MBUS 873 — Session 5

US/China Rivalry Part II: Geopolitics of High Technology

Queen's Smith AMBA 2026 · Prof. David Detomasi · Participation-Ready Prep
Semiconductor Value Chain TSMC & the Silicon Shield CHIPS Act vs. Made in China 2025 Export Controls Taiwan, Semiconductors, and a "New Cold War"? (HBS 9-722-035)
Block 1 — The Lens: Technology as National Security

Why This Session Treats a Chip Factory Like an Oil Field

AMD founder Jerry Sanders opens the intellectual frame of this case in a single 1979 line, quoted on the case's cover page: "Semiconductor processing technology is today's crude oil.... The strength of all industrialized nations will largely depend on semiconductors by the end of this century." Session 4 treated critical minerals as a geopolitical lever because they sit at the base of every supply chain that matters. This session applies the identical logic one rung up the value chain: whoever controls the manufacturing chokepoint for a general-purpose technology — the chips inside every phone, missile, data center, and AI model — accumulates leverage disproportionate to that chokepoint's share of global GDP. Semiconductors generated $550 billion in global revenue in 2021, a rounding error next to the $100+ trillion world economy. But the case's entire argument is that dollar share understates strategic weight: a modern economy, a modern military, and a modern AI program simply do not function without leading-edge logic chips.

Strategic Leverage = f(Chokepoint Concentration), not f(Market Size)
A segment's geopolitical weight is set by how hard it is to route around — not by its share of industry revenue. That is why the front-end fabrication stage, not the whole $550B industry, is the fulcrum of this case.

The Semiconductor Value Chain: Four Segments, Four Different Geographies

The case (Exhibit 10) breaks the value chain into four segments, each with a different geography, capital intensity, and defensibility — which is exactly why the chain fractures so cleanly along geopolitical lines:

1

Pre-Competitive Research & EDA/Core IP

The software that designs chips and the foundational research behind them. Three US firms — Synopsys, Cadence, and Mentor — hold a combined 70% of the design-software market, making this a quiet American chokepoint.

Concentrated: United States
2

Chip Design (Fabless)

Firms like Nvidia, Qualcomm, AMD, and Apple design chips but own no factories — the "fabless" model TSMC's founding made viable. High value-add, low capital intensity, highly portable.

Concentrated: United States
3

Front-End Manufacturing (Foundry / Fab)

Fabricating the physical chip. A cutting-edge fab cost ~$20 billion and took two years to build as of 2020. By 2020, TSMC alone held 50% of global foundry capacity and the "lion's share" of leading-edge capacity — the case's central chokepoint.

Concentrated: Taiwan
4

Back-End Manufacturing (Assembly, Packaging, Testing)

Lower-value, more labor-intensive finishing steps. Historically the segment Taiwan itself entered via — dispersed across Taiwan, mainland China, and Southeast Asia; far easier to relocate than a fab.

Dispersed: Taiwan / China / SE Asia
The chokepoint is segment 3, not the industry. Design (segment 2) is portable — a fabless firm can headquarter anywhere with engineering talent. Back-end assembly (segment 4) is low-barrier and has already diffused across Asia. Front-end fabrication at the leading edge, by contrast, requires a single fab to cost $20 billion, take two years, and run at 99% yield across roughly 700 discrete process steps — a natural-monopoly economics that has concentrated almost the entire world's leading-edge output on one 13,826-square-mile island. That concentration, not the industry's dollar size, is why Taiwan sits "at the forefront of tectonic shifts in geopolitics and geoeconomics," in the case's own words.
$550B
Global semiconductor industry revenue, 2021
50%
TSMC's share of global foundry capacity, 2020
$20B / 2 yrs
Cost and build time for one leading-edge fab, as of 2020
12% → 14%
BCG's estimate of the US global production share shift from $50B in CHIPS subsidies

This is also the session's bridge from Session 4: critical minerals (rare earths, copper, cobalt) are the geological input; semiconductors are the manufactured chokepoint one step downstream, and China's own hand in that upstream layer — 35% of known rare-earth reserves, over 60% of production, and more than 85% of global refining capacity, per the case — means Beijing holds a countervailing lever even as it plays catch-up in chips themselves.

