Taiwan, Semiconductors, and a "Cold New War"? · Prep

The Professor's Questions

The United States and China competed over dominance of the global semiconductor industry in the 2020s. Beijing’s Made in China 2025 initiative aimed to achieve 70 percent self-sufficiency in semiconductor technology. Meanwhile, the U.S. Congress debated a bill authorizing $52 billion in support for the semiconductor industry. The so-called “tech war” posed pressing questions for Taiwan, the world’s leading manufacturer of semiconductors. Would the competition between the United States and China erode the “silicon shield” that Taiwanese leaders counted on to preserve their island’s autonomy? And what consequences could the world expect from the seeming deglobalization of the semiconductor industry?

Assess the effects of government policies on the development of the semiconductor industry in the United States, Taiwan, and China.
How should Taiwan navigate the rise of Sino-American tensions? Is its semiconductor industry a source of protection or vulnerability?
Does the competition over Taiwan and the global semiconductor industry indicate a world of deglobalization? Why or Why not? If so, what likely consequences might follow from deglobalization?

My Answers

Q1. Government policy and the industry

Government policy built this industry in all three places. It just built three different kinds of power.

In the United States, the state created the science, then stepped back from the factory. Bell Labs needed wartime purification work at MIT. The Pentagon paid for transistors, set standards, and financed plants. The Justice Department forced AT&T to open its patents. That open IP cadence, plus venture capital, produced Fairchild, Intel, and later the fabless model. When Japan’s share jumped from 28 percent in 1978 to 46 percent in 1986, Washington did not stay liberal: it threatened anti-dumping duties, taxed $300 million of Japanese electronics, and stood up SEMATECH at about $200 million a year. The later U.S. edge is not in making wafers. Exhibit 11: America still takes 38 percent of value added and 74 percent of EDA and core IP. What it lost is the leading-edge factory. Intel missed EUV. Nvidia, a U.S. design firm that does not own a fab, became the most valuable chip company — and it is a TSMC customer. The CHIPS Act’s $52 billion is an attempt to buy the factory back. BCG’s own math in the case: $50 billion of subsidy lifts the U.S. production share from 12 to 14 percent. Morris Chang called it an expensive exercise in futility. Policy can seed an industry. It cannot, in one bill, rebuild a work culture that fixes a tool at 2 a.m.

In Taiwan, policy chose a narrow bet and stuck to it. Land reform, export zones, vocational schools, and a 30 percent engineering share of university students built the human base. ITRI did applied research, not basic science, and leaked 12,000 engineers into firms. ERSO bought RCA’s obsolete seven-micron node, beat RCA’s own yields, and spun out UMC. Chang then refused to become another integrated device maker. He used the state to build a pure-play foundry, licensed Philips technology only nine months off the frontier, and let design houses bloom in Hsinchu — more than forty in a year. That single institutional choice created the global fabless industry. Apple, Qualcomm, and Nvidia exist in their current form because Taiwan’s government absorbed manufacturing risk. By 2019 Taiwan held 92 percent of wafer capacity for logic below 10 nanometers. The developmental state did not pick “electronics” in the abstract. It picked the one slice of the chain that matched what Taiwan could do, and what it could not.

In China, policy has been bigger, broader, and less precise. The National IC Plan aimed at $150 billion. China supplied 86 percent of worldwide state equity support for chip firms from 2014 to 2018. SMIC’s government stake rose from under 15 percent to over 45 percent. Made in China 2025 wrote down a 70 percent self-sufficiency target. The toolkit included forced joint ventures, $35 billion of outbound chip M&A in a single year, talent programs, and civil-military fusion. The result, by the case’s own scoreboard, is mixed. About 200 new chip firms were registered per day in early 2020. SMIC announced 7-nanometer chips. Tsinghua Unigroup went bankrupt. Chinese-owned firms still made only 5.9 percent of the chips China used. Huawei’s Mate 60 raised the China-origin share of components from 29 to 47 percent — real progress, still without EUV, still on a process analysts called unreliable. Money created activity. It has not yet created TSMC.

