The Semiconductor Foundry Industry Has a Map Problem
Here is a thing the semiconductor industry likes to say: TSMC holds more than 90% of advanced chip production. Here is another thing it likes to say: the foundry market is a three-horse race. Neither statement is wrong, exactly. Both are misleading in ways that matter enormously if you are making policy, placing a bet, or trying to understand why your iPhone chip comes from one building in Taiwan.
Yole Group published its annual foundry landscape report last month, and the analysts did something useful: they catalogued where the industry gets the numbers wrong. The result is less a story about one company's dominance than a story about how the industry's own vocabulary obscures its structural realities.
The first misconception is about nodes. "Advanced" is not a fixed point. Depending on who is speaking, it means anything below 20nm, anything below 10nm, or anything below 5nm. Yole settles on the most technically honest definition: focus on the 3-nm class in high-volume manufacturing. By that measure, TSMC holds roughly 95% of the market. Samsung has not consistently matched TSMC's yield at that node, and Intel's 18A is ramping slowly and largely serving internal products. The companies that needed leading-edge capacity in the past few years — Apple, NVIDIA, Google — had essentially one real option, and TSMC's pricing reflects that.
That near-monopoly at the leading edge is why the 2nm ramp matters. TSMC confirmed that volume production of N2 began in the fourth quarter of 2025. The capacity situation is not subtle: both 2026 and 2027 are reportedly sold out, with TSMC asking customers to define their production needs through at least the second quarter of 2027. Reservation lead times can stretch to six quarters — meaning a chip designer needs to commit to production roughly a year before wafers ship. The A16 variant, aimed at high-performance AI chips, enters volume production in the second half of 2026.
Samsung's position is more complicated than the standard narrative allows. The company lost Apple, NVIDIA, and Google because it could not stabilize yields at 3nm. That is a real failure with real consequences. But by late 2025, Samsung's 2nm yield had reportedly climbed to 55–60%, according to TrendForce citing Chosun Daily. The company is recovering. Samsung began shipping its Exynos 2600 smartphone chip — the first commercial product on its SF2 process — in the last quarter of 2025, with mass production starting earlier than some reports expected. The yield trajectory matters: it climbed from roughly 30% in the first quarter to 50% by fall, with a target of 60–70% by year-end. The price gap with TSMC is also significant. With TSMC's 2nm wafers reportedly priced about 50% higher than previous generations, Samsung's more flexible pricing has already attracted a Tesla contract for next-generation AI chips, along with orders for Apple image sensors and crypto-mining ASICs from Chinese firms. If yields hold, that is a real alternative for cost-sensitive designs.
The second misconception is about the size of the market. The foundry industry is projected to reach $402 billion in 2026, according to Yole Group's latest estimates. But the word "foundry" hides a split: roughly 53% of wafer manufacturing is captive — IDMs like Samsung, Intel, and Texas Instruments running fabs for their own products. The open foundry segment is the other 47%. TSMC holds about 72% of the open segment, but only about 34% of total manufacturing including captive fabs. Both numbers are accurate. Most coverage uses the open-foundry number. The aggregate number tells a different story about concentration, competition, and who actually makes the world's chips.
The third misconception is the "China is 10 to 15 years behind" framing. SMIC and Hua Hong are at 7nm, roughly equivalent to TSMC's 2019 capability. Huawei used that node to retake the number one position in China's 5G smartphone market in 2025 — with more silicon per phone, but no catastrophic performance gap for most users. The gap is real. It is not a clean 10-to-15-year lag across the entire ecosystem. And it exists partly because U.S. export controls have cut off Chinese fabs from ASML's extreme ultraviolet equipment, which TSMC and Samsung both rely on. The constraint is not purely technical.
The geographic mismatch is the point that should keep people up at night. U.S. companies account for 57% of global wafer demand but control only roughly 10% of foundry capacity locally. Taiwan controls 23% of global capacity and generates just 4% of demand. China held 21% of foundry capacity in 2024 while generating only 5% of global wafer demand. By 2030, China is on track to hit 30% of installed capacity, according to Yole. Foundry capacity is expected to grow at 4.3% per year from 2024 to 2030, with global utilization staying around 70% through the end of the decade — a relatively low figure that will become the new normal.
That is the structural picture: the country that designs the most chips has almost no fabs, the country that makes the most chips uses very little of what it makes, and the country building fabs fastest has almost no ability to produce the most advanced logic. No single region operates in isolation. ASML's Dutch machines, Tokyo Ohka's Japanese photoresist, Applied Materials' U.S. deposition tools — the leading edge is a global project. The question is who controls it.
The CHIPS Act was designed to change the U.S. numbers. TSMC signed a preliminary agreement for up to $6.6 billion in direct funding plus up to $5 billion in loans from the U.S. government, supporting a planned investment of more than $65 billion in three leading-edge fabs in Phoenix. Intel is building out. The timelines are long and the capital requirements are staggering — a new advanced fab can exceed $30 billion. The return on those investments is not obvious.
Pierre Cambou, the Yole analyst who has been tracking this for years, put it plainly: "The foundry market is more of a capitalistic game than a product competition." Ownership, location, and utilization now have to be read through national interests, economic security, and long-term technology strategy. That is the real framework. The market share numbers are just the scoreboard.