Block 2 — Case Analysis: Taiwan, Semiconductors, and a "New Cold War"? (HBS 9-722-035)
Case Summary

The case is built around one man's improbable bet and one company's improbable dominance. Morris Chang left a 25-year career at Texas Instruments in 1983 feeling "put out to pasture," only to be summoned back to Taiwan in 1985 by ROC Premier Sun Yun-suan to run the Industrial Technology Research Institute (ITRI). Chang's insight was structural: fab costs were doubling roughly every five years, reaching $250 million by 1985, while a small cohort of "fabless" design firms was emerging with no capacity of its own. In 1987 he spun ITRI's Electronics Research and Service Organization into a "pure-play" foundry — Taiwan Semiconductor Manufacturing Company (TSMC) — that would manufacture chips designed by others rather than compete in design or marketing. Over the next three decades TSMC became the world's most advanced semiconductor manufacturer, contradicting AMD CEO Jerry Sanders's 1992 boast that "real men have fabs" so thoroughly that AMD itself divested its foundries in 2009. By 2020, TSMC held 50% of global foundry capacity and effectively the entire market for the most advanced process nodes.

That concentration is now a geopolitical fact both superpowers are racing to undo. Washington's 2022 CHIPS and Science Act authorized $278 billion in federal funding, more than $52 billion of it for domestic semiconductor manufacturing; Beijing's National IC Plan (2014) and Made in China 2025 initiative set a target of 70% semiconductor self-sufficiency within a decade, backed by a $150 billion state-investment target — forty times China's previous investment in the industry. Yet as of the case's writing, Chinese-owned firms still produced only 5.9% of all semiconductors used in China, and Exhibit 14's count of firms worldwide capable of manufacturing below 10 nanometers shows almost none headquartered in China. Both superpowers, in other words, are spending unprecedented sums to reduce their dependence on an island neither fully controls — and that effort is precisely what has made Taiwan's semiconductor industry a live strategic asset rather than a purely commercial one.

That asset has a name in the case: the "silicon shield" — the idea, voiced by President Tsai Ing-wen and TSMC chairman Mark Liu alike, that Taiwan's centrality to the global chip supply makes an invasion too costly for China, and for the world, to tolerate. Liu told 60 Minutes that "the world all needs Taiwan's high-tech industry" and "will not let war happen in this region." The case opens in autumn 2021 with Chinese warplane incursions at record levels and Taiwan's defense minister warning China could mount an invasion "at the lowest possible cost" by 2025, and closes in winter 2024 with newly elected President Lai Ching-te repeating the shield's logic on the world stage. The case's central, unresolved tension is whether that shield still holds — or whether the very policies (CHIPS, Made in China 2025) both powers are pursuing to reduce their Taiwan dependence are quietly dismantling the deterrent they claim to fear losing.

Why This Case Belongs in Sessions 4–5

Session 4's Zambia case showed a smaller, weaker state navigating great-power competition over a resource it could not fully control. Taiwan is the same structural position at a vastly higher level of stakes: a 23.5-million-person democracy whose formal statehood is contested by its largest trading partner, sitting on top of the single most consequential manufacturing chokepoint in the modern economy. Where Zambia had copper and cobalt as bargaining chips, Taiwan has TSMC — and where Zambia's leverage was largely passive (its minerals are in the ground regardless of policy), Taiwan's leverage is the product of forty years of deliberate, state-orchestrated industrial policy that this session asks you to evaluate directly.

Block 3 — Government Policy Effects: US, Taiwan, China (Case Question 1)

Three States, Three Industrial-Policy Models, Three Very Different Results

The case is, among other things, a natural experiment in industrial policy — the same technology, pursued by three governments with radically different tools, time horizons, and starting conditions.

