been revised by the people with the most information about the risk. TSMC's board, the Taiwanese government, Washington, Tokyo, and Brussels have all arrived at the same conclusion through different analytical paths: a world in which the entire leading-edge semiconductor supply chain sits on a single island in a contested strait is a world with a single point of failure that no insurance policy can adequately cover.
That conclusion does not require a belief that conflict over Taiwan is probable. It requires only a belief that the consequences of such conflict — for the global technology infrastructure, for military capability on both sides, for the AI programs that both superpowers have made central to their strategic competition — are severe enough to justify paying a continuous margin tax to reduce the probability of a catastrophic single-point failure.
This is antifragility logic, not efficiency logic. The architecture is spending margin to buy resilience. The fact that it is doing so — visibly, at scale, at a cost that shows up in every quarterly earnings call — is the clearest possible signal that the people running the architecture believe the underlying risk is real.
The redundancy tax, in the end, is the architecture's honest answer to the question the FSA Wall in Post 5 could not resolve: whether the design is stable under maximum stress. TSMC cannot answer that question in words — the scenario is too sensitive, the variables too uncertain, the planning too classified. It answers it instead in the only language that a publicly traded company can use with full credibility: capital allocation. Fifty-two to fifty-six billion dollars in 2026 capex, spread across four continents, weighted toward geographies chosen for political alignment rather than operational efficiency.
The architecture believes the risk is real. The tax is the proof. Post 7 asks the final question: in a world where both versions of the blueprint are running simultaneously, which one generates more durable capability through 2036?
- TSMC Arizona (Fab 21): 4nm volume production 2024; 3nm Phase 2 2025–2026; ~$65B total committed investment
- TSMC Arizona (Fab 22): 2nm and A16 announced 2023; production timeline 2028+
- CHIPS Act direct award to TSMC Arizona: up to ~$6.6B (2024 agreement)
- JASM Kumamoto Fab 1: 12/16nm and 22/28nm, volume production 2024; Japanese government subsidy ~¥476B (~$3.2B)
- JASM Kumamoto Fab 2: 6nm announced 2024; production ~2027; additional Japanese government subsidy committed
- ESMC Dresden: 12/16nm, production target 2027; EU subsidy ~€5B; total investment ~€10B
- Overseas fab gross margin dilution vs. Taiwan baseline: estimated 2–4 percentage points (TSMC management guidance and analyst estimates)
- TSMC 2026 total capex guidance: US$52–56B — highest in company history
- Arizona workforce: thousands of Taiwanese engineers deployed for knowledge transfer (2022–present)
- Taiwan leading-edge operations: 2nm volume production (late 2025); A16 ramp 2H 2026 — frontier unchanged
The precise gross margin impact of each overseas fab — disaggregated by site, by node, and by phase of ramp — is not publicly disclosed by TSMC at that level of granularity. The 2–4 percentage point aggregate dilution estimate is derived from management commentary, analyst models, and comparison of TSMC's reported margins against periods of pure Taiwan operation. The actual per-site cost structure, yield trajectories, and knowledge transfer timelines are proprietary. The Arizona workforce deployment numbers — the scale of Taiwanese engineer secondments — are known through news reporting and worker accounts but not formally disclosed by TSMC. The Wall stands at the boundary between the observable aggregate margin impact and the site-level economics that would allow a full cost-benefit analysis of each location independently.
Primary Sources · Post 6
- TSMC Earnings Calls and Investor Days (2022–2026) — overseas expansion capex; margin guidance; management commentary on overseas cost structure
- U.S. Department of Commerce, TSMC Arizona CHIPS Award (2024) — direct funding terms; investment commitments
- Japan Ministry of Economy, Trade and Industry (METI) — JASM subsidy announcements; ¥476B commitment documentation
- European Chips Act (2023), EU Regulation 2023/1781 — framework for ESMC Dresden subsidy; ~€5B allocation
- TSMC press releases: Arizona Fab 21/22 announcements (2020–2023); JASM Fab 1/2 announcements; ESMC announcement (2023)
- Reuters, Bloomberg, Nikkei Asia reporting on Arizona workforce deployment and knowledge transfer challenges (2022–2023)
- TSMC Annual Reports (2022–2025) — geographic segment reporting; capex breakdown by region
No comments:
Post a Comment