TSMC 1987: Inventing the Pure-Play Foundry

Situation

It is 1987, in Taiwan. Morris Chang, a veteran of the American semiconductor industry, is founding a company around an idea most insiders find backwards.

At the time, the great chip companies — Intel, Texas Instruments, Motorola — are vertically integrated (Integrated Device Manufacturers, or IDMs). Each both designs its chips and manufactures them in its own fabrication plants (fabs). Owning a fab is the non-negotiable price of entry — and it is staggering: billions in capital that only the largest firms can afford. This locks out a generation of brilliant chip designers who have ideas but no capital to build a fab.

Chang's idea is to split the industry in two. TSMC will be a "pure-play foundry": it will manufacture chips and nothing else. It will never design its own chips, never sell its own branded products, and never compete with its customers.

That last promise is the strategic heart of the whole model. A foundry that also had its own chip line would be a threat — why hand your secret design to a company that competes with you and might learn from or copy it? But a foundry with no design business of its own is a safe partner. By being the manufacturer that can never threaten them, TSMC can become the trusted factory for an entirely new species of company: the "fabless" designer who owns no fab and rents TSMC's. The decision is whether to bet on disaggregating one of the most capital-intensive industries on earth.

The decision moment

It is 1987. Chang must choose TSMC's model:

1. Be a pure-play foundry — manufacture only, never compete with customers. Build world-class fabs and offer them as a service to anyone who designs chips, guaranteeing TSMC will never enter the design business. Bet that this trust position unlocks a whole ecosystem of fabless designers who could never afford their own fabs — and that concentrating all their manufacturing demand into TSMC's fabs creates unmatched scale and process-learning advantages over time. Radical and unproven: it depends on an industry of fabless companies that barely exists yet.
2. Be a conventional IDM — design and manufacture your own chips. Do what Intel and TI do: build a fab and a design business, selling your own products. Proven and respected — but it pits a small Taiwanese startup directly against the world's strongest integrated giants on their own terms, and forfeits the foundry trust position.
3. Be a contract manufacturer that also dabbles in its own designs. Hedge — take foundry work and sell some of your own chips. Seemingly safer, but it poisons the trust that makes the pure-foundry model work: customers would never fully trust a foundry that competes with them, even a little. The hedge destroys the very thing that creates the advantage.

You are Morris Chang.

Key datapoints (for reference)

Frameworks invoked

• Business Model Innovation (disaggregation). The breakthrough wasn't a better fab — it was unbundling a vertically integrated industry into separate design and manufacturing layers. By specialising in only one layer, TSMC let designers specialise in the other, and the whole industry reorganised around the split. Disaggregating a vertical value chain can create more total value than any integrated player captured alone.
• Trust Positioning ("never compete with customers"). The promise not to have a design business is a strategic asset, not a limitation. It makes TSMC the only manufacturer a designer can fully trust with their crown-jewel IP. Deliberately forgoing a business (chip design) is what unlocks the far larger one (being everyone's foundry).
• Enabling Complementors. Rather than competing with chip designers, TSMC created and empowered them — every fabless startup that could now exist because of TSMC became a customer. Building an ecosystem of complementors who depend on you is a more powerful and durable position than fighting them for the same pie.
• Scale + Learning Curve. Because all the fabless companies' manufacturing flows into TSMC's fabs, TSMC runs more volume and more process variety than any single IDM — accelerating down the learning curve, funding ever-more-expensive leading-edge nodes, and compounding a process lead that eventually surpassed even Intel. Concentrated demand is the flywheel.

Discussion questions

1. The foundry model's power comes partly from a promise not to do something (never compete with customers). Why is deliberately forgoing the design business such a strong strategic move, and where else does "refusing to compete with your customers" create advantage?
2. TSMC disaggregated a vertically integrated industry. When is unbundling a value chain the winning move, and when does integration (doing it all yourself) win instead?
3. TSMC's advantage compounds because all fabless demand concentrates in its fabs. Why is that scale/learning-curve flywheel so hard for an integrated rival like Intel to counter?
4. The whole bet depended on a fabless industry that barely existed in 1987. How do you justify a strategy whose customers don't exist yet — and how is this like (or unlike) the SK Hynix HBM bet?
5. By enabling complementors instead of competing with them, TSMC made itself indispensable to Nvidia, Apple, Qualcomm, and others. What are the risks of a model where your success is tied to enabling companies that grow more powerful than you?

The real outcome (revealed at session end)

1987 onward: TSMC commits to the pure-play foundry model and the ironclad promise never to compete with its customers. Initially modest, the model proves transformative: it enables the entire fabless industry. Designers who could never have afforded a fab — the future Nvidia, Qualcomm, Broadcom, AMD, and eventually Apple's silicon team — can now exist, designing chips and letting TSMC build them.

The flywheel turns: Because all of that demand concentrates in TSMC's fabs, TSMC accumulates unmatched scale and process-learning, funding ever-more-advanced and ever-more-expensive leading-edge nodes. Over decades it surpasses even Intel in manufacturing process leadership. By the 2010s–2020s, TSMC makes the world's most advanced chips and becomes one of the most strategically critical companies on earth — the foundry on which Apple, Nvidia, and the AI boom all depend.

The lesson: TSMC won by splitting an industry rather than competing within it. The decisive insight was that refusing to compete with your customers — forgoing the chip-design business entirely — was not a sacrifice but the source of unbeatable trust, and that enabling complementors (the fabless designers) rather than fighting them would build a more powerful position than any integrated giant. Concentrating the whole industry's manufacturing into one specialist created a scale-and-learning flywheel that even Intel couldn't catch. Sometimes the strongest strategy is to define yourself by the lucrative business you won't enter — and to make everyone else's success run through your factory.

Sources

• Morris Chang interviews and TSMC corporate history.
• Chris Miller, Chip War: The Fight for the World's Most Critical Technology (2022).
• Analyses of the fabless/foundry industry structure and TSMC's rise vs Intel.
• TSMC annual reports and semiconductor-industry histories.