[News] TSMC’s 3nm Reportedly to Overtake 5nm as Top Revenue Node in 2H26; Output Value Could Top NT$400B
TSMC’s 3nm node is set for a stronger ramp in 2H26, fueled by surging demand for AI and high-performance computing (HPC), particularly as NVIDIA’s Rubin platform ramps up. Citing analysts, the Economic Daily News reports that 3nm is expected to overtake 5nm in revenue contribution for the first time in 2H26, with quarterly output value projected to exceed NT$400 billion in 3Q26—a record high and more than 30% of total revenue.
According to TSMC’s earnings report, while 5nm accounted for 33% of its 2Q26 sales, 3nm’s revenue contribution rose to 30% during the period from 25% in 1Q26. The rising 3nm mix is also expected to bolster TSMC’s gross margin, the Economic Daily News adds.
Meanwhile, strong follow-on orders from major AI chipmakers including NVIDIA, AMD, and Broadcom are driving further 3nm capacity expansion, Commercial Times reports. The report, citing Wedbush, suggests TSMC is expected to lift monthly 3nm wafer starts to 180K in early 4Q26, up from 150K in 1H26.
With 2nm also ramping up, TSMC’s combined monthly capacity for 2nm and 3nm is expected to exceed 400K wafers by 2028, based on the company’s official 2nm growth guidance and market estimates, the report notes.
While TSMC’s 3nm family remains capacity-constrained, the company also speeds up expansion, both in Taiwan and overseas. As previously reported by MoneyDJ, TSMC plans to add another 3nm fab at the Tainan Science Park, with mass production expected to begin in 1H27.
In Arizona, the company’s second fab will also adopt 3nm technology, with mass production slated to start in 2H27, the report notes. In Japan, TSMC plans to introduce N3 technology at its second fab, with mass production expected to begin in 2028.
A16 Ramps Up, With Potential Clients in Focus
TSMC is also advancing its next-generation node roadmap. According to the Economic Daily News, A16, the company’s first angstrom-class process, is expected to be production-ready in 2H26, with industry sources widely expecting Fab 20 in Hsinchu to lead production.
Apple and NVIDIA are expected to be among A16’s first customers, while OpenAI is reportedly poised to become another key client. The development has drawn particular attention as OpenAI’s first-generation Jalapeño chip, co-developed with Broadcom, is built on TSMC’s 3nm process. A separate Economic Daily News report in June said its successor could move to A16 and adopt TSMC’s CoWoS advanced packaging technology.
NVIDIA is also lining up A16 for its next-generation products. As reported by Commercial Times, the chipmaker plans to use A16 for its next-gen Feynman GPU, while also weighing the process for its Rosa CPUs.
Read more
- [News] TSMC 3nm Monthly Wafer Starts to Hit 180K by Early 4Q26 on Strong Demand, 2–3 Months Ahead of Expectations
- [News] TSMC 3nm Monthly Capacity May Hit 180K Wafers by 2026, Up Over 40% YoY on AI Demand
(Photo credit: TSMC)
