In the spring of 1987, a 55-year-old engineer named Morris Chang walked into a converted industrial plot in Hsinchu, Taiwan, and staked the future of the entire global electronics industry on a single, counterintuitive idea: that the companies designing chips and the companies manufacturing them should be completely separate. Until that moment, every major semiconductor firm — Intel, Texas Instruments, Motorola — designed and built its own chips in-house. Chang believed the future belonged to a company that built chips for everyone and owned nothing else. He called his new firm Taiwan Semiconductor Manufacturing Company. Nobody outside Taiwan paid much attention. Thirty-seven years later, TSMC controls 72 per cent of the world's pure-play foundry market, manufactures over 90 per cent of the world's most advanced logic chips, and has become the single most geopolitically consequential factory complex in human history. Apple's iPhones run on it. Nvidia's AI accelerators run on it. The F-35 fighter jet's avionics run on it. If TSMC stopped operating tomorrow, the global economy would begin seizing within weeks — not months, not years. Weeks. This is not an exaggeration. This is arithmetic.
The Invisible Empire — Part 2
The Microchip Chokepoint: How a 36,000 km² Island Controls the Nervous System of the Modern World
Where the World's Chips Are Born: The Hsinchu Campus
TSMC Fabrication Campus, Hsinchu Science Park, Taiwan
The buildings inside this compound produce more than 90% of the world's most advanced semiconductors. A single fab costs upward of $20 billion to construct and takes three to five years to become operational — making its concentration in one island one of the most severe single-point failures in the global economy.
What Is a Semiconductor, and Why Does It Rule Everything
Before you can understand why Taiwan's chip industry is a geopolitical weapon, you need to understand what a semiconductor actually does — and why making a good one is so extraordinarily difficult that only a handful of places on earth have ever managed it at commercial scale.
A semiconductor is a material — usually silicon, derived from ordinary sand — that conducts electricity under some conditions and blocks it under others. This switching behaviour is the physical basis of all digital logic. A transistor, the fundamental component of every chip, is essentially a tiny switch: current flows through it or it doesn't, representing the binary states 0 and 1 that all computing reduces to. The history of semiconductors is, at its core, the history of miniaturisation — fitting more transistors into the same space, making each switch faster and cheaper to operate.
In 1971, Intel's first commercial microprocessor, the 4004, contained 2,300 transistors. In 2024, TSMC's most advanced 2-nanometre chips pack approximately 100 billion transistors into a die roughly the size of a postage stamp. Those transistors are each about 2 nanometres across — roughly the diameter of ten hydrogen atoms placed side by side. To manufacture them requires a precision that defies ordinary physical intuition. The tolerances involved are not mechanical. They are quantum mechanical.
This extraordinary difficulty is the first reason why chip manufacturing concentrated in so few places. It is not a coincidence that Taiwan dominates. It is the result of four decades of deliberate government investment, engineering culture, talent accumulation, and — crucially — a business model that nobody else had the courage or clarity to replicate at the moment it mattered.
The Morris Chang Bet: How One Decision Built a Monopoly
When Morris Chang founded TSMC in 1987, the conventional wisdom in the semiconductor industry was that integration was strength. Companies like Intel, AMD, and Texas Instruments controlled their entire value chains — from chip design to fabrication to packaging. Chang's insight was precisely the opposite: that specialisation, at sufficient scale, would create advantages that integrated companies could never match.
His pure-play foundry model — take orders, build chips, own nothing proprietary — had three critical effects over the following decades. First, it enabled the rise of fabless chip design companies: Qualcomm, AMD, Apple, Nvidia, Broadcom, and others who could focus entirely on design without maintaining ruinously expensive fabs of their own. Second, it gave TSMC a volume advantage that no integrated competitor could match — because it built chips for all of them simultaneously, it could invest in process technology at a scale that single-brand fabs could not justify. Third, and most consequentially, it made TSMC indispensable to its customers in a way that created a self-reinforcing lock-in: if you are Apple, you cannot easily switch chip manufacturers mid-product-cycle. The investment in co-designing your chip architecture with TSMC's specific process nodes is enormous. Leaving means starting over.
