Foundry Market Share 2026: TSMC Holds 67%, Samsung Slips to 8% — Why the Gap Keeps Widening

What kind of business is a foundry, and why does it concentrate around one company?

A foundry is a company specializing in contract semiconductor manufacturing. As explained in how the chip industry split into design-only and manufacturing-only companies (in Japanese), it takes circuit designs from fabless companies such as NVIDIA and Apple and mass-produces their chips across tens of thousands of wafers. Because a pure-play foundry has no products of its own, it never competes with its customers — and by serving many customers at once, it keeps its equipment utilization high.

Why the market concentrates in so few hands is equally clear. The cost of building capacity for the most advanced nodes (2nm and 3nm, the frontier of miniaturization) balloons by tens of billions of dollars with each generation. TSMC's 2026 capital expenditure is expected to reach US$52–56 billion, and only a company that can secure anchor customers large enough to justify that outlay can advance to the next generation. The result: TSMC is effectively the only company operating at the leading edge, and its share has approached 70%.

What does the global share picture look like as of Q4 2024?

According to TrendForce, global foundry market share in the fourth quarter of 2024 breaks down as follows.

The top five account for roughly 90% of the world market — and TSMC alone far exceeds the combined share of positions two through five. It is an extreme concentration structure.

Nothing has changed in 2026. TSMC's Q2 2026 revenue hit a record US$40.2 billion (+33.7% YoY), lifted by the first-ever revenue contribution from leading-edge 2nm. HPC (AI and data center) accounted for 66% of revenue, surging 20% quarter over quarter.

Why can TSMC sustain this level of dominance?

The core of TSMC's staying power is yield — the share of good chips — on leading-edge processes. As covered in our explainer on semiconductor yield (in Japanese), yield determines how many good chips come off each wafer and therefore directly sets manufacturing cost.

TSMC's 2nm (N2) generation, which entered volume production at two Taiwan fabs in December 2025, uses gate-all-around (GAA) transistors and is reported to yield 80–90%. That is the source of its overwhelming cost advantage. N2 capacity for 2026 is effectively sold out to major customers including Apple, NVIDIA, Qualcomm and AMD. See also our deep dive on TSMC's 2nm ramp (in Japanese).

The other barrier to entry is trust built on zero competition with customers. TSMC designs and sells no chips of its own. NVIDIA can hand its next-generation GPU crown jewels to TSMC without fear the information leaks to AMD. That trust has compounded over more than 30 years, and it is structurally difficult for a latecomer to replicate quickly.

Why did Samsung's foundry business stall?

Samsung's foundry unit (under its DS division) posted losses in the trillions of won in 2024, and the market had hoped for a recovery in 2026. But in June 2026, foundry chief Han Jin-man said annual profitability was realistically a 2028 story and that a recovery next year would not be easy.

The root of the gap is yield on leading-edge 2nm (the SF2 family). Reports put Samsung's 2nm yield at roughly 55% as of April 2026, with some accounts of improvement past 60% in the first quarter of 2026. But against TSMC's 80–90%, that is still a 25–35 point deficit — and the deficit translates directly into fewer good chips per wafer.

That yield gap drove customers away. Qualcomm shifting Snapdragon production from Samsung to TSMC is one example.

Samsung's Taylor, Texas fab (Fab 1) plans to finish building out 2nm capacity within 2026, but full customer production has slipped to early 2027.

On the other side of the ledger, Samsung has secured a US$16.5 billion contract (through 2033) to manufacture Tesla's AI6 chips — the single biggest revenue pillar in its path to profitability by 2028.

How is Intel trying to catch TSMC?

Intel carried over US$13 billion in operating losses in its foundry business across 2023–2024, yet keeps pursuing external customers with Intel 18A as its weapon.

As detailed in our deep dive on the Intel 18A process (in Japanese), 18A is the industry's first process to combine a GAA structure (RibbonFET) with backside power delivery (PowerVia). Production began on the Oregon development line in November 2025, and in June 2026 Intel announced the transition to the higher-performance 18A-P. The follow-on 14A and 14A-E are slated for full operation in 2027–2028 and will be the first to use High-NA EUV lithography tools.

Intel's biggest win is Google. Reports indicate Google will order over 3 million TPUs by 2028 — potentially about half of a projected 6 million units heading to Intel. NVIDIA is also said to be running MPW evaluations on 18A for a four-die integrated design of its next-generation GPU (Feynman).

For Intel Foundry to reach profitability, it must lift external-customer revenue substantially. 2026–2027 is the credibility-building window: if large orders keep landing, the competitive structure could genuinely shift.

What does Japan's Rapidus bring to the race?

Japan's Rapidus is developing a 2nm process in Chitose, Hokkaido, targeting mass production within fiscal 2027. It builds on IBM's 2nm technology, co-developed with Belgium's imec, and cumulative Japanese government support reached 2.354 trillion yen as of April 2026.

In June 2026 it signed a memorandum of understanding with the UK Semiconductor Centre (UKSC), creating a framework to offer UK companies access to 2nm prototyping and production from the second half of 2027.

Rapidus's planned capacity remains far smaller than TSMC's, Samsung's or Intel's. But from a geopolitical risk-diversification standpoint, it can expect real interest from customers uneasy about concentration in Taiwan — particularly defense, telecom and automotive players across Japan, the US and Europe. Foundry competition ultimately hinges on technology and accumulated production track record, and Rapidus is likely to become a meaningful presence only from 2028–2030 onward.

Key takeaways

• Q4 2024 global foundry share: TSMC 67.1%, Samsung 8.1%, SMIC 5.5%. TSMC alone exceeds everyone else combined, and the structure persists in 2026.
• TSMC's moat is 2nm yield (80–90%) and non-compete trust with customers. 2026 capacity is effectively sold out, with annual capex of US$52–56 billion.
• Samsung's 2nm yield of roughly 55% trails TSMC badly. Foundry profitability is targeted for 2028, anchored by the US$16.5 billion Tesla contract.
• Intel is counterattacking with 18A, built around a Google order of over 3 million TPUs. Customer wins in 2026–2027 are the watershed.
• Japan's Rapidus, backed by 2.35 trillion yen in cumulative state support, targets 2027 production. Its scale is orders of magnitude smaller, but it is positioned as the geopolitical diversification play.