TSMC still the linchpin

Taiwan Semiconductor Manufacturing Co. sits under almost every serious AI chip story because the company is not just big. It is the place where the most advanced chips actually get made. TSMC says it is the technology leader in advanced logic, and industry trackers now put its share of the pure-play foundry market above 70%, powered by heavy demand for 3-nanometer and 5-nanometer chips used in AI systems. (tsmc.com) That matters because the AI boom has turned chipmaking into a capacity problem, not a design problem. Nvidia can design a new GPU. Broadcom can design a custom accelerator. Apple, AMD, Qualcomm, and the big cloud companies can all line up with their own plans. None of that changes the harder fact that only a tiny number of fabs on Earth can manufacture the leading-edge logic those chips require, and TSMC runs most of them. TrendForce reported in February that TSMC’s N2 process will enter mass production in 2026, with demand already tight from both mobile and high-performance computing customers. (trendforce.com) That is why the report that TSMC’s 2-nanometer capacity is effectively booked through 2026 landed so hard. The exact customer mix is still mostly rumor, but the shape of the story is clear. Apple is expected to take a large share early. High-performance chip companies are crowding in behind it. TSMC’s own roadmap says N2P, the next step in the 2-nanometer family, is already scheduled for volume production in the second half of 2026. In other words, customers are not fighting over a distant science project. They are fighting over real factory slots on a process node that is about to matter. (wccftech.com) The bottleneck is not only the wafer fab. It is also packaging. AI chips have become so large and memory-hungry that advanced packaging now decides how many systems can ship. TSMC’s CoWoS platform has become the key choke point, and industry reports say Nvidia locked up the largest share of that capacity as Blackwell ramps. Even when wafers start coming out of Arizona, the most advanced AI parts still depend on packaging capacity centered in Taiwan. (trendforce.com) That helps explain why TSMC’s US expansion keeps growing in the retelling. The official number is already enormous. In March 2025, TSMC announced it would add another $100 billion to its Arizona plan, bringing total US investment to $165 billion. The company says that buildout includes three fabs, two advanced packaging facilities, and an R&D center. Its Arizona site says the first fab is already producing 4-nanometer chips, while the third fab broke ground in April 2025 and is slated for N2 and A16 technologies. (pr.tsmc.com) The rumor now is much bigger than the official plan. Posts and reports circulating through the market describe a possible long-run US footprint closer to $500 billion and as many as 12 fabs in Arizona. That is not what TSMC has formally announced. What it has said, clearly and on the record, is that Arizona is becoming a “gigafab” cluster and that the company is accelerating parts of its US roadmap to meet customer demand. The rumor feels believable because the smaller version already happened. The number moved from $65 billion to $165 billion in one step. (pr.tsmc.com) Investors still flinch at the idea because every part of the expansion cuts both ways. More US capacity lowers geopolitical exposure to Taiwan. It also raises costs. Bloomberg reported in July 2025 that AMD’s Lisa Su said chips from TSMC’s Arizona facilities cost more than chips from similar fabs in Taiwan, by more than 5% and less than 20%. TSMC is not moving into a cheaper production base. It is building a second one because the world’s appetite for advanced compute has become too large, and too strategic, to leave on one island. (bloomberg.com)