The comparative lesson: policy is not a on/off switch. U.S. policy created an open knowledge system and then offshored the plant. Taiwanese policy created a specialist plant and invited the world to use it. Chinese policy is trying to buy the whole chain at once, under a security clock. The first two worked because they chose a slice. The third is still paying for wanting every slice.

Q2. How Taiwan should navigate

Taiwan should treat the silicon shield as a wasting asset. Keep the leading edge on the island. Put just enough capacity in the United States to keep Washington’s skin in the game. Build a porcupine. Do not declare a new republic. Do not bet the island on “one country, two systems.” The industry is still protection. It is becoming vulnerability faster.

Why it is still protection. In 2019 Taiwan held 92 percent of wafer capacity for logic below 10 nanometers. Only three firms on earth could do 5 nanometers. A war that breaks TSMC is not a local war. It is a shutdown of phones, clouds, cars, and AI training. That is why Tsai wrote to Foreign Affairs, why Liu told 60 Minutes the world would not let war happen, and why Lai, on the night he won, called the industry “a common asset of the world.” The shield is real in one direction: it raises the cost of a kinetic war for everyone who uses chips.

Why it is becoming a vulnerability. Both patrons are now paying to stop needing Taiwan. Beijing’s target is 70 percent self-sufficiency. Washington’s CHIPS Act and the October 2022 export rules are designed so that advanced logic does not have to live in one strait. Every successful Arizona tool-install makes the sentence “the world can only make this chip in Taiwan” a little less true. The customer base is already political: Huawei was TSMC’s second-largest customer, more than 15 percent of revenue, and disappeared by regulation, not by the market. China still takes 44 percent of Taiwan’s exports. That is not a diversified book of business. It is a single counterparty that has already used pineapples as a sanction.

The military numbers make the vulnerability worse. Conscription is four months. Headcount is about 80 percent filled. Defense spending fell from roughly 5 percent of GDP around 1990 to about 2 percent (Exhibit 8a). Half the public says it would fight; most believe it will never have to, because America will come. Pentagon war games often have the red team win. Buying 66 F-16s for $8.1 billion is the opposite of a porcupine. High-prestige platforms die in a first strike. Anti-ship missiles, mines, and air defense do not look like a silicon shield. They are what you buy when you stop using TSMC as a substitute army.

So the navigation is a portfolio, not a slogan.

Keep N-2 and below, and the yield knowledge, on the island. That is the last piece of leverage that is actually scarce. Do not let CHIPS money relocate the frontier.
Accept a costly, limited U.S. footprint. Arizona at four times Taiwan’s construction cost is an insurance premium paid to Congress and to TSMC’s American customers. Chang already said the quiet part: globalization is almost dead, so you build there anyway. The point is political entanglement, not cost.
Lengthen conscription, fill the ranks, and buy the ugly weapons. The shield cannot remain the only defense. A four-month summer camp plus a $8.1 billion fighter deal is how a rich island pretends a foundry is a navy.
Hold the political status quo in words and in law. Exhibit 6 is the mandate. Immediate independence invites the invasion the case says Beijing has promised. Immediate unification on Hong Kong’s formula is what Taiwanese voters rejected after 2019. Ambiguity is not weakness here. It is the product the median voter already bought.
Slowly reduce the 44 percent. Not a sudden decoupling — Taiwan is the world’s second-most exposed economy to global value chains. A multi-year shift of assembly and final demand toward Europe, Japan, Southeast Asia, and the U.S. makes the pineapple episode harder to repeat at chip scale.

The cost of this package is explicit. Arizona raises TSMC’s cost base. A porcupine raises the defense budget. Diversifying away from the mainland costs exporters. The cost of the alternative is larger: a shield that Washington and Beijing are both working to make obsolete, sitting on an island whose young men binge McDonald’s to fail the physical, while 92 percent of the world’s most advanced wafers wait in one typhoon path.

Q3. Deglobalization or not

No. This is not deglobalization. It is the same global chain, with permission to ship now treated as a weapon.

A deglobalized semiconductor world would look like three complete stacks: American design plus American fabs plus American tools; a Chinese stack that can live without EDA, EUV, and Nvidia; a Taiwanese or Korean stack that does not sell to either. The case does not show that world. It shows a fight over a still-shared chain.