United States

Defense-Seeded, Then Reactive

  • The industry itself is a Pentagon creation: military demand funded transistor R&D at Bell Labs and financed Fairchild Semiconductor's founding (three days before Sputnik). A DOJ antitrust consent decree forced AT&T to license its patents openly, seeding a uniquely dynamic, IP-diffusing ecosystem.
  • 1980s Japan competition: SEMATECH consortium (1987), $200M/year, half DARPA-funded — GAO called it a success; critics said it missed the PC/microprocessor wave entirely and mainly helped Intel.
  • CHIPS Act (2022): $278B total, $52B+ for manufacturing, 25% investment tax credit; conditions include no stock buybacks and no new advanced-node investment in China.
  • Morris Chang called it an "expensive exercise in futility," citing TSMC's own failed 1990s US fab attempts — "cost problems," "people problems," "cultural problems." TSMC's Arizona build cost ~4x the Taiwan equivalent.
Verdict: highly effective at seeding the original industry (1950s–80s); much weaker as a reshoring tool today — capital cannot manufacture Taiwan's engineering culture or build-out speed on a subsidy timeline.
Taiwan

Patient, Coordinated, Decisive

  • Land reform (tenancy 64%→10%, 1949–73) freed capital and labor into export-oriented industrialization; the world's first export processing zone opened in 1966.
  • Manpower Development Plan pushed vocational enrollment from 40% to 60% (1965–75); by 1990, 30% of university students studied engineering (vs. 5% in India, 20% in South Korea).
  • ITRI (1973) and ERSO (1974) transferred a seven-micron process from RCA and trained 40 engineers; ERSO spun off UMC (1980) and then TSMC (1987) under Morris Chang's direct state mandate.
  • Hsinchu Science-Based Industry Park (1980), modeled on Silicon Valley, offered government equity stakes up to 49%, tax exemptions, and low-interest loans.
Verdict: the clear success case — decades of concentrated, patient state-industry coordination built one purpose-designed firm rather than diffusing capital broadly, and that firm now holds 50% of global foundry capacity.
China

Massive Capital, Partial Results

  • Three prior decades of state-owned semiconductor investment (1980s–2000s) produced facilities so dependent on foreign tech and talent that domestic spillovers were "all but nonexistent," per the case.
  • National IC Plan (2014) targeted $150B — 40x prior investment — funding national champion SMIC; Made in China 2025 (2015) set a 70% self-sufficiency target within a decade.
  • Tools included forced joint-venture technology transfer (25 firms surveyed), aggressive outbound M&A ($35B in 2015 alone, until Western CFIUS-style screening began blocking deals like Aixtron in 2016), and outright IP theft (SMIC's founder recruited 100+ former TSMC staff; TSMC sued and won 9% of SMIC plus $325M).
  • SMIC hit a 7-nanometer breakthrough in 2020 without EUV lithography — enough to power Huawei's 2023 Mate 60 — but analysts call it "years behind" TSMC, Intel, and Samsung.
Verdict: mixed — real, faster-than-expected progress at the margins, but Chinese-owned firms still produce only 5.9% of semiconductors used domestically versus a 70% target, proving capital alone cannot compress a multi-decade manufacturing learning curve.
The pattern across all three: the policy that worked best (Taiwan's) was the narrowest and most patient — one government, one small set of state-linked institutions (ITRI/ERSO/Hsinchu), one firm built over 15 years before it dominated anything. The policies struggling hardest (the US's reshoring push, China's self-sufficiency drive) are both trying to compress that multi-decade build-out into a single presidential term or five-year plan, using capital as a substitute for time. The case's own data (BCG: $50B only moves US share from 12% to 14%; China at 5.9% domestic production against a 70% target) suggests that substitution does not work cleanly.
Block 4 — The Silicon Shield: Protection or Vulnerability? (Case Question 2)

Does TSMC Protect Taiwan, or Paint a Target on It?

The case frames this as its central open question, and it deserves to be argued both ways before landing on a position.

The Case for Protection

  • TSMC chairman Mark Liu, on 60 Minutes: "the world all needs Taiwan's high-tech industry" and will therefore "not let war happen in this region" — mutual economic dependence as deterrent.
  • President Lai Ching-te, on winning the 2024 election: "the semiconductor industry in Taiwan is a common asset of the world... we hope China and the international community will cherish this industry as well."
  • An invasion would destroy the very asset China would be fighting over, and would trigger catastrophic, self-inflicted damage to China's own tech-dependent economy — raising the cost of conflict for Beijing specifically, not just for outside powers.
  • The shield gives the US, EU, Japan, and South Korea an economic — not just values-based — reason to back Taiwan's defense, widening the coalition that would respond to an invasion.

