Bull & Bear

Bull and Bear

Verdict: Lean Long, Wait For Confirmation — the Bull's hard evidence (FY2023 trough margins above every rival's peak, ROIC 43.7%, NT$1.00T FCF on record capex, clean forensics) is durable in a way the Bear's "peak everything" framing cannot match, but it is not yet enough to underwrite a stock priced at the highest fiscal-year-end EV/EBITDA in its own 11-year dataset.

The decisive question is not whether AI is real — it is whether the 66.2% Q1 FY26 gross margin survives the first hyperscaler capex digestion. Bull frames the FY23 mini-cycle (revenue −4.5%, OM 42.6%) as proof TSMC is now a "non-cyclical compounder"; Bear is right that we have not yet seen this margin regime tested against a real AI capex pause. Q2 FY26 — Bull's own stated catalyst (≥64% GM with HPC mix ≥60%) — is a cheap evidence marker that resolves the tension at low time cost. Until that prints, paying 18× EV/EBITDA on margins 10 points above management's own through-cycle anchor is a willingness-to-be-wrong trade, not a margin-of-safety trade.

Bull Case

Bull's strongest cards are facts that already happened in the income statement, not promises about AI.

Bull's price target is NT$12,600 per share on 28× FY26E EPS of NT$450, the upper end of the 11-year multiple range, on the view that AI/HPC mix sustains a 62–66% gross-margin band and US$ revenue compounds ~25% through FY29. Timeline is 12–18 months, anchored to Q2 FY26 confirming ≥64% GM with HPC mix ≥60%. Bull's stated disconfirming signal is Samsung Foundry returning to double-digit YoY growth with a publicly named HPC customer (Nvidia, AMD, Broadcom, or Qualcomm) at 2nm — the technical fact that would invalidate "no second source." The dropped fourth point ("the market is paying for the base case, not the bull case") is a valuation framing, not new evidence.

Bear Case

Bear's strongest cards are facts about price and concentration, not predictions about AI demand.

Bear's downside target is NT$1,200 per common share on 20× (10-year P/E mean) × normalized FY27 EPS ~NT$60, reflecting GM reverting to the company's own 56% through-cycle anchor and a single-digit revenue decline year. Timeline is 12–18 months. Primary trigger is two consecutive quarters of flat-to-down aggregate hyperscaler AI capex (Microsoft + Google + Meta + Amazon + Oracle) coupled with TSMC's capex productivity (Rev / Net PP&E) drifting from 1.2× toward 1.0×. The cover signal is Samsung Foundry posting a third consecutive quarter of segment revenue decline AND TSMC sustaining GM ≥62% across two quarters of growing trailing-12-month hyperscaler AI capex — both pieces required. The dropped fourth point ("pricing-power moat shows cracks" via the 2025 IP indictments) was rhetorically sharp but its P&L impact is speculative.

The Real Debate

Two tensions matter; both rest on the same underlying facts, with Bull and Bear reading them in opposite directions.

Verdict

Lean Long, Wait For Confirmation. The Bull carries more weight because the durable variables — return on capital, free cash conversion, accrual quality, share-count discipline, and the FY23 mini-cycle stress test — are already in the record and dominate any rival's best-ever year, whereas the Bear's strongest argument depends on a hyperscaler capex pause that has not yet happened. The single most important tension is whether Q1 FY26's 66.2% gross margin is the new regime or a cycle peak — and unlike the other two, this one has a near-term observable answer rather than a multi-year arc. The Bear could still be right: a 10-point gap between today's GM and management's own 56% through-cycle anchor cannot logically be permanent, and at 18× EV/EBITDA the stock prices the optimistic interpretation. The condition that flips this to Lean Long (full) is a clean Q2 FY26 print at ≥64% GM with HPC mix ≥60% — Bull's own catalyst — which is the near-term evidence marker; the condition that flips it to Avoid is the durable thesis breaker, namely a named major HPC tape-out moving to Samsung 2nm or Intel 18A within 2026–27, because that severs the "single asset, single operator" premise the entire margin regime rests on. Until then, the asymmetry of waiting one quarter is favorable: the cost of a Bull-confirming Q2 print is one quarter of opportunity cost, while a Bear-confirming print at peak multiple carries materially more downside.