By 2024, TSMC deployed 288 distinct process technologies and manufactured 11,878 different products for 522 customers — a figure that captures the extraordinary breadth of its dependency network. North America accounted for 70 per cent of TSMC's net revenue in 2024, according to the company's annual report filed with the U.S. SEC. The company's two largest customers — Apple and Nvidia — represent two of the most valuable companies on earth. TSMC is the factory that makes their most critical products possible.
The 90% Number: What It Actually Means
The statistic that appears most consistently in serious analysis of Taiwan's semiconductor position is this: Taiwan produces over 90 per cent of the world's most advanced logic chips — those manufactured at nodes of 7 nanometres and below. This figure, confirmed by the U.S. Trade and Development Agency, the Stimson Center, and multiple semiconductor industry bodies, requires some unpacking to appreciate fully.
Not all chips require the most advanced nodes. The chip controlling the windshield wipers in your car is probably manufactured at 40nm or larger. The chip in your microwave is simpler still. These chips are produced in many countries — South Korea, Japan, Germany, the United States, China. Disrupting Taiwan would not immediately eliminate these products from global supply.
What it would eliminate — completely, with no short-term substitute — are the chips that power everything at the frontier of technology. Nvidia's H100 GPU, the hardware backbone of the AI revolution, requires TSMC's advanced process nodes. Apple's A-series and M-series chips, which power every iPhone and MacBook, are manufactured exclusively at TSMC's facilities in Taiwan. AMD's EPYC server processors, which run a significant fraction of the world's data centres, come from the same fabs. Google's custom TPU chips, used for AI training and inference. Amazon's Graviton processors. The list continues without interruption.
In every modern automobile — not just electric vehicles, but conventional petrol and diesel cars — there are between 1,000 and 3,500 individual semiconductors. The pandemic-era chip shortage of 2020–2022, caused by nothing more dramatic than demand forecasting errors and a few factory shutdowns, produced a global automotive production shortfall of millions of vehicles and cost the industry an estimated $210 billion in lost revenue. That was a disruption affecting a fraction of semiconductor supply. A Taiwan shutdown would be categorically different in magnitude.
The $200 Million Machine That No One Else Can Build
Inside an EUV Lithography Machine — The Tool That Makes Advanced Chips Possible
Each ASML EUV machine weighs approximately 180 tonnes, contains over 100,000 individual components, requires 40 shipping containers to deliver, and costs upward of $200 million. Only ASML — based in Veldhoven, Netherlands — manufactures them. Without this machine, no chip below 7 nanometres can be produced at commercial scale anywhere on earth.
Alt text: Macro close-up of EUV lithography machine interior showing precision mirror arrays and wafer optics in purple and gold lighting
The Machine That Makes the Machine: ASML's Unique Stranglehold
If TSMC is the world's most consequential factory, ASML — headquartered in Veldhoven, a modest city in the southern Netherlands — is the company that makes TSMC's most critical tools. Without understanding ASML, you cannot fully grasp the semiconductor chokepoint, because TSMC's dominance is itself contingent on a machine that only one company on earth can build.
The machine is called an Extreme Ultraviolet (EUV) lithography system. Lithography is the process by which circuit patterns are printed onto silicon wafers — the equivalent of a photographic process, but operating at scales measured in nanometres. Conventional deep ultraviolet (DUV) lithography systems use light wavelengths of around 193 nanometres to print these patterns. At chip nodes below 7nm, the patterns are smaller than the wavelength of the light being used — which creates a fundamental physical problem that multi-patterning techniques can partially address but not eliminate. EUV lithography uses light at a wavelength of 13.5 nanometres — roughly 14 times shorter — which allows patterns to be printed at the atomic scale required for today's most advanced chips.
Generating EUV light is not a simple engineering challenge. It involves firing a high-powered laser at a droplet of molten tin, 50,000 times per second, turning the tin plasma into an extreme ultraviolet light source. The entire process happens inside a vacuum chamber because EUV light is absorbed by virtually all matter, including air. The machine that orchestrates this contains over 100,000 components, weighs approximately 180 tonnes, requires 40 shipping containers to deliver in pieces, and must be assembled on-site by ASML technicians. Each machine costs upward of $200 million. ASML currently produces around 60 of them per year.