The Case for Vulnerability

  • Taiwan's defense minister, Chiu Kuo-cheng, warned in 2021 that China would be able to invade "while keeping the cost and losses to the lowest possible level" by 2025.
  • China maintains a life-size replica of Taiwan's Presidential Office in Inner Mongolia to drill for a "decapitation strike," and the Global Times reported 2019 civilian-ship transport drills explicitly linked to "a potential landing mission... on the island of Taiwan."
  • Xi Jinping, 2019: reunification "should not be passed down from one generation to the next," and while he invoked "peaceful reunification" 18 times in one speech, he also stated "we do not renounce the use of force."
  • Pentagon wargames simulating a Chinese invasion of Taiwan have the "red team" (China) regularly beating the "blue team" (the US) — the shield's deterrent value depends on a political response speed the case gives no evidence will actually materialize.
  • Taiwan's own defense posture is underweighted for the threat: conscription fell to just four months by 2021 ("strawberry soldiers"), only ~80% of budgeted military positions were filled, and procurement still favors prestige platforms (66 F-16s for $8.1B in 2020) over the "porcupine strategy" of asymmetric anti-ship and anti-tank defenses some analysts recommend.
  • Crucially: the shield's very existence is what's driving Washington and Beijing to spend hundreds of billions of dollars building around Taiwan — which shrinks the shield's own deterrent value over time.

My Position: Real Protection Today, With a Shrinking Half-Life

The silicon shield is not a myth — TSMC's centrality genuinely raises the cost of conflict for China, for the US, and for every economy dependent on advanced chips, and that cost is real leverage Taiwan should not surrender lightly. But the shield's logic contains the seed of its own erosion: because Taiwan's concentration is precisely what threatens both superpowers, both are now rationally investing to reduce their exposure to it. Every CHIPS-funded fab in Arizona and Ohio, every additional percentage point SMIC claws toward self-sufficiency, is a small withdrawal from the account the shield draws on. The shield does not disappear overnight — Exhibit 14's chipmaker count shows the leading edge will remain Taiwan-and-Korea-concentrated for years — but its deterrent value is a depreciating asset, not a permanent guarantee.

Recommended navigation strategy for Taiwan: a dual-track approach. First, do not accelerate offshoring the leading edge — keeping TSMC's most advanced nodes physically on the island preserves what remains of the shield's deterrent value, and Morris Chang's own skepticism about US fab execution suggests the leading edge will stay Taiwan-anchored regardless for the medium term. Second, treat the shield as a asset to be spent down deliberately rather than relied upon passively: use TSMC's continued centrality now to lock in deeper defense commitments, embed Taiwan further into allied coordination mechanisms like the Chip 4 forum, and fund the "porcupine strategy" asymmetric defenses that would matter regardless of whether the shield holds. Taiwan's best strategy treats the shield as a clock, not a wall.
Block 5 — Deglobalization Implications (Case Question 3)

My Position: Yes, But It's Bifurcation in Tech, Not a General Retreat From Trade

The case gives Morris Chang — the architect of the most globalized, most specialized supply chain in existence — the closing word on this question, and it is a stark one: "globalization is almost dead and free trade is almost dead." I largely agree, with one important qualifier: what the evidence supports is not deglobalization across the whole economy, but a deliberate bifurcation of the technology stack specifically, into US-aligned and China-aligned blocs.

Evidence for Bifurcation

  • Legislated decoupling: CHIPS Act recipients are contractually barred from expanding advanced-node capacity in China — an industrial subsidy with an explicit decoupling clause attached.
  • Export controls with extraterritorial reach: the Biden administration's October 2022 controls blocked Chinese access to advanced semiconductors, manufacturing equipment, and even the intellectual property behind them, extended via the Foreign Direct Product Rule to reach a Dutch firm's (ASML's) EUV lithography machines. One industry analyst called the package "a declaration of full-scale technological economic cold war."
  • Allied bloc formation: the Netherlands and Japan joined US export controls in January 2023; the "Chip 4" forum (US, Taiwan, South Korea, Japan) formalizes coordination among a "trusted" manufacturing bloc.
  • China's mirror-image response: Made in China 2025's self-sufficiency drive plus China's own leverage over rare earths (35% of reserves, 60%+ of production, 85%+ of refining capacity) — both sides are building redundant, parallel capability rather than deepening the single integrated network.

