TSMC

Taiwan Semiconductor Manufacturing Company is the world's largest contract chip manufacturer, a "pure-play foundry" that builds the chips others design while selling none of its own. Founded by Morris Chang in 1987, it inverted an industry that had always designed and manufactured under one roof. The through-line across its stories: by choosing to make only what others could not credibly promise, neutrality, TSMC built a position the integrated incumbents could not copy without destroying themselves, then compounded it into a cornered leading-edge resource.

Counter-Positioning: the model Intel could not copy

The problem. When Chang founded TSMC in 1987, the industry was vertically integrated: Intel, Texas Instruments, and Motorola each designed and built their own chips as integrated device manufacturers (IDMs). These IDMs did contract work only with spare capacity, and a fabless startup handing over its design risked being copied or out-competed by the very fab building its product. There was no trusted, dedicated manufacturer a designer could rely on.

The approach. TSMC committed to a pure-play foundry model: it would manufacture for customers and never design or sell chips of its own, promising to "never compete with customers." Renouncing the high-margin design business was the entire point, because it made the trust that IDMs structurally could not offer into TSMC's product.

How it solved it. An IDM like Intel could not credibly become a neutral foundry, since trusting Intel with your design meant trusting the company trying to beat you, and Intel's fat product margins made the lower-margin foundry business look unattractive for decades. TSMC adopted a model incumbents rationally refused to copy because copying it threatened both their product margins and their customers' trust. Nearly 40 years on, that refusal underpins about NT$2,894 billion (~$90B) in 2024 revenue, up 33.9% year over year.

Moats: outspending everyone on every node

The problem. Counter-positioning bought TSMC an opening, but a neutral foundry is only valuable if it stays at the technological frontier, and leading-edge fabrication is staggeringly capital-intensive. No single chip designer, however large, can justify the tens of billions per year required to keep advancing process nodes, and any rival foundry that falls behind loses the customers who need the newest transistors.

The approach. By aggregating manufacturing demand from the entire fabless industry, TSMC could spend more on each new process node than any individual customer or narrower rival ever could. Its 2025 capital expenditure guidance ran $40 billion to $42 billion, with roughly 70% directed at advanced process technology.

How it solved it. That spending advantage compounds: more volume funds better process technology, which wins more customers, which funds still more volume. In 2024, chips at 7nm and below made TSMC 69% of its revenue and its 3nm node alone 18.3%, while its pure-play foundry share held above 64%. Leading-edge fabrication is now a cornered resource, and replicating it requires not just capital but decades of process know-how that cannot simply be bought.

Platform & Ecosystem: making fabless design possible, then orchestrating it

The problem. Removing the need to own a fab made "fabless" chip companies viable, but a designer still cannot ship silicon from raw manufacturing alone. They need design tools, verified IP blocks, and process design kits tuned to each specific node, and coordinating that across dozens of EDA vendors and IP suppliers was a barrier that could stall a customer's tape-out for months.

The approach. In 2008, when 40nm was the most advanced node, Chang launched the Open Innovation Platform (OIP), later folded into TSMC's "Grand Alliance." It knits customers together with EDA partners, IP providers, design-center and cloud partners, all aligned to TSMC's technology roadmap and its qualified process design kits.

How it solved it. The ecosystem grew to more than 40,000 IP titles, over 2,600 process design kits, and 100-plus partners orbiting TSMC's roadmap. That platform is why the fabless generation exists at scale: Nvidia, Qualcomm, Broadcom, AMD, and Apple Silicon all design chips that only exist because someone else owns and coordinates the factory and its tooling.

Switching Costs: designed into a node only TSMC operates

The problem. A foundry customer does not buy a generic commodity. It co-designs a chip against a specific process node, using that node's design rules, PDKs, and qualified IP, then invests millions and many months validating the layout. For a leading-edge product, walking away would mean re-engineering the design for a different manufacturer's process from scratch.

The approach. TSMC's approach makes that lock-in structural rather than contractual: it stays years ahead at the bleeding edge (3nm in volume, 2nm following), so a customer's most advanced designs are committed to a process where no equivalent alternative exists. The OIP ecosystem deepens this, since a design built on TSMC-qualified IP and PDKs is not portable without redoing the work.

How it solved it. With 7nm-and-below representing 69% of 2024 revenue and TSMC holding the overwhelming majority of sub-7nm capacity, customers needing the newest transistors have effectively nowhere to switch to. Samsung and a struggling Intel trail at the leading edge, so for an Nvidia AI GPU or an Apple flagship SoC, re-tapeout elsewhere is not merely costly, it means a worse chip on a slower node.