Canon and Nikon — ASML's Japanese competitors in lithography — poured billions of dollars into developing EUV systems and failed. The technical barriers were simply too high. ASML holds roughly 90 per cent of the EUV market, with no meaningful competition in sight. Its nearest potential competitor is SMEE in China, which has built a reported EUV prototype but which semiconductor analysts estimate won't reach commercial viability until 2030 at the earliest — and that timeline assumes no technical setbacks. ASML has never shipped a single EUV system to a Chinese customer.
The chain of dependency runs like this: advanced chips require EUV lithography → EUV lithography requires ASML machines → ASML machines require components and expertise concentrated almost entirely in the Netherlands, Germany, and the United States. This supply chain, mapped fully, passes through perhaps six to eight countries in a sequence where the failure of any single link stops production of the world's most advanced chips entirely. This is not a supply chain. It is a chokepoint system with multiple locks on the same door.
The Export Control War: Technology as Sanction
The United States recognised the strategic value of this chokepoint well before most governments did. In October 2022, the Biden administration introduced the most sweeping export controls on semiconductor technology in American history — restricting the sale of advanced chips, chip manufacturing equipment, and associated software to China. The controls targeted not just U.S. companies but any foreign company using U.S. technology in its supply chain, a jurisdictional reach that brought TSMC, ASML, and Japanese equipment manufacturers like Tokyo Electron and Shin-Etsu within Washington's regulatory perimeter.
The Netherlands, under sustained U.S. pressure, followed with its own export controls in 2023, requiring ASML to apply for licenses before shipping even its older DUV immersion systems to Chinese customers. In January 2024, ASML's export license for several already-sold machines was partially revoked by the Dutch government, forcing the company to halt planned shipments. ASML has never shipped a single EUV machine to China — and under current rules, it cannot.
The result is a technology war conducted entirely through paperwork. No shots fired. No sanctions in the traditional sense. Just export control lists, license applications, and regulatory guidance documents — and the effect has been to partially wall off China from the most advanced tier of semiconductor manufacturing. China's self-reported chip self-sufficiency stands at roughly 16 per cent, against an official target of 70 per cent by 2025. The gap between ambition and reality is a direct consequence of ASML and TSMC export controls.
But here is the uncomfortable counter-fact: ASML sold 70 per cent of its DUV lithography systems to Chinese customers in 2024, according to CNAS research published in December 2025. Chinese firms have been stockpiling older machines — the ones still permitted under current controls — before those too get restricted. Those older machines, used with a technique called multi-patterning, can produce chips at nodes that previously required EUV. The loophole is significant, and it is being actively exploited.
The Taiwan Strait: Where Geography Meets Catastrophe
Taiwan sits 160 kilometres off the southeastern coast of mainland China, separated by the Taiwan Strait — a stretch of water that has been one of the world's most militarily tense zones since 1949. The People's Republic of China claims Taiwan as a breakaway province whose unification with the mainland is, in Xi Jinping's formulation, a "historical inevitability." Taiwan's democratically elected government disagrees. The United States maintains a policy of strategic ambiguity — neither recognising Taiwan as independent nor accepting China's territorial claim — while supplying Taiwan with weapons and maintaining unofficial diplomatic relations.
In 2025, the U.S. approved a $2 billion arms deal with Taiwan that included air defence systems, and budgeted $500 million for HIMARS rocket systems, tactical missiles, and anti-ship batteries. In March 2025, Taiwan received the first of a planned 66 F-16 Block 70 fighter jets. The military build-up on both sides of the strait is accelerating by every measurable indicator.
The RAND Corporation posed the scenario starkly: if China seized Taiwan, one of two outcomes would follow for its chip factories. Either China controls TSMC — and with it, 90 per cent of the world's advanced chip manufacturing capacity — or the factories are destroyed in the conflict. RAND's Jason Matheny described the second scenario as potentially producing "an economic crisis the likes of which we have not seen since the Great Depression."
The Institute for Economics and Peace estimated that a full-scale Chinese invasion of Taiwan could reduce global economic output by up to 2.8 per cent. On a global GDP base of roughly $105 trillion, that is nearly $3 trillion in lost output — and that figure likely understates the cascading effects. Chip shortages do not only affect consumer electronics. They hit automotive production, medical devices, telecommunications infrastructure, military systems, and agricultural equipment. Every sector with a control system of any kind takes a hit.
