Orders tied to Nvidia’s Rubin platform are pushing TSMC’s 3nm process toward the top of the company’s revenue stack. Market reports gathered by TrendForce on Sept 7 say 3nm will out-earn 5nm in the second half of 2026 for the first time, with quarterly output value heading past NT$400B, a record worth more than 30% of total sales.
The process contributed 30% of TSMC revenue in Q2, up from 25% a quarter earlier, though 5nm still led at 33%. AMD and Broadcom are adding follow-on orders to the Rubin wave, and analysts at Wedbush expect monthly 3nm wafer starts to climb to about 180,000 early in Q4 from 150,000 in the first half.
TSMC is stretching 3nm capacity across three sites. A new fab in the Tainan Science Park is slated for mass production in the first half of 2027, its Arizona second fab will run 3nm from the second half of 2027, and Japan’s second fab brings N3 online in 2028. A richer product mix also cushions gross margin as the company absorbs heavier depreciation.
Beyond 3nm, the first angstrom-class node A16 is expected to be production-ready in the second half of 2026 at Fab 20 in Hsinchu, with Apple and Nvidia tipped as launch customers and OpenAI named as another possible buyer. OpenAI’s first custom processor, the Broadcom-built Jalapeno, currently runs on 3nm, and industry watchers expect its successor to move to A16 with CoWoS packaging.
