TSMC Won Chips by Building Nothing of Its Own. That Was the Only Move It Had.

Open an iPhone, an Nvidia AI accelerator, or an AMD server processor, and you will find a piece of silicon that the company selling it never manufactured. They designed it, marketed it, and put their name on it — but the actual atoms were laid down in Taiwan, by a company that designs nothing, sells no products of its own, and competes with none of its customers. TSMC built more than 9,920 products for about 465 customers in a single year³, and it has a name on exactly zero of them. That is not an accident of strategy. It is the strategy.

The official story is that Morris Chang was a visionary who foresaw the fabless revolution and built TSMC to enable it. That is a beautiful legend, and it is a post-hoc reconstruction. Chang wasn't trying to reinvent the industry. He built a foundry because the Taiwanese government wanted a semiconductor business, and Taiwan was weak in both design and marketing — so manufacturing-only was the only path he could see.⁴

The weakness that turned out to be the weapon

Every other chipmaker in 1987 was an IDM — an integrated device manufacturer that designed its own chips, made them in its own fabs, and sold them under its own brand. The whole stack, owned end to end. Taiwan couldn't play that game; it had no army of world-class chip designers and no consumer brands to sell against. So Chang did the only thing the constraint allowed: he ripped one slice out of the integrated stack — the manufacturing — and made it a business that did nothing else. TSMC opened in February 1987 with $220 million in capital, half of it from the government.⁴ It was less a bold bet on the future than a company built around what Taiwan happened to be able to do.

Here is the turn that makes the whole story work. By owning less of the stack, TSMC became the only player nobody had a reason to distrust. An IDM that offers to manufacture your chip is also your competitor — it sees your design, your volumes, your roadmap, and it sells products into the same markets you do. Why would you hand it your secrets? TSMC sold the one thing an integrated rival structurally cannot: neutrality. It competes with no one, so everyone can build with it. The geography did the suppressing of design ambition for them — and what looked like Taiwan's handicap became the entire product.

Its first customers weren't the startups it was supposed to save

The tidy version of the legend has TSMC born to serve the scrappy fabless startups it would go on to enable. It didn't happen that way. TSMC's first customers were the giants — Intel, Motorola, Texas Instruments — offloading older, out-of-date products they no longer wanted to run in their own fabs.⁴ The IDMs used TSMC as overflow capacity for the work they'd outgrown. Only after the foundry had proved it could deliver did the fabless designers arrive in force, knowing they could now build a billion-dollar chip business without ever pouring a fab. The model didn't create the demand; the demand discovered the model.

Why a thin slice compounds into an unassailable moat

A foundry that builds other people's designs sounds like a low-margin contract manufacturer. TSMC is the opposite, and the reason is the flywheel hiding inside Chang's learning-curve obsession. Pool the volume of hundreds of customers into one set of fabs, and you ride down the cost curve faster than any single IDM building only for itself ever could. That cost lead funds the next process node; the next node attracts the most demanding customers; their volume funds the node after that. In 2024 TSMC turned US$90.08 billion of revenue into US$36.52 billion of net income at a 56.1% gross margin² — and held 64% of the pure-play foundry market, against Samsung's 12%.⁶ The advanced nodes are where the gap becomes a chasm: chips at 7nm and below already account for 74% of TSMC's wafer revenue.⁸

Doesn't TSMC make 90% of the world's chips — and isn't that just luck?

Two honest objections. First, the dominance is often overstated: you'll read that TSMC makes 90% of the world's advanced silicon. It doesn't. Fold in DRAM and NAND and TSMC is roughly 12% of worldwide advanced-silicon capacity.⁷ The true claim is narrower and sharper — TSMC owns the most advanced logic nodes almost completely, which is exactly the silicon the AI boom needs most. Overclaiming the share actually obscures how concentrated and how defensible the real position is.

Second, the fair objection is that this was luck — a government push, a constraint, and good timing, dressed up afterward as genius. There's truth in it. Chang didn't foresee the fabless wave; he stumbled into a structure the future happened to reward. But notice what survived the luck: thirty-seven years of compounding trust and scale that no integrated rival has matched. Samsung has the fabs and the capital and still sits at 12%⁶, because it cannot offer the one thing it would have to give up its own chip business to sell — being nobody's competitor. The origin was contingent. The moat is not. And the proof is in the pricing power: TSMC can reportedly raise prices 5–10% across every advanced node, and Apple, Nvidia, AMD, Qualcomm, Broadcom, and MediaTek all pay it⁸, because there is nowhere else to go.

TSMC conquered chips by refusing to do almost everything a chip company does. It designs nothing, brands nothing, and competes with nobody — and that triple subtraction is the whole asset. Chang didn't set out to revolutionize the industry; he set out to build the one thing a small, design-poor island could build, and the constraint quietly handed him a position no fully integrated rival can ever copy without dismantling itself. The world's most advanced silicon is made by the company that wanted the least from each chip — and got, in exchange, the only seat every chip has to pass through.