Consider this specific dependency: Apple's most advanced chips — the ones powering every iPhone 16 — can be manufactured in a single building within TSMC's Taiwan campus. Not a metaphor for concentration. A literal description of the physical address of a product used by 1.5 billion people worldwide. One building. One address on one island in one contested strait.
The Most Dangerous 160 Kilometres of Water on Earth
The Semiconductor Supply Chain Web — East Asia's Most Dangerous Geography
Every major consumer electronics product sold anywhere on earth passes through a supply chain node located within this map — most of them within 200 kilometres of the Taiwan Strait. A conflict in this stretch of water would not be a regional war. It would be a global supply chain event with no historical precedent in the modern era.
China's Paradox: The Country That Cannot Afford to Win
Here is the strategic paradox that makes the Taiwan situation genuinely unlike any other territorial dispute in the modern world. China is simultaneously the country most motivated to seize Taiwan — because doing so would solve its semiconductor dependency problem in one stroke — and the country most likely to destroy its own economy in attempting it.
China's trade with Taiwan totalled $224 billion in 2023. Chinese entities had over $100 billion invested in Taiwan at that time. China imports over $400 billion worth of semiconductors annually and, despite a decade of aggressive domestic investment, has achieved only about 16 per cent self-sufficiency in chip production. An invasion would trigger immediate international sanctions targeting China's financial system, technology exports, and key industries — and would simultaneously cut China off from the Taiwanese chips it relies on to manufacture the consumer electronics, electric vehicles, and solar panels that drive its export economy.
Vietnam would lose 14 per cent of its export value from a China-Taiwan conflict, according to estimates compiled by the Indo-Pacific Defense Forum. Cambodia would lose 13 per cent. The supply chain damage would radiate outward across Southeast Asia in ways that directly affect China's own Belt and Road investments and regional economic relationships. China would not simply be conquering an island. It would be detonating the supply chain that sustains its own economy — and the economies of every country it considers an ally or trading partner.
This mutual dependence — sometimes called the silicon shield theory of Taiwan's deterrence — argues that Taiwan's semiconductor dominance itself provides a form of protection. The logic: any country contemplating invasion must account for the economic damage it would inflict on itself, on the global economy, and on its own diplomatic relationships. The silicon shield is not a military deterrent. It is an economic one. It works by making the cost of aggression so diffuse, so globally distributed, that even China cannot absorb it without catastrophic self-harm.
The theory has limits. The Stimson Center noted in November 2025 that Taiwan's government, mindful of this dynamic, actually prohibits the overseas production of its most advanced chips — ensuring that only Taiwan itself can manufacture the leading-edge nodes that provide the shield. The rule is deliberate: the moment TSMC can fully replicate its most advanced production in Arizona or Japan, the silicon shield weakens. Taiwan's government is managing the dispersal of its strategic asset very carefully, and this caution is, paradoxically, in Taiwan's security interest.
The CHIPS Act and the Limits of Geographic Diversification
The Biden administration recognised the single-point-failure problem and passed the CHIPS and Science Act in August 2022, allocating $52.9 billion in subsidies to rebuild U.S. semiconductor manufacturing. TSMC responded with a commitment to build facilities in Arizona — initially announced as a $12 billion investment, later expanded dramatically. By early 2025, TSMC had committed $100 billion to build five additional chip facilities in the United States over the coming years, one of the largest manufacturing investments in U.S. history.
Japan secured TSMC investment with heavy government subsidies, establishing an $8.6 billion joint venture with Sony and Denso in Kumamoto producing 22–28nm chips. A second, more advanced Japanese fab is under discussion. The European Chips Act allocated €15.9 billion through 2030 to build European semiconductor capacity, with Intel planning fabs in Magdeburg, Germany.
These are real investments, and they will meaningfully reduce some categories of chip dependency. But they carry a caveat that deserves plain statement: the Arizona fabs will not reach 2nm production until 2028 at the earliest, and by then, TSMC in Taiwan will already be manufacturing at the next node. The overseas plants will always be a generation behind. This is partly logistics and partly the deliberate policy of Taiwan's government, which prohibits TSMC from replicating its leading-edge processes abroad.
