TSM · Taiwan Semiconductor ManufacturingSemiconductors
Business Model
TSMC is the world's largest contract chipmaker: companies like Apple, NVIDIA and AMD design chips but do not build them — TSMC does. It is the only place on earth that can mass-produce the most advanced chips, which is why nearly every leading AI processor is made in its fabs.
TSMC is a pure-play foundry: it manufactures chips it does not design, selling wafer capacity per-wafer to fabless customers, with more advanced nodes priced higher. It sells into an unusual market structure — at the leading edge it is the only supplier able to make chips at volume, giving it a scarcity-pricing position.
The core economic engine is the leading-node premium. Each new process generation (N3→N2→A14) commands higher per-wafer pricing; because AI/HPC demand keeps outstripping industry capacity, TSMC converts that capacity bottleneck into pricing power that flows straight to gross margin.
Revenue splits by end platform: high-performance computing (AI accelerators, now the largest and fastest-growing slice), smartphone, then automotive and IoT; by node, advanced nodes (7nm and below) carry the bulk of wafer revenue. The direction is clear — the mix keeps ratcheting toward the most advanced, highest-margin nodes.
This is an extreme fixed-cost, capital-intensive model: each leading-edge fab costs tens of billions, and profitability hinges on spreading that fixed cost across high wafer volume. When capacity runs full, incremental wafers drop through at very high margin, creating powerful operating leverage; the same intensity turns into a liability the moment utilization slips.
Investment Verdict
As the sole manufacturer of leading-edge AI silicon, TSMC turns a sold-out capacity monopoly into durable pricing power — the Street has priced the growth but still under-models that margin ceiling, leaving a quality-compounder to hold and add on weakness.
px · close · 2026-08-18
AI/HPC sustains ~30% USD revenue growth into FY2027 (Goldman already at 28%); the N2 ramp plus the 3nm hike push gross margin structurally toward 65%+; overseas fabs ramp with less-than-feared dilution. EPS ~$21 at 26.7x = $560.
The FY2026 above-30% guide converts into steady execution (Jan–May already +30% YoY), FY2027 grows ~25–28%; N3/N2 pricing stays firm (the 3nm hike realizes in H2 GM) while non-AI demand normalizes; capital intensity stays high but cash generation is robust. EPS ~$19.2 at 25x = $480, a one-turn re-rate.
AI buildout growth decelerates sharply to 10–15% by FY2027 and utilization falls below 80%; gross margin compresses toward 58–60%; the multiple derates to 20x on geopolitical and cyclical risk premium. EPS ~$16.5 at 20x = $330.
EntryThe margin-durability thesis proving out — the 3nm hike realizing in H2 gross margin and N2 yield heading to 75%+, or a Q2 print that lifts FY2027E EPS above $21 with 65%+ GM sustained; alongside a pullback that widens the risk-reward past the add line.
ExitThe AI-capex-plateau bear trigger: utilization commentary shifting from supply-constrained to demand-balanced, two consecutive months of negative YoY monthly revenue, or a top-5 customer converting a Samsung 2nm evaluation into a volume dual-source win that cracks the leading-edge monopoly.
Consensus FY27E EPS $21.78 sits closest to our Bull EPS $21.00 — the market already prices the Bull earnings as its base case.