Long-Term Thesis

Long-Term Thesis — 5-to-10-Year View

Figures converted from TWD at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.

1. Long-Term Thesis in One Page

The long-term thesis is that TSMC compounds because the leading-edge foundry is becoming structurally less cyclical and more concentrated — every new node costs more, takes longer, and consolidates a smaller pool of customers onto the only foundry that can hit yield. The 5-to-10-year case works only if (a) Samsung Foundry and Intel Foundry continue to under-deliver on advanced-node yield, (b) AI/HPC remains the dominant logic-design driver, and (c) management keeps recycling 30%+ of revenue into capex that earns 30%+ ROIC. This is not a "great franchise" story to be priced at any multiple — at $40B of FY25 capex and 33% capex/revenue, the duration of the runway hinges on every wafer dollar continuing to earn 44% returns on capital, and the geographic/customer concentration that secures pricing power today is also the only configuration of facts that could end the thesis abruptly.

2. The 5-to-10-Year Underwriting Map

The thesis is not one bet — it is six bets, each individually observable. Five of the six must hold for the compounder narrative to remain intact through FY2030; the sixth (geopolitics) is binary tail-risk that sits outside the moat framework and should be sized as such.

Driver 1 is the load-bearing one. Every other driver — AI demand, margin regime, capex productivity, allocation discipline — collapses in scenario terms if a competitor lands a major HPC customer at 2nm with volume yield. The reverse is not true: even a deep AI-cycle correction (Driver 2 failing) compresses earnings without breaking the franchise, as FY23 proved (revenue –4.5%, op margin held 42.6%, ROIC 27%). The thesis is robust to demand cycles; it is fragile to a single yield-gap closure.

3. Compounding Path

TSMC is a capital-heavy compounder: it converts roughly 50–60% of net income into free cash flow over a full cycle and reinvests the bulk of operating cash into next-node capacity. The compounding signal is not FCF yield in any one year but the trajectory of operating margin, ROIC, and EPS through the cycle. The 11-year record shows revenue 4.7× ($25.7B → $121B), EPS per ADR-equivalent 5.9× ($1.78 → $10.54), operating margin lifted from 38–40% to a 50.8% record, all on a share count that did not grow — the cleanest compounding pattern in semiconductors.

The two charts tell the durable story. Revenue and EPS compounded steadily through one normal mini-cycle (2019 handset glut), one global pandemic, and one inventory correction (2023). The margin chart tells the cause: operating margin held in a 35–40% band through FY2014–2021 — the classical foundry era — and then stepped up to a 43–51% band as advanced-node mix took over. ROIC followed, reaching 43.7% in FY25 — the highest of the decade and ~5× any peer pure-play. The thesis is that the upper band is the new through-cycle, not a peak. Management's own anchor (56%+ GM, high-20s ROE) explicitly raises the floor in 2026.

Capex absolute spend grew 5.1× over the decade — from $7.8B in 2015 to $40B in 2025 — but capex/revenue fell from a 53% peak (FY21 super-cycle) to 33%. That divergence is the moat in cash-flow form: the absolute dollar commitment is record-large, but the asset base earns enough to fund itself without compressing the intensity ratio. The forward guide (FY26 capex $52–56B at the top of historical absolute) keeps the bet sized for the AI demand to continue.

These are management's own anchors — not analyst extrapolations. The 5-to-10-year thesis is essentially a bet that TSMC executes against this set with the same 9-of-10 hit-rate it ran in FY2020–FY2025. The compounding math from here is: revenue at ~25% USD CAGR through FY29 implies FY29 revenue ~$295B; at the 56%+ GM floor and a mid-40s op-margin band, FY29 operating profit could land in the $145–160B range — i.e., today's revenue base. That is what "right tail" looks like; the left tail is FY27 reversion to mgmt's own 56% floor on flat revenue, which compresses op profit by $12B+.

4. Durability and Moat Tests

The wide-moat conclusion is unconditional only for the leading-edge segment (~61% of revenue at 3nm/5nm). Five tests separate durable thesis evidence from short-term noise — at least one is competitive, at least one is financial, and at least one is observable inside any single quarter.