The evidence against the word. In 2019 the United States still did 74 percent of EDA and core IP. The only EUV firm is Dutch, and Washington can block a shipment only because American parts sit inside the machine. TSMC’s customers are American fabless firms. China still bought near-record volumes of chips and tools in 2023. Nvidia did not leave China; it redesigned cards to sit just under the threshold. The Netherlands and Japan joined export controls — that is alliance politics, not autarky. Chip 4 is a club inside the old network, not a breakup of the network. Taiwan remains the world’s second-most exposed economy to global value chains. A world that was deglobalizing would be reducing that score. It is arguing about who is allowed to use the same scoreboard.

The evidence that looks like deglobalization is real, and it is narrower. CHIPS money tries to pull leading-edge capacity onto U.S. soil. CFIUS now reviews deals with almost no U.S. footprint, like Magnachip. The Foreign Direct Product Rule claims jurisdiction over chips that contain no U.S. atoms. Chang said globalization is almost dead and free trade is almost dead. He is describing the political overlay, not the production graph. Wafers, software, and talent still cross borders. What no longer crosses freely is permission.

If we still want a word, use weaponized interdependence, or securitized globalization. The frontier is being recoupled along alliance lines. Mature nodes are still global, and may become more global in the wrong way: TSMC’s CEO said in 2024 there might be too much capacity being built for mature nodes, exactly what you get when every capital dumps subsidy into the easy part of the chain.

Consequences, even without full deglobalization:

Higher costs, forever. Arizona is about four times Taiwan. That cost will sit in every iPhone and every server, or in a government budget that never quite sunset.
A glut at the trailing edge, a shortage at the frontier. Duplicate mature fabs plus Chinese subsidy is how you get dumping. The scarce thing — EUV, yield, 5-nanometer discipline — stays scarce longer, because you cannot subsidy-clone a 700-step process.
A slower Moore’s Law. The old model socialized manufacturing risk in one specialist island so that a hundred design firms could iterate. Three half-copies of that island waste capital on concrete instead of nodes.
Two permission regimes, not two planets. A China-accessible stack and an America-accessible stack will share tools and older nodes, and diverge on AI accelerators and anything that looks like a military input. Families, carmakers, and cloud firms will keep two bills of materials.
The bill lands on small open economies first. Taiwan’s 44 percent, Korea’s memory plants, Dutch export licenses. Deglobalization talk in Washington is paid for, in the first round, by places that actually live inside the chain.

So if someone in class says “the chip war proves globalization is over,” the reply is: globalization of atoms is still running. Globalization of permission is being rewritten. Those are different sentences. The first one is why a war over Taiwan still crashes the world. The second one is why both superpowers think they can, in time, afford to care less.

Case Summary

In autumn 2021, Tsai Ing-wen addressed the world in Foreign Affairs. Chinese military aircraft were entering Taiwan's air defence identification zone in record numbers. Defence Minister Chiu Kuo-cheng judged that by 2025 China would be able to attack Taiwan "at the lowest possible cost and losses." Tsai wrote that if Taiwan fell, it would be a catastrophe for regional peace and the democratic alliance, tantamount to authoritarianism prevailing over democracy in the contest of values. She also knew that ideals often dissolve in international relations, so she offered another reason — TSMC, and Taiwan's semiconductor industry more broadly, was a "silicon shield." Chairman Mark Liu told 60 Minutes that the whole world needs Taiwan's high tech, and therefore "will not let a war happen in this region." Yet Beijing and Washington were both pouring in money to break their dependence on Taiwan — and on each other. Some began calling it a "new Cold War."

What Taiwan is, is itself contested. An island of 13,826 square miles, with about 23.5 million people in 2021, more than 95% Han Chinese. It meets Weber's functional definition of a "state," yet since the 1970s has struggled to maintain external recognition. The mainstream view on the island is that the Republic of China is "already independent," and that the ambiguous status quo beats the alternatives. Mainland China regards Taiwan as a breakaway province; its military has built a scale model of the Presidential Office in Inner Mongolia to rehearse "decapitation." Over thirty years, identity has shifted heavily towards "Taiwanese" (Exhibit 7), while positions on unification versus independence are heavily concentrated on "maintaining the status quo" (Exhibit 6).

