TSMC is the only foundry scaling advanced ≤7nm logic — and on the segment numbers the moat is widening

• Share is the proof. Advanced ≤7nm chips were 35% of wafer revenue in FY20 and are 74% in Q1 FY26; at 3nm and below, TSMC's economic share is roughly 90%. Apple, Nvidia, AMD, Broadcom and Qualcomm have no second source for current designs.
• Both credible rivals went backwards in the year TSMC grew 31.6%. Samsung Foundry's segment revenue fell 3.9% in FY25; Intel Foundry posted a US$13B operating loss on roughly US$18B of revenue. Each new node costs US$15–25B and 5+ years to ramp — only the incumbent earning 44% ROIC can fund the next chase.
• The 2023 downcycle was the cleanest stress test. Revenue fell 4.5% on inventory destocking, but operating margin held 42.6% and ROIC stayed 27% — both above every rival's best-ever year.

Q1 FY26's 66.2% gross margin — new regime, or the cycle peak?

• Where Q1 sits. Gross margin 66.2% and operating margin 58.1% — above the FY22 super-cycle peak, with HPC at 61% of revenue versus 41% in FY22. Management's own through-cycle anchor was raised in 2026 from 53% to 56%+ — at peak, after the print.
• The disclosed headwind stack management has not yet absorbed. N2 ramp dilution of 2–3pp through 2026, overseas-fab dilution 2–3pp widening to 3–4pp through 2028, plus material and gas cost increases — stacked, a 5–7pp peak gross-margin headwind sits ahead. Q1 absorbed only the start.
• Each percentage point matters. At this revenue base, every point of gross margin is worth ~NT$38B of operating profit. A return to management's own 56% anchor would compress operating profit by NT$380B+ on flat revenue — and the 29× P/E and 18× EV/EBITDA (the highest fiscal-year-end multiple in the 11-year dataset) leave no room for that math.

Record revenue, record margin, record cash, and a balance sheet built like a sovereign

Capex of NT$1.27T in FY25 — larger than the combined annual revenue of UMC, GlobalFoundries and SMIC — funds the next two nodes plus the Arizona, Kumamoto and Dresden builds. Even after that bill, net cash sits at NT$2.10T, dividends grew at 12% CAGR over five years on a flat share count, and stock-based comp runs 0.07% of net income. The FY23 trough ROIC of 27% still tops any rival's best-ever year.

The FY23 stress test no longer applies — the company that printed it does not exist any more

• What protected TSMC in FY23 was that HPC was secular while smartphone cycled. HPC was 43% of mix then and the No.2 customer was 11% of revenue. In Q1 FY26, HPC is 61% and the No.2 customer is 17% (per the FY2025 20-F) — the segment that absorbed the last shock now is the next shock.
• FY26 capex of NT$1.6–1.7T pre-loads three years of depreciation against AI continuation. If aggregate hyperscaler AI capex (Microsoft + Google + Meta + Amazon + Oracle) flattens for two quarters with capacity already built, the setup is a classical foundry downcycle with peak fixed costs and a customer pool of five names rather than thirty.
• Concentration is widening on both sides. Top-10 customer share rose from 70% in FY23 to 78% in FY25; the No.2 customer doubled from 11% to 17%; ~63% of ≤7nm capacity sits in Taiwan, and Arizona's first ramp already dilutes gross margin 2–3pp through 2028. The marginal customer's incentive to qualify a second source has never been higher.