TSMC
What TSMC is up to now
New 08-25 (late catch, 08-24 event): TSMC is the fab behind Xiaomi's new in-house 3nm flagship SoC. Xiaomi launched the Xring/Xuanjie O3 — 24bn+ transistors, claimed first mobile SoC past a 5-million AnTuTu score — plus a 3nm Xuanjie D100 autonomous-driving part, both on TSMC 3nm; a 6nm Xuanjie O100 AI accelerator was announced alongside. ⚠️ Worth holding precisely: a Chinese national champion reducing its dependence on foreign chip DESIGNERS by deepening its dependence on this actor's FAB. Whatever else export-control policy is doing, it is not reducing Chinese demand for leading-edge Taiwanese capacity — which is the variable `tsmc-capacity-race` actually tracks. New 08-23: CoWoS acquired its first visible challenger from inside the memory supply chain — SK hynix disclosed it is evaluating Intel's EMIB packaging alongside CoWoS for future HBM-logic integration. Nothing has moved yet; what changed is that a major HBM maker said publicly it is looking at an alternative. New 08-16: still quiet since 08-14 — no TSMC-specific development in either day's sweep beyond a passing peer-mention (semiconductor-capex chain, 08-15). Standing picture below holds. New 08-11: TSMC's board approved roughly $29.4B in new capital appropriations for advanced-node capacity and packaging (citing AI/HPC demand) and separately executed a definitive agreement with Sony to jointly build a next-generation image-sensor fab in Kumamoto, Japan (Sony contributing ~¥465B, TSMC ~¥282B; production targeted 2029, aimed at sensors for AI-driven robotics and vehicles) — coming weeks after TSMC raised its 2026 equipment-spend guidance to $60-64B. No other TSMC- specific development surfaced this week. Still holding from 08-04→08-05: the Kumamoto earthquake's final toll (38 dead, 127 injured, ~48,300 households without power) with TSMC's own fab impact limited and its 1.4nm (A14) fab now ahead of schedule (first building before April 2027, mass production mid-2028); and the CoWoS packaging capacity constraint — TSMC expanding outsourcing of the CoW front-end bonding step to OSAT partners (ASE, Amkor, SPIL) as Nvidia/ASIC demand outstrips its own capacity, with TrendForce putting Nvidia's 2026 wafer reservation at over half of total CoWoS capacity and projecting a ~20% supply-demand gap through the year. New 08-25: N2 reached commercial shipment via Apple's M6, validating 2nm yield and mass-production maturity ahead of the smartphone, HPC and custom-ASIC customers queued behind it.
as of 2026-08-26
The chokepoint fab; raised 2026 capex to $60-64B + $100B Arizona.
The metrics
- Posture med
- expanding 3 src →
- Capital · available high
- ~$117B cash + ST investments (~$109B-$119B across FX 29.6-32.3); TWD 3,518,013M at 30 TWD/USD; total debt ~$33B (TWD 982,447M) 2 src →
- Capital · operating high
- ~$88B/yr operating cash flow (TTM TWD 2,634,679M at 30; FY2025 TWD 2,274,976M ~= $76B) — self-funded 1 src →
- Capital · deployed high
- ~$50B/yr capex actually spent (TTM TWD 1,491,123M at 30) climbing toward $60-64B FY2026 guidance; AZ/JP/DE fab buildouts; + CHIPS $6.6B grant + up to $5B loans 2 src →
- Capital · in high
- inflows: ~$148B/yr revenue (TTM TWD 4,440,492M at 30; FY2025 TWD 3,809,054M ~$127B) + US CHIPS $6.6B grant + $5B loans; equity — self-funded, no recent raise 2 src →
- Capital · out high
- outflows: ~$50B/yr fab capex (rising to $60-64B) + dividends; FCF TTM TWD 1,143,557M (~$38B) after capex 1 src →
- Optionality med
- mixed — FCF-generative and self-funding, but a rising slice is locked to the forced AZ/JP/DE buildout ($265B US commitment; $60-64B 2026 capex) 2 src →
- Gravity med
- GROSS ~$148B/yr foundry revenue sits under multi-trillion downstream — ~70% global foundry share, ~90%+ of <7nm; sole/main supplier to Nvidia/Apple/AMD/Broadcom/Qualcomm — CHOKEPOINT … 3 src →