The island's modern history is a crossroads of empires. Dutch and Spanish outposts, the Ming loyalist Zheng regime, annexation into Fujian by the Kangxi Emperor in 1684 (he had originally said "a ball of mud beyond the sea, of no great consequence"), cession to Japan in 1895. Japan made it a "model colony": primary school enrolment for boys rose from 39% to 80%, while dissent was suppressed. In 1945 the Kuomintang took over; the popular saying was "the dogs left and the pigs came." In the February 28 Incident of 1947, at least 10,000 people were killed. In 1949 about 1.5 million soldiers and civilians retreated to Taiwan and imposed decades of martial law and White Terror on the roughly 6 million islanders. Meanwhile, "Free China" received US$4.2 billion in American aid between 1949 and 1968. Land reform cut the share of tenant farmers from 64% to 10%, and the surplus capital and labour fed import substitution, which then pivoted to export orientation in the late 1950s. In 1966 the world's first export processing zone opened. Upstream state-owned enterprises supplied cheap raw materials, and downstream small and medium-sized firms did contract manufacturing — in Toy Story, Buzz Lightyear finds Made in Taiwan stamped on his wrist.

Semiconductors were the next stop on this upgrading path. A 1966 plan aimed to make Taiwan an "electronics industry centre." The share of vocational education rose from 40% to 60%, and in 1990, 30% of university students studied engineering (India 5%, South Korea 20%). The Industrial Technology Research Institute (ITRI) was founded in 1973 to do applied rather than basic research; in 1974 its Electronics Research and Service Organization (ERSO) bought RCA's already outdated 7-micron process and trained 40 engineers. The pilot line's yields soon exceeded those of RCA's plant in the US. In 1980 it spun off UMC, the first company in the Hsinchu Science Park, which three years later was the most profitable company in Taiwan. Hsinchu deliberately copied Silicon Valley: the government could hold up to 49% of equity, with tax exemptions, low-interest loans, matching R&D funding, plus housing, schools and hospitals. ITRI staff turned over at 10–15% a year, and from 1973 to 2000 about 12,000 engineers moved into industry; in 1999, 110 of the park's 284 companies had been founded by engineers who had returned from the United States.

In the summer of 1985, Morris Chang put down the phone. Born in Ningbo, he went to Harvard to escape the civil war, felt that "in the early fifties, the only career a Chinese American could seriously pursue was technology," transferred to MIT, and spent 25 years in semiconductors at Texas Instruments, rising to group vice president, before leaving in 1983 feeling he had been "put out to pasture." He saw two things: the cost of building a fab was roughly doubling every five years, reaching US$250 million by 1985; and a crop of fabless design companies had begun renting idle capacity from IDMs. Premier Sun Yun-suan asked him to lead Taiwan's semiconductor push. He accepted the presidency of ITRI and proposed spinning off from ERSO a pure-play foundry — one that only manufactures for others, with no design and no brand of its own. This used Taiwan's strengths in manufacturing and engineering while sidestepping its weaknesses in IP-intensive design and marketing. He signed with Philips, bringing in technology only nine months behind the frontier, and founded TSMC. Within a year, more than forty design companies sprang up in Hsinchu. In 1992 AMD's Jerry Sanders was still saying "real men have fabs." In 2009 AMD sold off its own fabs. By 2020 TSMC held half of global foundry capacity, and the overwhelming majority of capacity for the most advanced chips.

This industry has never grown out of the market on its own. Bell Labs produced the germanium transistor in 1947; the Pentagon provided money, set standards and built plants; the Justice Department's antitrust action forced AT&T to open up its patents. Shockley, the "Traitorous Eight," Arthur Rock and venture capital grew Fairchild and Intel in the Stanford Industrial Park. Moore's Law became the metronome. In the 1970s American companies moved assembly and testing to developing economies including Taiwan; Japan's MITI coordinated R&D, subsidised loans and restricted competition, and its share rose from 28% in 1978 to 46% in 1986 while the US fell from 55% to 40%. The Reagan administration forced Tokyo to set export price floors, raise foreign companies' share of the Japanese market from 8% to 20%, and imposed tariffs on US$300 million of Japanese electronics. In 1987 the Defense Science Board said dependence on foreign microelectronics was "unacceptable." Hence SEMATECH, with an annual budget of about US$200 million, split evenly between DARPA and industry. America's share later came back — some credit the consortium, others say it was really the PC and Intel's microprocessors, while the consortium was still focused on already commoditised DRAM.