There is a second limit to geographic diversification that rarely gets discussed: it is not just about factories. It is about people. The Truman National Security Project published an analysis in September 2025 arguing that Taiwan's semiconductor workforce — its engineers, process designers, and yield specialists — is itself a strategic asset as significant as the silicon. TSMC's extraordinary production efficiency is not encoded in its equipment alone. It is encoded in the accumulated knowledge of tens of thousands of engineers who have spent careers refining specific processes at specific fabs. An invasion that displaces or coerces this workforce could destroy capabilities that would take decades to rebuild even with functioning equipment in place.
AI, the GPU Crisis, and Why the Stakes Keep Rising
Everything described above was already true before the AI revolution began accelerating in 2022. The arrival of large language models, generative AI systems, and the GPU arms race between Microsoft, Google, Amazon, and Meta has added an entirely new layer of urgency to the semiconductor chokepoint — because AI models require chips at a scale and at specifications that only TSMC's most advanced nodes can deliver.
Nvidia's H100 GPU — the chip powering ChatGPT's training infrastructure, Google's Gemini, Meta's Llama, and virtually every major AI model in production — is manufactured exclusively at TSMC. Its successors, the H200 and the Blackwell B100 series, run on TSMC's most advanced nodes. The demand for these chips was so intense that Nvidia's order backlog extended over a year at peak AI investment, with customers offering premiums for earlier delivery slots.
The U.S. export controls on AI chips — restricting the sale of Nvidia H100s and equivalent hardware to China — are a direct application of the semiconductor chokepoint as a foreign policy instrument. By controlling TSMC's customer list and restricting advanced chip exports, Washington is attempting to maintain an AI compute gap between itself and China. The logic runs: countries that cannot access frontier AI hardware cannot train frontier AI models, and countries that cannot train frontier AI models fall behind in the technologies — autonomous weapons, intelligence analysis, economic planning, cybersecurity — that will define strategic power over the coming decades.
This is silicon geopolitics in its most direct form. A chip is not just a commercial product. It is a unit of strategic capability. Controlling who can manufacture it and who can buy it is, functionally, an act of power projection — executed not through aircraft carriers or blockades, but through export control lists and licensing regimes administered by the U.S. Department of Commerce.
China's Counter-Move: Huawei, SMIC, and the Domestic Chip Race
China has not accepted this constraint passively. In August 2023, Huawei released the Mate 60 Pro smartphone — and quietly revealed it contained a 7nm chip manufactured domestically by SMIC, China's largest semiconductor foundry. This shocked Western analysts who had estimated China was years away from 7nm production without EUV machines. SMIC had achieved it using multi-patterning techniques on older DUV equipment — precisely the loophole CNAS flagged in December 2025 as the most dangerous gap in current export controls.
The achievement was real, but its limits were equally real. SMIC's 7nm yield rates — the proportion of chips on a wafer that pass quality standards — are far lower than TSMC's equivalent process, making the chips significantly more expensive per unit. Scaling to the volumes required to compete commercially is a different problem than demonstrating the technical capability in a laboratory or limited production run.
China has also reportedly built an EUV machine prototype, covered by Reuters in late 2025, but analysts place commercial viability at 2030 at the earliest, assuming no technical setbacks. In the meantime, the technology gap between TSMC's 2nm production and China's best domestic capability remains measured in years and multiple process generations. That gap is the most important strategic distance in the world right now — more consequential, in day-to-day global affairs, than any military front line.
The Feynman Question: What Happens If Taiwan Stops for Six Months?
Richard Feynman had a technique for understanding complex systems: pick a specific, bounded scenario and trace its consequences rigorously, without flinching from where the logic leads. Apply it here. TSMC stops production for six months. Not destroyed — just halted. What happens?
