The Foundry Nobody Pitches: Selling When You Are Not the Default Choice

A hardware team can spend two years building a chip that actually works and still fail, not because the design is wrong, but because no fab will put it into production on a timeline the company can survive. Taiwan's funding conversation rarely gets that far. It stops at whether TSMC will take the call.

The number that decides who ships

TSMC's Q4 2025 earnings call put a number on where the industry's attention sits. Nodes at 7 nanometers and below accounted for 77% of that quarter's wafer revenue: 28% from 3nm, 35% from 5nm, 14% from 7nm. That is not a surprise. It is the business TSMC is built to run, and it is the business every Taiwanese pitch deck implicitly assumes when it says we are talking to TSMC.

The part that gets skipped is what happens to everything that is not that business. On the same call, CFO Wendell Huang laid out how TSMC plans to spend its US$52 to 56 billion 2026 capital budget: roughly 60% to 80% into advanced process technologies, 10% to 20% into advanced packaging, testing and mask production, and about 10% into specialty technologies. Ten percent of the world's largest foundry's capital plan is the entire budget line for RF, power, automotive and sensor processes, the categories most hardware startups actually need.

The whole specialty pure-play is smaller than one quarter of the leader

Put the two side by side and the scale gap becomes concrete. TSMC's full-year 2025 revenue was about US$122.4 billion. GlobalFoundries, the world's third-largest pure-play foundry and one of the few sizable players built specifically around specialty and mature-node manufacturing, closed FY2025 at roughly US$6.79 billion. That is about 5.5% of TSMC's year, and smaller than TSMC's single Q4 2025 quarter of US$33.7 billion. This is not a story about one company losing to another. The two do not compete for the same die. It is a story about the size of the market that exists once you take the advanced-node conversation out of the room.

The growth is not where the story says it is

The specialty side is usually described as automotive and IoT. The segment that actually grew in 2025 was communications infrastructure and datacenter, up 32% year on year.

This changes which question is diagnostic.

A different question, not a smaller answer

TrendForce's tracking of the top ten mature-process foundries points at the same divide from the demand side. Nodes at 28nm and above ran under 80% capacity utilization through 2025, a moderate recovery rather than a boom, while 5/4nm and 3nm ran at full utilization on AI server and flagship smartphone demand. Two markets, two demand curves, and almost none of the capital story crosses between them.

For a hardware startup, which side of that line its chip sits on is decided early, usually before the term sheet, by the physics of what the product actually does. RF front ends, power management ICs, automotive-grade sensors and most edge silicon are not waiting for a cheaper version of an advanced node. They run on FD-SOI, SiGe or RF-SOI because those processes are the right physical fit, not a fallback. A team that has already scoped which process family its die needs, and lined up a fab relationship for it, has done a piece of diligence that a slide about a roadmap to TSMC does not replace.

What this means for anyone underwriting a hardware deal

Three things follow. First, the fab question belongs in the first diligence call, not the technical annex. Ask which process node the die actually needs and why, before asking about market size. Second, do not read a mention of TSMC as a maturity signal on its own. For a large share of hardware categories, the correct answer to the process question is a name most Taiwanese investors have never had to look up. Third, treat specialty foundry relationships the way you already treat channel partnerships: a form of go-to-market risk that sits outside the product roadmap slide, and one worth asking about directly.

There is a version of this problem that has nothing to do with semiconductors. The first meeting is about working out whether they want a second source, or a lever to move the first one.

Where EvoScale reads this

Operator-led investing means the earliest useful question is often the most boring one. For hardware and deep-tech deals, that question is which process family the die needs and whether the founding team has already had that conversation with a fab, not whether the deck can name TSMC. A team that has mapped its production path onto a real process node rather than an aspirational one has already removed the part of the risk a spreadsheet cannot see. That is the difference between reading a roadmap slide and having sat on the other side of the PDK.

The category is not glamorous and it will not trend on a fundraising deck. It is the line that decides which chip reaches volume and which one stays a very expensive prototype.

Sources: TSMC Q4 2025 earnings call.