TSMC Pulls Forward 3nm to 180,000 Wafers per Month as 1.4nm Fab Stays Ahead of Schedule
TLDR
- TSMC now expects to hit 180,000 3nm wafer starts per month by early 4Q26, two to three months earlier than the industry had pencilled in
- Strong pull from NVIDIA, AMD, and Broadcom is forcing the accelerated ramp, with 3nm monthly starts already near 150,000 in H1 2026
- 2nm is not far behind: combined 2nm and 3nm monthly wafer starts are tipped to clear 260,000 by early Q4
- TSMC’s 1.4nm fab in the Central Taiwan Science Park is ahead of schedule, with pilot production possible from 3Q27 and mass production slated for 2H28
- For Malaysia, the ripple effect lands in Penang and Kulim where OSAT, photomask, and substrate vendors feed the advanced-node pipeline
Taiwan Semiconductor Manufacturing Company is accelerating its 3nm ramp faster than anyone outside Hsinchu expected. According to TrendForce, citing Taiwan’s Economic Daily News, TSMC is on track to reach 180,000 3nm wafer starts per month by early in the fourth quarter of 2026, a full two to three months ahead of the timeline most analysts had assumed.
The pull is coming from the usual heavy hitters. NVIDIA, AMD, and Broadcom are stacking orders for AI accelerators, data-centre silicon, and high-end networking chips that all land on TSMC’s 3nm family, and the foundry is bending its capacity plan to match. Monthly 3nm wafer starts already sat near 150,000 in the first half, so the jump to 180,000 by October is a real capacity step rather than a marketing line.
2nm Closes In From Behind
3nm is the headline, but 2nm is the bigger story underneath. TSMC is converting 5nm production tools in Taiwan to feed additional 3nm lines and has committed to three more 3nm fabs across Taiwan, Arizona, and Japan. Five 2nm fabs, two in Hsinchu and three in Kaohsiung, are ramping through 2026, and the company has guided to a first-year 2nm wafer output that is roughly 45 percent higher than what 3nm managed in 2023.
Monthly 2nm wafer starts already ran around 50,000 to 60,000 in H1 and are expected to clear 80,000 by early Q4. Add that to the 180,000 3nm target and TSMC’s combined advanced-node output crosses 260,000 wafers a month, a number that explains why rivals are scrambling to keep their roadmaps credible.
1.4nm Is Also Ahead of Plan
The aggressive posture extends past 3nm and 2nm. TSMC’s 1.4nm fab at the Central Taiwan Science Park is reportedly running ahead of construction schedule, with the first building expected to wrap up before April 2027. If the pace holds, pilot production could start as early as Q3 2027, with mass production targeted for H2 2028. That puts the 1.4nm node into the conversation for next-generation AI accelerators and flagship mobile silicon far sooner than the conservative 2029 timelines floated last year.
What This Means for Malaysia
Penang and Kulim sit one tier down the supply chain from TSMC’s front-end fabs, but they ride the same wave. Malaysian OSAT players, photomask suppliers, substrate vendors, and equipment service companies see order books swell when TSMC pulls capacity forward. The accelerated 3nm ramp also pulls forward demand for advanced packaging, including CoWoS, much of which is supported by Taiwanese supply chain partners with growing footprints in Malaysia.
Our Take
The story here is not really that TSMC is fast. The foundry has been pulling nodes forward for years. The story is how lopsided the AI demand picture has become, so much so that even TSMC’s famously conservative capacity planning is bending. Two to three months sounds small until you remember that 180,000 wafers a month at 3nm wafer prices is a multi-billion-dollar revenue shift per quarter.
Samsung Foundry and Intel Foundry will read this carefully. Samsung’s 4nm line is already booked out through next year, and its 2nm yield story is improving, but TSMC pulling 3nm forward by a quarter effectively narrows the window in which rivals can win over defect-tolerant designs. Intel’s 18A and 14A strategy needs a marquee external customer win in the next two quarters to stay in the conversation.
For Southeast Asia, the practical takeaway is that any plan tied to the AI hardware cycle, including Malaysia’s New Industrial Master Plan 2030 targets for the electrical and electronics sector, is now building on a faster-moving base than the policy documents assume.