In the first month: companies with existing chip inventories continue operating. Nvidia ships from stock. Apple uses existing wafer stockpiles. The disruption is not yet visible to consumers, and markets have not yet fully priced the risk. In months two and three: automotive manufacturers begin announcing production cuts. The smartphone industry starts rationing chip allocations across product lines. New AI data centre deployments slow as GPU procurement becomes impossible. Technology sector valuations drop sharply. In months four through six: new car production falls by 30 to 60 per cent at major manufacturers. Smartphone launches are delayed indefinitely. Cloud computing capacity expansion halts. The economic disruption crosses $1 trillion in annualised output losses. Hospitals and telecom providers begin prioritising chip allocations for critical systems. Consumer electronics prices spike by 40 to 80 per cent where supply remains available at all.
That is a six-month halt. A permanent destruction of Taiwan's fabrication capacity would be categorically worse. The capacity cannot be replicated quickly. Building a comparable fab takes three to five years and $20 billion, and that assumes the equipment is available, the workforce exists, and the institutional knowledge survives. None of these assumptions holds after a conflict that destroys infrastructure and displaces engineers who took decades to develop their expertise.
Silicon Is the New Oil — With One Crucial Difference
The comparison between semiconductors and oil as strategic resources is now a cliché — but it is an accurate one, with an important qualification. Oil is extracted from the ground. Its geographic concentration — in the Middle East, Russia, Venezuela — is a geological accident. The dependency it creates is essentially physical: you need the substance itself. When alternative energy sources develop sufficiently, the dependency diminishes.
Taiwan's semiconductor dominance was created by human decisions. By Morris Chang's business model in 1987. By Taiwan's government investment in Hsinchu Science Park. By four decades of engineering talent accumulating in a specific industrial ecosystem. And unlike oil, which is consumed when burned, chips are manufacturing inputs whose precision requirements actually intensify over time. The more advanced the technology becomes — AI, quantum computing, autonomous systems — the more it depends on the most advanced chips, and the more concentrated that production stays.
Oil can eventually be substituted with solar panels and batteries. There is no substitute for a 2nm logic chip. You cannot manufacture it with a less advanced process. You cannot make it from a different material. You must have the specific equipment, the specific process knowledge, and the specific workforce that can operate both — and those things exist, at commercial scale, in one country. In 2024, Taiwan's semiconductor industry generated over $165 billion in revenue, representing approximately 20.7 per cent of the country's entire GDP. That one industry, in that one country, is the physical substrate on which the modern technological world runs.
Morris Chang built a company that the world cannot afford to lose. He built it in a place that an authoritarian neighbour claims as its own territory. He built it so well, and so far ahead of any competitor, that the world had no incentive to build a backup — until the backup became a matter of survival. The global semiconductor industry spent forty years optimising for efficiency and cost. It is now discovering what it sacrificed in the process. Rebuilding that resilience — if it can be rebuilt — will take another twenty years and hundreds of billions of dollars. In the meantime, the chokepoint remains. The Taiwan Strait remains. And somewhere in a nondescript building in Hsinchu, a machine is printing circuits at the width of two hydrogen atoms, and the entire connected world depends on it continuing to do so, uninterrupted, indefinitely.
References
- The Motley Fool — "Taiwan Semiconductor Controls 72% of the Global Chip Market" — March 2026.
- Stimson Center — "Why Taiwan Fears 'America First' Risks Eroding Its Silicon Shield" — November 2025.
- CNAS — "The Export Control Loophole Fueling China's Chip Production" — December 2025.
- Talos Network — "Etching Out Influence: ASML as a Strategic Asset in EU-US Relations" — September 2025.
- Council on Foreign Relations — "Will China's Reliance on Taiwanese Chips Prevent a War?" — July 2023.
- U.S. International Trade Administration — "Taiwan: Semiconductors including Chip Design for AI" — December 2025.
- CSIS — "Taiwan's Semiconductor Dominance: Implications for Cross-Strait Relations" — March 2022.
- Truman National Security Project — "Saving Taiwan's Silicon Scientists" — September 2025.
- Efficio Consulting — "China-Taiwan Tensions: Impacts on Global Supply Chains and Semiconductor Availability" — July 2024.
Disclaimer: While artificial intelligence is utilized for preliminary research, every post on Decoding Curiosity undergoes significant manual editing to ensure intellectual honesty, factual accuracy, and a purely human perspective. We rely strictly on verifiable facts.
Comments
Post a Comment