TSMC 2nm Monthly Capacity to Hit 120K Wafers by Year-End, Far Ahead of Market Expectations
nashnova research
TSMC has raised its year-end 2nm monthly capacity target to 120,000 wafers, roughly 20% above the street's 90,000–100,000 estimate — effectively pulling forward a milestone originally slated for 2027 by about two years, driven by a collective order surge from the entire AI compute chain.
What does 120,000 wafers actually mean — and why is it "far ahead"?
The market had broadly expected TSMC's 2nm monthly capacity to reach 90,000–100,000 wafers by year-end. The new target of 120,000 overshoots that by roughly 20%–33%.
This means → a capacity level originally penciled in for 2027 is arriving about two years early.
Senior VP Kevin Hou confirmed publicly: this is the fastest ramp in TSMC's history.
Who is placing the extra orders — and how much more?
The clients driving the surge are the core names in AI and high-performance computing: Apple, Nvidia, AMD, Qualcomm, and MediaTek.
Each added 10%–20% to their original bookings; both Apple and the non-Apple camp scaled up in parallel.
In plain terms = no single customer pulled the trigger alone. The entire AI compute supply chain is racing to lock in capacity, and that collective demand is what pushed the ramp to this scale.
Five fabs ramping at once — how unusual is that?
TSMC will simultaneously ramp five 2nm fabs this year — two in Hsinchu, three in Kaohsiung.
This has never happened before in TSMC's history; no prior node saw five fabs start volume production at the same time.
This means → TSMC is betting, with bricks and steel, that AI demand is not a short-term spike but a sustained migration.
How does the 2nm ramp compare with 3nm?
First-year 2nm wafer output will be 45% higher than 3nm's first-year output in 2023.
The 2026–2028 compound annual growth rate for 2nm capacity is projected at 70%; for 3nm over the same window, it is roughly 25%.
In plain terms = the 3nm ramp was already considered fast. The 2nm ramp is nearly three times that pace — a gap that directly reflects how quickly the market is migrating to more advanced nodes.
What is happening with 3nm and the global footprint?
Southern Taiwan Science Park 3nm monthly capacity will approach 180,000 wafers in Q4 this year, with an optimistic path to 200,000–210,000 by 2027.
TSMC is adding three new 3nm fabs — in Taiwan, Arizona, and Japan. The second Arizona fab is already built and targets volume production in H2 2027.
TSMC is also converting some 5nm equipment in Taiwan to support 3nm output — in plain terms = older lines are not sitting idle; they are being retooled to squeeze more out of existing sites.
What does the market watch next?
Verification point one: whether the 120,000-wafer monthly target is met on schedule by year-end.
Verification point two: whether the 70% CAGR holds through 2026–2028.
This means → TSMC's expansion narrative has moved from "planning" to "delivery" — whether the numbers land will directly shape how the market prices long-term advanced-node demand.
市场有风险，内容仅供研究参考，不构成投资建议。