The Counterargument — and Why It Doesn't Fully Hold

The case also supplies real evidence that chokepoint controls leak. Despite the 2022 controls, Chinese imports of semiconductors and manufacturing equipment reached near-record levels in 2023; Nvidia simply redesigned its most advanced chips to fall just below the regulatory threshold; and the Russia precedent is instructive — Russian semiconductor imports fell more than 50% in the first half of 2022 after sanctions, only to rebound and exceed 2021 levels by year-end, mostly rerouted through China and Hong Kong. BCG's own estimate that $50 billion in CHIPS subsidies shifts US global production share only from 12% to 14% suggests the "reshoring" story is more a hedging exercise than a genuine unwind of the existing network. I take this seriously — full deglobalization (a total severing of the US-China tech relationship) looks unlikely — but partial, deliberate bifurcation at the leading edge is already happening and accelerating, even if trade in mature-node chips and everything else continues largely as before.

Consequences of Bifurcation

Cost

Duplicative Capital

Every additional geography building a $20B, two-year fab raises the industry's cost base without adding net global capacity — TSMC's Arizona build alone cost roughly 4x the Taiwan equivalent.

Redundancy

Resilience Over Efficiency

"Trusted" supply chains (Chip 4) sacrifice the specialization gains of letting Taiwan do what it does best, cheaper, for everyone — trading efficiency for redundancy on both sides of the split.

Innovation Diffusion

Slower on Both Sides

Cutting China off from EUV and design software slows its innovation curve (SMIC remains "years behind" without EUV) — but US firms also lose a China market that historically supplied 20–30% of their revenue, a mutual drag on R&D funding.

The overlooked consequence: bifurcation doesn't just reshape US-China trade — it forces every third country to choose a compute stack it had no hand in designing. This is the thread I pick up directly in Block 7's Taju's Edge — a bifurcated tech world lands hardest on emerging-market builders who depend on affordable, undifferentiated global chip and cloud access, and who now face a choice that is increasingly not theirs to make.
Block 6 — Official Case Discussion Questions
Q1. Assess the effects of government policies on the development of the semiconductor industry in the United States, Taiwan, and China.
Three very different models, three very different results. The US industry was defense-seeded (Bell Labs, Pentagon-funded transistor R&D, the Fairchild founding tied to Sputnik-era defense spending) and grew inside an open-IP ecosystem forced by AT&T's antitrust consent decree — genuinely effective in the 1950s–80s, but the 2022 CHIPS Act's $278B is proving a much weaker reshoring tool: BCG estimates it moves US global production share only from 12% to 14%, and Morris Chang calls it "an expensive exercise in futility" given TSMC's own failed 1990s US fab attempts. Taiwan's model — patient, narrow, state-coordinated (ITRI, ERSO, Hsinchu Park, Morris Chang's direct 1985 mandate) — is the clear success, producing a firm that now holds 50% of global foundry capacity. China's National IC Plan and Made in China 2025 mobilized unprecedented capital ($150B target, 40x prior investment) and produced real but partial results — SMIC's 2020 7nm breakthrough powered Huawei's 2023 Mate 60, but Chinese-owned firms still produce only 5.9% of domestic semiconductor consumption against a 70% self-sufficiency target. See Block 3 for the full comparison.
Q2. How should Taiwan navigate the rise of Sino-American tensions? Is its semiconductor industry a source of protection or vulnerability?
Both, and the balance is shifting. TSMC's centrality is real protection today — Mark Liu's and Lai Ching-te's "silicon shield" logic genuinely raises the cost of conflict for China and widens the coalition (US, EU, Japan, Korea) with an economic stake in Taiwan's defense. But it is simultaneously the single highest-value target in the rivalry, and the very policies both superpowers are pursuing to reduce Taiwan-dependence (CHIPS Act, Made in China 2025) are structurally eroding the shield's own deterrent value over time — a self-defeating dynamic the case never resolves. My recommendation: Taiwan should not accelerate offshoring its leading-edge capacity (that would remove the remaining deterrent for no defensive gain), while treating the shield as a depreciating asset to be spent deliberately now — deepening allied coordination (Chip 4), locking in defense commitments, and funding "porcupine strategy" asymmetric defenses rather than relying on the shield as a permanent guarantee. See Block 4 for the full argument.
Q3. Does the competition over Taiwan and the global semiconductor industry augur a world of deglobalization? If so, what likely consequences might deglobalization bring?
Yes, but as targeted bifurcation of the technology stack rather than a general retreat from global trade — Morris Chang's own "globalization is almost dead" verdict, backed by CHIPS Act decoupling clauses, extraterritorial export controls, and allied bloc formation (Chip 4, the Netherlands and Japan joining US controls). The counter-evidence — near-record 2023 Chinese chip imports, Nvidia's threshold-skirting redesigns, the Russia sanctions-evasion precedent — shows controls leak and full severance is unlikely, but partial, accelerating bifurcation at the leading edge is already underway. Consequences: duplicative capital costs (TSMC's Arizona fab at ~4x Taiwan's cost), resilience purchased at the expense of specialization efficiency, and slower innovation diffusion on both sides of the split. The consequence most often missed: a bifurcated tech world forces every other economy to choose a compute stack it did not build — hitting emerging-market technology builders hardest. See Block 5 for the full argument.
Block 7 — Participation Hooks

