Taiwan Semiconductor Manufacturing
The world's dominant contract chipmaker and the manufacturing chokepoint of the AI buildout: roughly 72% of global foundry revenue, over 90% of leading-edge (sub-5nm) capacity and more than 90% of CoWoS advanced packaging. Q2 2026 revenue reached $40.2B (+36% YoY) at a 67.7% gross margin, with HPC - mostly AI accelerators - up 20% QoQ to 66% of revenue. Management lifted FY2026 growth guidance to slightly above 40% in USD and 2026 capex to $60-64B, and August 2026 monthly revenue hit a record NT$514.8B (+53% YoY).
Scenarios
Every leading AI accelerator (Nvidia, AMD, Broadcom and hyperscaler custom ASICs) is built on TSMC N3/N2 and packaged with its CoWoS, so AI capex flows through TSMC regardless of which chip designer wins. N2 ramps toward ~140K wafers/month by end-2026 and CoWoS toward 120-130K wafers/month, both effectively sold out, and no rival has matched its yields at 3nm and below.
Over 90% of advanced-node capacity sits in Taiwan, making the stock a direct proxy for cross-strait risk. US semiconductor tariff threats and export-control politics add policy risk, Arizona fabs carry higher cost and only reach volume at 2nm from 2H 2027, and a pause in hyperscaler AI capex would hit a business now two-thirds exposed to HPC. Samsung and Intel Foundry remain long-shot but funded alternatives.
Key Factors to Watch
- ●Q2 2026 revenue $40.2B (+36% YoY), gross margin 67.7%; HPC 66% of revenue, up 20% QoQ
- ●FY2026 growth guided slightly above 40% (USD); capex raised to $60-64B from $52-56B
- ●Over 90% of leading-edge logic and CoWoS packaging capacity - the physical bottleneck of AI compute
- ●Taiwan concentration, US tariff threats and export controls are the dominant risks, not AI disruption
- ●Deploys Nvidia AI in its own fabs (cuLitho computational lithography, ML process control, defect-detection vision)
Manufacturing Peers
Last researched: 2026-09-26
This is research and analysis, not financial advice. Scores reflect AI impact potential, not investment recommendations.