TSMC warns shortages to 2027+
- TSMC warned that AI-driven chip demand will keep advanced-node shortages stretching into 2027 and beyond. - The company said it is investing $56 billion in new fabs but delayed Arizona 3nm volume production until H2 2027. - The statement makes advanced-node capacity a structural constraint for product roadmaps and silicon-dependent timelines ( ).
Taiwan Semiconductor Manufacturing Co. said on April 16 that demand for advanced artificial-intelligence chips is still running ahead of supply, and the squeeze could last into 2027 and beyond. (investor.tsmc.com) TSMC gave that warning alongside first-quarter results showing revenue of NT$1.13 trillion, up 35.1% from a year earlier, with second-quarter sales guided to US$39.0 billion to US$40.2 billion. (investor.tsmc.com) The company is spending at the top end of a US$52 billion to US$56 billion 2026 capital budget, after first setting that range in January. Chief Executive C.C. Wei said the added spending is aimed at leading-edge process technology and advanced packaging used in artificial-intelligence systems. (investor.tsmc.com, investor.tsmc.com) An advanced node is the newest generation of chipmaking, measured in nanometers, and it is where the fastest graphics processors and custom artificial-intelligence chips are built. TSMC says it started volume production of 2-nanometer chips in the fourth quarter of 2025, while 3-nanometer entered high-volume production in 2022. (tsmc.com) That matters because TSMC is the main manufacturer for many of the chips that power data-center artificial-intelligence systems, and those products cannot ship on schedule if wafer starts or packaging slots are unavailable. TSMC said high-performance computing, the category that includes many artificial-intelligence chips, accounted for 59% of first-quarter revenue. (investor.tsmc.com) The bottleneck is not only the silicon wafer itself. TSMC has repeatedly pointed to advanced packaging, including chip-on-wafer-on-substrate, or CoWoS, as a second choke point because artificial-intelligence processors combine multiple dies and high-bandwidth memory in one package. (investor.tsmc.com, investor.tsmc.com) TSMC is also expanding in the United States, but that capacity arrives on a long clock. The company said in March 2025 that it planned to raise its U.S. investment to US$165 billion, adding three new fabs, two advanced-packaging facilities and a research center in Arizona. (pr.tsmc.com) Arizona is already producing 4-nanometer chips at its first fab, but the next step is later than many customers would like. Industry and local reports tied to TSMC’s Arizona buildout say the second fab’s 3-nanometer production is targeted for 2027, with equipment move-in around the third quarter of 2026. (tsmc.com, trendforce.com) TSMC’s own filings show how concentrated demand has become at the cutting edge. In the first quarter of 2025, 3-nanometer chips made up 22% of wafer revenue and 5-nanometer made up 36%, meaning more than half of sales already came from the newest process families before 2-nanometer ramped. (investor.tsmc.com) For chip designers, the message from TSMC is that capacity planning is now part of product planning. If demand stays at current levels, the
