TSMC currently manufactures nearly 70% of the world's contract-produced chips. For the most advanced nodes, its market share climbs to around 90%. These are the components powering artificial intelligence, iPhones, high-end graphics cards, and a growing proportion of global digital infrastructure. It represents one of the most spectacular industrial concentrations in modern history. How did Taiwan get here?

The title says "almost all", and that "almost" matters. Manufacturing a chip involves one of the most globalized value chains in existence, divided into three major segments. First, design: this is primarily the domain of American "fabless" companies that draw the blueprints. Nvidia, Apple, Qualcomm and AMD are among them. Second, manufacturing: etching that blueprint onto a silicon wafer is the job of the foundry, and this is where Taiwan reigns supreme. Finally, equipment: the most critical machine, which allows for the etching of the finest patterns, is produced by only one group in the world, the Dutch firm ASML.

TSMC's grip is less absolute in certain sectors, such as memory chips, analog components, sensors, or microcontrollers. But for the most advanced logic chips, the formula accurately summarizes the situation: the brain is often American, the hand is Taiwanese, and the tool is European. And today, it is the hand that forms the most spectacular bottleneck.

The Schism of 1987

Until the 1980s, designing and manufacturing chips went hand in hand. A phrase attributed to the founder of AMD summarized the industry's mindset: "real men have fabs." Morris Chang would overturn this dogma. A MIT and Stanford-educated engineer, he founded TSMC in 1987 with a simple but revolutionary idea: to create a pure-play foundry that would only manufacture other people's chips without ever selling its own.

The choice was bold but visionary. First, it reassured customers: a foundry that would never compete with them had less reason to plunder their intellectual property. Nvidia, Apple or Qualcomm could thus entrust it with their designs without fear of nurturing a future rival. It then gave birth to an entire industry: by removing the requirement to own a multi-billion dollar factory, the model paved the way for hundreds of chip designers. Taiwan had transformed its weakness into a weapon.

Four decades of dominance rest on barriers that reinforce one another. First, capital: a leading-edge fab costs over $20bn, requires high-precision processes, and becomes obsolete in a few years. Second, know-how: manufacturing yields only improve through years of repetition and cannot be bought off the shelf. Third, the ecosystem: the Hsinchu Science Park concentrates suppliers, laboratories, and tens of thousands of specialized engineers. Finally, trust: because it competes with no one and meets its deadlines, TSMC captures orders years in advance, finances its research and widens its lead. Samsung, Intel and other rivals currently remain far behind in advanced foundry services.

The Monopoly Behind the Monopoly

Why are hundreds of billions not enough to replicate this industry? Because one step further up the chain hides another monopoly. Etching the finest patterns requires EUV (Extreme Ultraviolet) lithography, and only one company in the world knows how to build these machines: ASML. Each one weighs dozens of tons, costs up to $400m, and is the result of 30 years of research. The group holds 100% of the EUV market. It is the "picks and shovels" strategy: whether Nvidia, AMD or another wins the AI race, their most advanced chips will pass through ASML and TSMC.

The bulk of the world's advanced production is thus physically located in Taiwan, making the island indispensable to the global economy. The Taiwanese speak of a "silicon shield": no major power can let the island fall without endangering its own industry. Yet Beijing claims this territory, posing a very concrete threat to global supply.

To measure what is at stake, one precedent suffices. The 2021 semiconductor shortage lasted only a few months and mainly affected less advanced components. Yet it reportedly cost approximately 1% of US GDP, or nearly $240bn. A prolonged disruption of the most advanced chips would be of an entirely different magnitude.

An Unbeatable Quality/Cost Combination

The debate over Western countries reclaiming production has long agitated power circles. Since 2022, the United States and Europe have been massively subsidizing the reshoring of part of the production. TSMC has committed to investing up to $165bn in Arizona, where a first plant is already producing for Apple and Nvidia.

But TSMC's founder himself is skeptical. In 2022, on the Brookings Institution and CSIS podcast "Vying for Talent", Morris Chang recalled a stubborn fact. TSMC had opened a plant in Oregon in 1997. 25 years later, it still cost about 50% more than in Taiwan, and the gap had never closed. The root cause, in his view, lies in talent: America allowed its production engineers to migrate towards design, then towards finance, starting in the 1980s. To him, subsidizing local manufacturing with tens of billions is a costly and largely futile exercise, at least as long as no war breaks out in the Taiwan Strait.

"I assume there will be no war. Frankly, if there were a war in the Taiwan Strait, I think the United States would have many more concerns than electronic chips," Chang noted at the time. This confidence has aged. Semiconductors now command artificial intelligence, data centers, and autonomous weaponry, from drones to guided missiles. Their interruption would no longer be mere collateral damage drowned in a broader conflict, but would constitute a first-order shock in its own right. Losing Taiwan's fabs would add a global economic earthquake to the military one: a problem within a problem.

Dependence is therefore not truly dissipating, despite the proliferation of projects elsewhere in the world. The most advanced technologies continue to be developed in Taiwan, while foreign sites often inherit previous generations or less critical capacities. Even by 2030, the chain will probably be slightly less concentrated, but far from truly diversified.

This situation offers three lessons for the investor. First, the strongest competitive moats are cumulative: it is the stacking of barriers, rather than a single one, that makes TSMC and ASML so difficult to catch. Second, in a gold rush like AI, infrastructure providers often capture more predictable value than the visible stars of the chain. Third, concentration is a risk as much as a strength: depending on a single site, a single equipment manufacturer, or a handful of clients creates a fragility that TSMC's own official reports acknowledge.