The honest read: four of five tests currently validate the thesis. The fifth (overseas fab dilution) is open — mgmt's own guide has the dilution widening before it rolls off, and the political conditionality (tariffs, equity stakes, IP licensing) is the part of the model that lives outside corporate playbook. The thesis is robust to demand cycles and rival catch-up attempts because both have been stress-tested already; it is most exposed to political shocks because those have not.

5. Management and Capital Allocation Over a Cycle

The 5-to-10-year thesis cannot survive a management transition that breaks capital-allocation discipline — and TSMC has lived through exactly one transition (Mark Liu/C.C. Wei split → C.C. Wei consolidated chair/CEO in June 2024) without breaking pattern. The chair/CEO consolidation is the cleanest A→A– governance flag: a structure that worked under prior chairs (Liu chair + Wei CEO, with founder Morris Chang as honorary chair) was streamlined, and the board did not designate a Lead Independent Director to backstop the consolidation. The board itself is 7/10 independent with four directors who have deep semiconductor operating experience (Splinter ex-Applied/Intel; Gavrielov ex-Xilinx; Bonfield ex-NXP; Reif ex-MIT) and 100%-independent audit and comp committees.

The capital-allocation pattern is the part that matters most for compounding. Reinvestment dominates: 70–80% of operating cash flow has gone into capex every year of the last decade. Dividends grew 12% CAGR ($8.7B → $15.0B over 5 years) on a stable ~27% payout. Buybacks are negligible ($0.1B in FY24, zero in FY25) and the share count has been flat at ~5.19B ADR-equivalent for a decade. SBC is 0.07% of net income — a rounding error. The capital structure is what an owner-aligned compounder looks like: zero financial engineering, zero share count creep, dividends growing in line with through-cycle earnings, and the discretionary capital recycled into the only asset base in the industry earning >40% ROIC. The right test is whether mgmt continues this when ROIC compresses — the policy mix should swing toward higher dividends and buybacks when the leading-edge runway flattens, and an investor should want that swing because it would be the honest read on a moat that has narrowed.

Compensation reinforces the pattern. CEO C.C. Wei's FY25 total of $77M (a Taiwan-listed record) is overwhelmingly variable, tied to TSR vs peers and ESG metrics; base salary is 0.7% of the total. The articles of incorporation cap directors at 0.3% of profit and require at least 1% of profit to employees — $3.3B of profit share went to employees in FY25. There is no founding-family blockholder, no dual-class structure, and the largest single owner is the R.O.C. government's National Development Fund at 6.38% — a stabilising rather than controlling position. The succession map is also unusually deep: Co-COOs Y.P. Chyn (39 years operations) and Y.J. Mii (32 years R&D) sit one tier below C.C. Wei, and CFO Wendell Huang has 27 years inside the company. The credibility of the multi-year guidance frame (12 of 13 valuation-relevant promises hit since 2018; raised twice) is itself a moat in the sense that the market should pay a higher multiple for promises that have actually been met.

6. Failure Modes

The thesis breaks in specific, identifiable ways. Six failure modes — three "company-level" (concentrated, observable in P&L), three "policy/geopolitical" (uncorrelated to operations and harder to underwrite). "Execution risk" is not on this list because TSMC's execution record is the closest thing in semiconductors to a known constant.

7. What To Watch Over Years, Not Just Quarters

Five multi-year milestones — none is a quarterly earnings event — that will update the thesis decisively. Quarterly margin oscillations and monthly revenue prints are noise relative to these.

The long-term thesis changes most if Samsung Foundry returns to sustained double-digit growth with a publicly named major HPC customer on 2nm. That single observable — visible inside Samsung's own quarterly business report and confirmed by TrendForce/Counterpoint segment estimates — is the highest-value multi-year signal. Every other milestone (N2 ramp, capex productivity, dilution roll-off, customer concentration) updates the thesis at the margin; Samsung 2nm with a named HPC customer is the evidence that would flip the moat conclusion from wide to narrow on a single earnings cycle.