Stack L2
TSMC
Sole-source CoWoS bottleneck for AI accelerators; capacity doubling again in 2026 to ~150k wpm
HQ TW
Fit Score
avg of 4 segmentsRecent Business Signals
Strategic Position
What they do best
Industrialize CoWoS-L + HBM stack integration at a scale no merchant OSAT can match — 75k wpm by end-2025 doubling again in 2026 (f9208998).
Their bet
That CoWoS-L with 8-Hi/12-Hi HBM4 on N3P (signal d6da49ee) becomes the default substrate for every 2026-2027 frontier-AI accelerator, not just NVIDIA Rubin.
Top risk
ASE FOCoS-Bridge or Amkor Arizona qualifying at NVIDIA/AMD before end-2026 would cap CoWoS pricing power and strand part of the doubled 2026 capacity.
Compared to peers
Direct competitor
ASE Technology
Merchant OSAT pushing FOCoS-Bridge as the cost-relief alternative to CoWoS, but lacks TSMC's co-located logic node tie-in.
Substitute
Amkor Technology
US-sourced CHIPS-funded Arizona advanced-packaging line — geopolitical hedge, not yet a capacity-equivalent threat.
Why someone would join
- 1.Capex guided to $38-42B in 2025 (signal fc8c5297) plus a $100B incremental US commitment on March 3, 2025 — every new hire ships into funded, customer-locked CoWoS-L and SoIC lines.
- 2.Active reqs across CoWoS HBM integration, hybrid bonding/SoIC, TSV, and 2.5D test (5 senior+ roles since Feb 2026) — the exact stack feeding NVIDIA Rubin and Micron HBM4 base die.
Recent Hiring (60 days)
- advanced packaging1
- CoWoS1
- HBM integration1
- hybrid bonding1
- SoIC1