TSMC ramps 2nm and packaging

Taiwan Semiconductor Manufacturing Co. said its 2-nanometer chip capacity will grow 70% a year through 2028 as it speeds up expansion for artificial intelligence demand. (focustaiwan.tw) A nanometer node is a chipmaking generation: smaller features let more transistors fit into the same space, which can improve speed and power use. TSMC said five fabs will begin volume production of 2nm chips in 2026, with two in Hsinchu and three in Kaohsiung. (focustaiwan.tw) (taipeitimes.com) TSMC said output in the first year of 2nm production, which began in the fourth quarter of 2025, should be 45% higher than first-year 3nm output in 2023. The company also said 3nm capacity is still growing at about 25% a year through 2027. (taipeitimes.com) (focustaiwan.tw) The bottleneck for many AI chips is no longer only the logic die. Advanced packaging is the step that stacks or links several chips together so they work like one larger processor. (reuters.com) TSMC said its Chip on Wafer on Substrate, or CoWoS, packaging capacity is growing more than 80% a year from 2022 through 2027. It said System on Integrated Chips, or SoIC, is growing more than 90% a year over the same period. (focustaiwan.tw) Reuters reported on April 22 that TSMC plans to open an advanced packaging plant in Arizona by 2029. Executive Kevin Zhang said the company wants to build a complete artificial intelligence supply chain in the United States. (reuters.com) TSMC’s Arizona chip output is also scheduled to rise before then. The company said its first Arizona fab is expected to increase output by 80% in 2026 after starting high-volume production in late 2024. (focustaiwan.tw) The timing matters for Nvidia, Advanced Micro Devices and other AI chip designers that depend on TSMC to manufacture and package their accelerators. Reuters reported that advanced packaging has become a supply bottleneck for Nvidia and other customers. (reuters.com) For now, TSMC is trying to expand both the chip factories and the “glue” that binds multiple dies into one product. The company’s message in late April was that AI demand is still outrunning every part of that chain. (focustaiwan.tw) (reuters.com)