Taiwanese trade press reported on August 3 that TSMC's 3nm output will reach 180,000 wafer starts per month by early in the fourth quarter of 2026, up from roughly 150,000 in the first half and two to three months ahead of the expected ramp. Demand is attributed to Nvidia, AMD and Broadcom.
The wider ramp
2nm moves from 50,000-60,000 wafer starts in the first half to more than 80,000 by early Q4 and about 100,000 by year-end. Combined, the two leading nodes pass 260,000 wafer starts a month. Longer term, three further 3nm fabs are planned across Taiwan, Arizona and Japan, and five 2nm fabs in total — two in Hsinchu, three in Kaohsiung.
Whose numbers these are
TSMC does not publish wafer capacity by node. Every figure above is supply-chain reporting, and trade estimates for 3nm over the same period have ranged from 150,000 to 175,000 — so part of the apparent increase depends on which estimate you started from. "Two to three months ahead of expectations" means ahead of analysts' modelled ramp. TSMC has committed to nothing.
Where the wafers actually come from
The near-term addition comes from converting existing lines and moving tools, not from the three additional fabs, which are years out. The 1.4nm dates in the same reporting — first building complete before April 2027, pilot production possibly early Q3 2027, mass production targeted for the second half of 2028 — are aspirations on the same footing.
Why 3nm and not 2nm
3nm is the node today's accelerators ship on. Pulling roughly 30,000 wafer starts forward by a quarter is the most concrete supply signal available for 2027 volume.