Consensus Point

The class will quickly agree that TSMC's dominance is real, structurally hard to replicate on a subsidy timeline, and now central to both US and Chinese strategy. Don't spend airtime re-litigating that — the more interesting question is what each side does about it, not whether it's true.

Provocative Push

Challenge the room: is the silicon shield's own logic self-defeating? If Taiwan's centrality is the reason Washington and Beijing are spending hundreds of billions of dollars building around it, then talking up the shield as leverage may be accelerating its own expiration — every dollar of CHIPS or Made in China 2025 spending is a small bet against the shield's premise. The shield may be most valuable in the years right now, while it's being actively dismantled by the powers who fear it most.

Taju's Edge — Supply Chain Concentration as a Product Risk

Think like a product operator, not a political scientist: TSMC's foundry model is a single point of failure the entire global tech industry rationally chose to accept for thirty years, because splitting fabless design (Nvidia, Apple, Qualcomm) from foundry (TSMC) let every fabless firm skip $20B capex and two-year build cycles. That's the same chokepoint economics I see one layer up in AI infrastructure today — nearly every AI product, including work I've done in African tech and edtech, ultimately depends on a small number of GPU and cloud chokepoints. The concentration that makes the industry efficient is exactly what makes it fragile precisely when geopolitics compounds it.

Taju's Edge — What Bifurcation Means for Emerging-Market Builders

If the tech world splits into a US-aligned and a China-aligned compute stack, African and other emerging-market tech companies don't get to build a third stack — they inherit whichever bloc's chips, cloud, and AI models are cheapest and most accessible in their market, and increasingly that choice is not theirs to make. This is directly relevant to how affordable-compute-dependent African products source infrastructure, and it will only get more consequential as export controls expand from semiconductors into AI model access itself — a preview of Session 7's Nvidia case.

Block 8 — Where This Connects in the Course
Session 1 — Singapore

Singapore's pivot toward biomedical sciences and a "Global-Asia" tech hub (Biopolis, A*STAR) is a smaller-scale version of the same state-directed tech-industrial playbook that built TSMC — a resource-poor state manufacturing competitiveness in a strategic technology sector through deliberate, patient institution-building rather than market forces alone.

Session 4 — Zambia / Critical Minerals

Zambia's copper and cobalt sit one layer further upstream in the exact same supply chain this case sits atop — the rare earths, copper, and critical minerals debate is the raw-material precursor to the chip fabrication chokepoint this session examines, and China's leverage in both layers (mineral supply and, increasingly, mature-node chips) is not a coincidence.

Session 7 — Nvidia

This session's export-control theme continues directly into Session 7's Nvidia case — the same October 2022 controls that cut Chinese AI firms off from Nvidia GPUs (95% of China's AI chip market) become the central strategic dilemma when examined from a single company's perspective rather than a national-policy one.