In 2014, Beijing switched to a bigger industrial policy. The Snowden affair convinced the leadership that core technologies could not be in other people's hands. Xi Jinping's words in 2016: core technology is our greatest vital weakness, and being controlled by others is our greatest hidden danger. The National Integrated Circuit Industry Development Guidelines aimed to reach "advanced international levels" by 2030; Made in China 2025 set a ten-year self-sufficiency target of 70%. This round differed from the 1980s and 1990s: it was about 40 times larger in scale, the strategy was to build national champions through overseas acquisitions plus technology transfer, and execution was handed to local private equity funds. The Big Fund, provincial and municipal funds, and the Phase II fund together added up to the hundred-billion-dollar range. From 2014 to 2018, 86% of the world's state equity support for semiconductors came from China. The government's stake in SMIC rose from under 15% in 2004 to over 45% in 2018. Alongside this came mandatory joint ventures, US$35 billion of outbound acquisitions in 2015 alone, the Thousand Talents Plan, and poaching Richard Chang and over a hundred former employees from TSMC — TSMC's litigation ultimately won it a 9% stake in SMIC and a US$325 million settlement. In the first ten months of 2020, about 200 new semiconductor companies were registered in China every day. SMIC announced a 7-nanometre breakthrough, though analysts still said it was "years behind"; Tsinghua Unigroup went into bankruptcy in 2021. Chinese-owned companies still produce only 5.9% of the chips China uses.

In the 2010s, Washington rewrote Beijing from "responsible stakeholder" into strategic competitor. The export control Entity List, the extraterritorial reach of the Foreign Direct Product Rule (FDPR), and CFIUS's expansion to minority stakes step by step removed Huawei from TSMC's customer list. In September 2022, Jake Sullivan said that it used to be enough to stay a generation or two ahead, but for advanced logic and memory the US now had to maintain "as large a lead as possible." In October, the Commerce Department imposed sweeping controls on advanced chips at roughly 14 nanometres and beyond, manufacturing equipment, IP and support from "US persons," and through the FDPR reached all the way to the EUV machines of the Netherlands' ASML — the only company in the world that can make them. Some analysts called it "a full-blown techno-economic cold war." The results were mixed: the Netherlands and Japan followed in 2023, and Chip 4 took shape; China's imports of semiconductors and equipment that year were still near record highs; Nvidia modified its chips to sit just below the thresholds; Huawei's Mate 60 brought domestic components up to 47%. Romney, who had once attacked Obama for wasteful subsidies to Solyndra, turned to supporting CHIPS: the CHIPS and Science Act authorised US$278 billion, more than US$52 billion of it for semiconductors. Intel, TSMC and Samsung planned fabs in Ohio, Arizona and Texas respectively. Conditions included childcare, sharing of excess profits, no buybacks, and no new high-end investment in China. Boston Consulting Group estimated that US$50 billion in subsidies would only lift the US share of production from 12% to 14%. Morris Chang called it "an expensive exercise in futility." Building in Arizona costs about four times as much as in Taiwan. He said: if a machine breaks at 1 a.m., in the US it gets fixed the next morning; in Taiwan it's fixed by 2 a.m. He also said: globalisation is almost dead, free trade is almost dead. That's why he was still optimistic about Arizona.

