VOLUME V · CHAPTER 54 · THE TAIWAN DILEMMA · INTERLEAF
THE BOOK FINALE · UNRESOLVED
The hedge that is not a substitute
The factory
that cannot be
replaced.
By 2024, TSMC was building fabs in Phoenix, Kumamoto, and Dresden. The total committed investment exceeded one hundred billion dollars. None of the new sites would manufacture the leading edge. Every wafer of N3 and below the world buys between 2022 and 2030 will still be made on Taiwan.
// CHAPTER 54 — THE BOOK ENDS HERE
// THE NEXT CHAPTER IS NOT YET WRITTEN
"Every advanced TSMC node still under volume manufacture in 2024, from N5 to N3, ran on Taiwanese soil, in three locations on a single island."
Chapter 54 — The Taiwan Dilemma
The hedge is real, expensive, and globally distributed. Phoenix Fab 21 entered N4 production in early 2025 at yields TSMC describes as comparable with Taiwan. The total Arizona commitment grew from $12B in 2020 to $65B across three planned fabs by November 2024 — the largest greenfield foreign direct investment in US history. JASM in Kumamoto opened February 24, 2024. ESMC in Dresden broke ground August 20, 2024.
Each project was strategically necessary. None was leading-edge. The hedge bought Phoenix, in steady-state, around three percent of TSMC's leading-edge capacity. The hedge would mean that, in the days after a war, the world would still possess the institutional knowledge of leading-edge logic manufacturing inside a US jurisdiction. The hedge was not a substitute. It was a residue.
// FIG. I — TSMC CAPACITY BY SITE
The visible iceberg · steady-state · ca. 2025
Three percent.
Twenty-eight nanometers.
The Sonoran Desert.
The three diversification projects expand mature-node capacity in three jurisdictions where Taiwan's geopolitical risk is absent. Their leading-edge capacity, in steady state, rounds to a single-digit percentage of total TSMC output. The leading edge stays in Hsinchu, Taichung, Tainan.
// % OF TSMC LEADING-EDGE CAPACITY (STEADY-STATE) · APPROX. CHART NARRATIVE
Taiwan · The Original
Hsinchu · Taichung · Tainan
3 sites · west coast · 200km strip
~90% of leading-edge global supply
~90%
NODES:N3 / N5 / N7 / N16 / N28
R&D:EXCLUSIVELY HERE
HEADCOUNT:~70,000 in TSMC alone
N3 volume began Dec 29, 2022. N2 ramping in Hsinchu and Kaohsiung 2025–26.
United States
Phoenix, AZ
Fab 21 · 3 phases planned · north Phoenix
~3%
NODES:N4 (live, 2025) · N3 / N2 (planned)
INVEST:$65B (was $12B in 2020)
SUBSIDY:$6.6B grant + up to $5B loan
Largest greenfield FDI in US history. Fab 21 entered N4 production early 2025.
Japan · Kyushu
Kumamoto
JASM · JV with Sony · Denso · Toyota
Mature nodes · autos & image sensors
~2%
NODES:28 / 22 / 16 / 12 nm
OPENED:FEB 24, 2024
FAB 2:2027 · FAB 3 NEGOTIATING
Targets the auto and CMOS image-sensor market that Sony and Denso run.
Germany · Saxony
Dresden
ESMC · JV with Bosch · Infineon · NXP
Mature nodes · automotive Europe (2027)
~1%
NODES:22 / 28 nm
BREAKING:AUG 20, 2024
EU AID:€5B · von der Leyen + Scholz on dais
Targets European automakers. First production: 2027.
// FIG. II — WHY DIVERSIFICATION IS HARD
The Morris Chang argument · April 14, 2022
"The cost premium is something I believe
the US will eventually be unwilling to pay."
Chang's most pointed version of the argument came at a Brookings/CSIS conversation. The interviewer asked him about the Pat Gelsinger line then circulating in Washington — that Taiwan was not safe and the US therefore needed its own leading-edge fabs. Chang gave an answer the Taipei Times paraphrased as exercise in futility.
Taiwan baseline
1.0×
Reference cost · all-in fab construction + operation per wafer
Arizona — Chang's claim
+50%
Higher than Taiwan, due to ecosystem, talent, and construction costs
Time to rebuild ecosystem
~1gen
"The supporting ecosystem the US has let migrate over forty years would take a generation to rebuild."
Morris Chang · age 90 · Brookings · April 14, 2022
If there is no war, the entire effort to onshore advanced semiconductor manufacturing in the United States is a wasteful and expensive exercise in futility, and if there is a war, everyone is going to have a great deal more to worry about than chips.
— paraphrased by Taipei Times · the implication was almost theological
// FIG. III — THE OPENING SCHEDULE
Four fabs · four jurisdictions · five years
The hedge has a calendar.
The diversification effort is sequenced. Each opening is a milestone. None replaces the leading edge in Tainan.
DEC 29, 2022
Tainan · Fab 18
N3 · VOLUME PRODUCTION
N3 RAMP
8-phase complex. >950,000 sq m. Begins 3-nanometer volume production two days before the calendar year closes.
FEB 24, 2024
Kumamoto · JASM
28 / 22 / 16 / 12 nm
~$8.6B (Phase 1)
JV with Sony, Denso, Toyota. Fab 1 opens. Fab 2 committed 2027. Fab 3 under negotiation.
AUG 20, 2024
Dresden · ESMC
22 / 28 nm
~€10B · €5B EU subsidy
JV with Bosch, Infineon, NXP. Groundbreaking with von der Leyen and Scholz. First production: 2027.
EARLY 2025
Phoenix · Fab 21
N4 (now) → N3 → N2
$65B across 3 fabs
N4 production yields described as "comparable with Taiwan." Fab 2 (N3) targeted ~2028. Fab 3 (N2) targeted ~2030.
Whether the dependency deters Beijing or attracts it,
whether the diversification effort can be scaled fast enough,
whether 2049 is a target or a metaphor —
are questions to which no one has a confident answer.
The CSIS wargame has been run twenty-four times. The runs have not been run for real. The lit fabs of Hsinchu and Tainan keep running. The wafers keep coming off the line, on a frequency that has not yet been interrupted. The book of how the world's most important industry came to depend on the political stability of a single island ends on this question. The answer is the next chapter, and the next chapter is not yet written.