TSMC N2 Capacity Squeeze
- TSMC's newest N2 node now matches older 18A yields, tightening cutting‑edge capacity across the industry. - NVIDIA has reportedly booked most N2 capacity, leaving AMD short and Intel pushing IFS tooling for 2027. - That concentrated bookings picture suggests advanced-node supply will stay uneven for the next planning cycles ( ).
The fight over TSMC’s 2-nanometer lines is no longer about lab demos; it is about who can still get wafers in 2026 and 2027. (tsmc.com, epium.com) A process node is the factory recipe for making the smallest, fastest chips, and TSMC’s N2 recipe started volume production in the fourth quarter of 2025. TSMC says N2 uses its first nanosheet transistors and that the follow-on N2P version is scheduled for volume production in the second half of 2026. (tsmc.com) TSMC is building that family at Fab 20 and Fab 22, but outside reports say most early N2 slots are already spoken for through at least late 2026 and, in some accounts, into 2027. Those reports describe customers being pushed to lock in allocations far earlier than on older nodes. (tsmc.com, trendforce.com, epium.com) That squeeze sits on top of TSMC’s broader expansion cycle. In its April 16, 2026 first-quarter presentation, TSMC reported $35.9 billion in revenue, a 66.2% gross margin, and said 70% to 80% of 2026 capital spending would go to advanced process technologies. (investor.tsmc.com) The practical bottleneck is not just silicon starts. Advanced chips also need scarce packaging capacity, and multiple 2026 reports say Nvidia has reserved more than half of TSMC’s advanced CoWoS packaging output for the year. (freshfromchina.com, financialcontent.com) That matters because a company can have a chip design ready and still miss a launch window if it cannot secure both the leading-edge wafers and the packaging line that stitches the dies and memory together. Silicon Analysts said in April 2026 that TSMC’s N2, N3 and N5 capacity was fully booked, while CoWoS lead times were running 50 weeks or longer. (siliconanalysts.com) Intel is trying to use that opening to sell its own foundry path. Intel says 18A is ready for customer projects, and its February 10, 2025 explainer said the process was progressing toward high-volume production in the second half of 2025. (intel.com, newsroom.intel.com) Intel has also kept extending that roadmap. Its foundry process page says Direct Connect 2025 added 18A-PT and 14A-E updates, while Intel’s October 2025 Panther Lake announcement said the first AI PC platform on 18A would enter high-volume production later that year at Fab 52 in Arizona. (intel.com, newsroom.intel.com) The missing piece is that neither TSMC nor Intel publicly breaks out customer-by-customer N2 allocations, so claims that Nvidia has taken most N2 capacity and that AMD is short rely on trade reports and analyst channels rather than company filings. What the public record does show is a market where TSMC’s newest node is in production, packaging is tight, and the companies with the biggest reservations are shaping everybody else’s launch calendar. (tsmc.com, investor.tsmc.com, intel.com, epium.com)
