TSMC's Real Invention Wasn't a Chip. It Was a Promise to Never Build One.

In the late 1980s, the smart money in chips lived by a single swaggering rule: build your own fabrication plant or you weren't a real semiconductor company. The line — 'real men have fabs' — belonged to Jerry Sanders, the CEO of AMD, and it captured the gospel of an entire industry.⁷ Owning the factory was the whole point of being powerful. Into that conviction walked a company in Hsinchu, Taiwan, that proposed the opposite: it would build the most expensive factories in the world and then promise, in writing and forever, to never design a single chip of its own. That company was TSMC. The swagger aged badly. AMD itself eventually went fabless.⁷

The official story of TSMC's dominance is that it simply out-engineered everyone — better yields, smaller transistors, the tightest process technology on earth. That's true, and it's also the wrong explanation. The engineering is the visible result; the cause sits underneath it, in a structural choice made on the day of incorporation. TSMC didn't win because it built the best fab. It won because of the one thing it promised never to build.

The promise that turned a factory into a moat

TSMC was incorporated on February 21, 1987, in Hsinchu, with the Taiwanese government's National Development Fund holding 48.3% and Philips 27.5%, on a single founding idea: it would be a pure-play foundry that pledged never to compete with its customers.¹ Morris Chang, recruited to Taiwan in 1985 to run the government's ITRI research institute, had conceived it there as a 'common fab' — a shared factory that would take any customer, do no design work of its own, and stay out of every product market its clients were in.⁶ Read coldly, that pledge looks like a limitation. It was actually the entire business model. A chip design is a company's crown jewel; handing it to a manufacturer that also designs chips means handing it to a future competitor. By promising to be only a manufacturer — never a rival — TSMC removed the single biggest reason a designer would refuse to outsource. The promise wasn't a constraint on the business. It was the product.

Why every customer's R&D becomes TSMC's switching cost

Here is the mechanism, worked down to where it actually bites. When a fabless company designs a chip, it does not design in the abstract — it designs against a specific foundry's process, its design rules, its libraries, its packaging. The customer pours its own R&D into making a product that works inside TSMC's particular factory. That investment is the trap, and it's self-laid. Switching to another foundry doesn't mean signing a new contract; it means re-engineering the chip, re-validating it, and re-tooling around a different process — at the customer's own expense, with the customer's own time, against a deadline a competitor isn't waiting on. So the more deeply a designer commits to TSMC, the more expensive it becomes to leave. The trust that the pure-play pledge created is what got the customer in the door; the co-engineering is what nails it to the floor. TSMC doesn't have to lock anyone in. The customer does it to themselves, one design cycle at a time.

Now watch that loop compound across decades. TSMC's early international customers — Intel, Motorola, Texas Instruments — outsourced older, trailing-edge work, not their crown jewels.⁵ The real prize was a generation of fabless startups that had no fabs at all. Broadcom, Marvell, Nvidia, Qualcomm: in Chang's own telling, they 'started with us when they were small.'⁵ They grew up designing against TSMC's process, so as they got large, their entire product architecture was already entangled with one manufacturer. Volume flowed in; that volume funded the next, more expensive process node; the better node attracted the next wave of designers. The flywheel runs on a promise — keep it, and the trust never has to be re-earned.

The numbers a near-monopoly looks like

By 2025 the loop had produced something close to a structural monopoly on the world's most advanced manufacturing. TSMC's revenue reached about US$122.54 billion for the year, up 36.1%, giving it 69.9% of the global foundry market — up from 64.4% the year before — while Samsung, the distant second, held 7.2%.⁴ In 2023 TSMC produced 28% of the entire world's non-memory semiconductor output value.² Concentration shows up inside the customer book too: in its 2025 SEC filing, the top ten customers accounted for 78% of net revenue, with the single largest at 19%.³ That's the signature of a business whose customers are deeply, expensively committed. And the leading edge is where the lock is tightest: advanced nodes of 7nm and below made up 74% of TSMC's wafer revenue in late 2025.⁸

The clearest proof of pricing power is the boldest move of all: in early 2026 TSMC notified customers of 5–10% price hikes across its most advanced sub-5nm nodes.⁸ A supplier raises prices on its most demanding products only when it knows the customer has nowhere else to go.

Isn't this just the best fab winning — and isn't it fragile?

The fair objection is that none of this requires a clever business model — TSMC simply makes better chips, and customers stay because the alternatives are worse. There's truth in that. Samsung's 3nm yields reportedly sat in the 30–40% range in 2025, too low to win large external orders, which means a designer who wanted to leave TSMC for the leading edge often had no credible place to go.⁸ But that argument actually strengthens the thesis rather than undercutting it. Technology leadership and the switching cost are the same flywheel seen from two angles: the volume that the pure-play trust attracts is what funds the next node, and the next node is what keeps the volume captive. Strip the model away and a rival could in principle out-engineer one factory. Leave it in place and the rival must out-engineer the factory and overcome years of a customer's own joint R&D and earn trust that TSMC never has to re-earn. A better fab is necessary. It was never sufficient.

The honest counter is that the position is not eternal, and it's not purely Chang's genius either. The lone-visionary story is too tidy: Chang was recruited by Taiwanese officials — Sun Yun-suan, Li Kuo-ting — into a state-backed semiconductor push, and the government held 48.3% of the company at founding.⁶¹ The pure-play structure was his strategic move; the runway under it was national policy. And concentration cuts both ways: a customer book where the top ten are 78% of revenue is also a book where the loss of a few relationships would hurt badly.³ Geography is the real fragility — the most indispensable factories on earth sit on one contested island. But none of that is a flaw in the model. It's the price of the model working too well.

TSMC's founders made the most counterintuitive promise in the industry's history: we will build the hardest, most expensive thing to build, and we will never use it to build a single thing of our own. The competitors who laughed — real men have fabs — were measuring power by who owned the factory. TSMC measured it by who could be trusted with it. Four decades later it manufactures the brains of the modern world and answers to almost no one for the price.⁸⁴ The genius was never the smallest transistor. It was choosing, on day one, to want nothing from its customers except their work — and to give them, in return, the one thing no rival with a design of its own could ever credibly offer: a manufacturer that would never become a competitor.