The silicon shield theory is being used both on the island and in politics. In 2020 Taiwan was the world's 22nd-largest economy, with exports often exceeding half of GDP, 40% of them semiconductors, and China absorbing 44% of exports. It is the economy with the second-highest exposure to global value chains. Only 13 countries still have formal diplomatic relations with it; the US supplies defensive weapons under the Taiwan Relations Act and "strategic ambiguity," and in Pentagon war games the red team often beats the blue team. Compulsory military service was cut to four months, which some called summer camp, with "strawberry soldiers"; units were filled to only about 80% of establishment, and military spending fell from about 5% of GDP to about 2%. Around half the population says it is willing to fight, and most believe they will never have to and can count on the US. Procurement favours expensive platforms like the F-16 — the Tsai government bought 66 for US$8.1 billion in 2020 — while analysts would rather see a "porcupine": anti-ship, anti-tank and air defence, to fight a survivable war of attrition. In the winter of 2024, Lai Ching-te, whom Beijing calls a "troublemaker," won the presidency, while the Kuomintang won a majority in the legislature. After his victory he said: Taiwan's semiconductors are a shared asset of the world, and he hoped both China and the international community would cherish them. The silicon shield narrative is still there. The case ends on a question: is it alive and well, or slowly and painfully dying?

Key Data

Dimension	Figure / Fact	Source
Global scale	2021 semiconductor revenue $550B; an advanced fab costs about $20 billion and takes two years	p.4
Foundry structure	2020 foundry $85.13B; TSMC about 50% of capacity, largest share of the most advanced nodes	p.8; Exhibit 15
<10nm capacity	2019: Taiwan 92%, South Korea 8%; China 0 at 10nm and below (2019)	Exhibits 12, 14
Who has 5nm	3 worldwide: one each in the US, Taiwan, South Korea	Exhibit 14
Value chain position	US captures 38% of value added and 74% of EDA/IP; Taiwan very strong in manufacturing, about 9% of total value added	Exhibit 11
China's gap	Consumes about 50%, produces (Chinese-owned) only 5.9%; design 20%, foundry 10%	p.8, p.11
China's subsidies	Target $150 billion; 86% of global state equity support 2014–18; government stake in SMIC <15%→>45%	p.9; Exhibit 17
Self-sufficiency target	Made in China 2025: 70% in ten years	p.8; Session description
US reshoring	CHIPS >$52 billion for semiconductors; BCG: +$50 billion only moves 12%→14%	p.12, p.15
TSMC's US bill	Arizona $40 billion, about 4x the cost; Morris Chang: expensive exercise in futility	p.15–16
Huawei shock	Formerly TSMC's second-largest customer, >15% of revenue; revenue down about 1/3 after controls	p.13
Taiwan's external dependence	Exports >50% of GDP, semiconductors 40% of exports, China 44% of exports; 2nd-highest GVC exposure in the world	p.16
Macro buffers	2021 current account 15.3% of GDP; 2023 FX reserves $563.3 billion, debt 26.6% of GDP	Exhibit 2
Security bill	Conscription 4 months; units about 80% filled; military spending/GDP ~2%; F-16s $8.1 billion / 66 aircraft	p.18; Exhibit 8
Diplomacy	13 diplomatic allies; 59 quasi-embassies; strategic ambiguity + Taiwan Relations Act	p.17
Yield discipline	~700 steps; 99% per step → <9% final	p.7
Rare earth countermeasure	35% of reserves, >60% of production, >85% of refining	p.10
Population	23.25 million in 2023, already declining; 18.4 million by 2060, dependency ratio worsening	p.16; Exhibit 2

Estimate: if Huawei contributed >15% of TSMC's revenue, with a 2020 foundry market of $85.13B and TSMC holding about half, then roughly 0.15 × 0.50 × 85.13 ≈ $6.4B a year of foundry demand tied to Huawei was cut off at a political stroke. If American companies get 20–30% of revenue from China, the controls simultaneously cut the cash flow they need to sustain their R&D lead — a contradiction the case itself points out.

Key People

Person	Role	Position / Role in the story	Relationships
Tsai Ing-wen / Lai Ching-te	DPP presidents (2016–24 / 2024–)	Turned the silicon shield into the language of international politics; Lai: semiconductors are "a shared asset of the world"	Facing Beijing and Washington; after Lai's election, the legislature is in KMT hands
Mark Liu / Morris Chang	TSMC chairman / founder	Liu: the world won't let war happen here. Chang: CHIPS is an expensive exercise in futility; globalisation is almost dead, which is why he still went to Arizona	Customers are Apple, Nvidia, Qualcomm; former second-largest customer Huawei was removed by the US
Xi Jinping / Wang Huning	CCP top leadership / ideology adviser	Core technology must not be controlled by others; military-civil fusion; national rejuvenation by 2049; no commitment to renounce force	Industrial policy and security policy tied to the same line
Romney / Biden administration / Sullivan	US politics and national security	Moved from opposing industrial policy to CHIPS; Sullivan: must maintain as large a lead as possible, no longer content with a generation or two ahead	Uses the dual tools of export controls + subsidies; FDPR reaches ASML
Pat Gelsinger	Intel CEO	Wants subsidies for American companies, not for "foreign chipmakers who will keep their IP at home"; meanwhile goes to Europe for subsidies	America's "national champion" fell behind, overtaken by TSMC and Samsung using EUV
Sun Yun-suan / ITRI–ERSO	Architects of Taiwan's developmental state	Brought Morris Chang back; applied research, science parks, equity stakes, spin-offs	Determined that Taiwan went pure-play foundry rather than IDM
Philips / RCA	Technology licensors	RCA gave the outdated 7-micron process; Philips gave a process 9 months behind	The first bite of technology in Taiwan's industrial policy came from foreign licensing, not from working in isolation
Huawei / SMIC / Tsinghua Unigroup	Chinese champions and cautionary tales	Huawei's revenue plunged after being cut off, then the Mate 60 proved the sanctions had gaps; SMIC's 7nm without EUV; Tsinghua Unigroup's bankruptcy	Shows money can pile up companies, but can't pile up stable advanced yields
US fabless firms (Apple, Nvidia, Qualcomm)	TSMC's real customers	Without them, the silicon shield is empty	They need TSMC's advanced capacity, and also Washington's export licences

Decision rights: Taiwan's leaders decide whether to keep advanced capacity on the island, whether to reshape conscription and procurement into a porcupine, and whether to stay ambiguous on unification versus independence. Washington decides export licences and subsidy conditions. Beijing decides technology and trade countermeasures. Morris Chang / TSMC decide where to build fabs — but can no longer do the sums on cost alone.

Core Tensions

Is the silicon shield protection, or is it becoming a weak spot (Professor's Q2, placed first): On the protection side — 92% of the world's most advanced logic capacity is in Taiwan, war would mean a global electronics shutdown, which is why Mark Liu says the world won't let war happen. On the weak-spot side — the US and China are both spending money to "escape Taiwan"; a single customer like Huawei with >15% of revenue can vanish for political reasons; the more concentrated advanced capacity becomes, the more Taiwan looks like a prize rather than a refuge; the island's military spending share has fallen to about 2%, and conscription is down to four months.
Industrial policy can build champions, and it can build waste: ITRI + pure-play foundry grew TSMC; China's hundred billion dollars bought only 5.9% self-sufficiency and a bankrupt Tsinghua Unigroup; America's US$52 billion, per BCG, buys only 2 points of production share. All called industrial policy, under completely different constraints.
Is interdependence a peace dividend, or leverage that can be weaponised: 44% of Taiwan's exports go to China; China consumes half the world's chips but can't make advanced nodes; the US holds EDA, software and the FDPR. Every party both depends on the others and is learning how to cut them off.
Is this deglobalisation, or a political licensing regime on the same chain (Professor's Q3): Morris Chang says globalisation is almost dead; but in 2023 China's imports of chips and equipment were still near record highs, what the Netherlands and Japan followed was controls rather than bringing fabs home, and Nvidia is modifying products rather than exiting Asia.
The security paradox after democratisation: The political miracle followed the economic miracle, and the Economist ranks it the most democratic in Asia; at the same time the military's role has shrunk, soldiers are called strawberries, the porcupine strategy is written in the analyses while F-16s are written on the purchase orders. The silicon shield is filling a gap the military should be filling.
Building fabs in the US: reinforcing the shield, or dismantling it: Arizona gives Washington skin in the game, but also begins to make the statement "the world can only make the most advanced chips in Taiwan" untrue. The fourfold cost is an insurance premium — and also an instalment plan for self-disarmament.