Moat
Moat — What Protects TSMC
Figures converted from TWD at historical FX rates — see data/company.json.fx_rates. Ratios, margins, and multiples are unitless and unchanged.
1. Moat in One Page
Conclusion: wide moat. TSMC has the clearest evidenced economic moat in semiconductors — a multi-source advantage (cost-at-scale, accumulated process IP, customer switching costs, and capital-intensity barriers) that has produced industry-outlier returns across four full demand cycles. The FY2025 gross margin of 59.9% sits 30–35 percentage points above every other pure-play foundry; ROIC of 43.7% is 4–5× peers; and the only rival with both the capital and the technology to attack the leading edge (Samsung Foundry) saw foundry revenue decline 3.9% in FY2025 while TSMC's grew 31.6%. The moat is not narrowing — it is widening, mostly because each new node ramp costs more, takes longer, and consolidates a smaller pool of customers onto the one foundry that can hit yield.
The 2–3 strongest pieces of evidence: (1) at ≤5nm, TSMC's economic share is roughly 90%+ — Apple, Nvidia, AMD, Broadcom, Qualcomm, Marvell have no second source for current designs; (2) FY2025 capex of US$40B is larger than the combined annual revenue of UMC, GFS and SMIC, and capex productivity (Rev / Net PP&E) of 1.2× is roughly double the rivals' 0.5×; (3) through the 2023 inventory correction, TSMC held a 42.6% operating margin and 27% ROIC — levels every competitor would call a record year.
The two biggest weaknesses: (1) geographic concentration — ~63% of ≤7nm capacity sits in Taiwan, with the strait risk a binary tail; (2) customer concentration — Apple plus the AI-accelerator complex (Nvidia, AMD, Broadcom) account for roughly 40–45% of revenue, and the entire moat now depends on hyperscaler AI capex continuing to size up the next two node ramps.
Moat rating (1-5; Wide moat)
Evidence strength (0–100)
Durability (0–100)
Weakest link (Taiwan geo)
For a beginner: a moat is a durable economic advantage that lets a company keep earning higher returns, margins, or pricing than rivals can match. TSMC's moat is "wide" because it shows up directly in the P&L (margins, ROIC, share), is multi-source (not one trick), and has survived every prior downturn without breaking the structure. The risks are concentrated and identifiable — they are not "the moat is fake," they are "specific tail events could end it."
2. Sources of Advantage
Each row below names a category of moat, what it would mechanically do to economics if real, what TSMC actually shows, and what could erode it. Three of the six sources are decisively evidenced; one (intangibles/IP) is strong but harder to measure directly; two (regulatory, location) are real but partial.
Two categories that are not the moat, despite being commonly named. A "Taiwan brand" is not a moat — TSMC's customers do not pay a premium for Taiwan-ness; they pay for yield at the leading node. "First-mover" is not a moat either — Intel was first to 22nm FinFET and got passed; UMC was older than TSMC and is now one-tenth its size. The leading-edge yield gap is the moat; everything else is consequence.
3. Evidence the Moat Works
The moat must show up in numbers, not narrative. The table below is the evidence ledger — seven items, each labelled as supporting or refuting the moat thesis, with confidence and known distortions.
The three-bar comparison is the cleanest single read on whether the moat works. The gross-margin gap (left bar) is the leading-edge pricing premium; the operating-margin gap (middle) shows that scale economies amplify it; the ROIC gap (right) confirms it survives the capex bill. No competitor earns positive ROIC on its capital base; TSMC earns roughly 44%.
4. Where the Moat Is Weak or Unproven
The wide-moat conclusion is genuine but not unconditional. Four soft spots warrant honest naming.
First, geographic concentration. Approximately 63% of global ≤7nm capacity sits in Taiwan, and even with the Arizona, Kumamoto, and Dresden ramps, that figure is still ~80%+ through 2027 on TSMC's own footprint. The strait risk is binary at extreme — TSMC could earn 60% gross margins for a decade and still see the equity zeroed in a tail event. The standard rebuttal (the "silicon shield" deters action) is plausible, not provable.
Second, customer concentration. Apple, Nvidia, AMD, Broadcom, and Qualcomm collectively are a meaningful majority of revenue. The behavioural switching cost is high, but each of these customers has a public custom-silicon strategy and the resources to insource at scale if economics shifted decisively. Apple in particular has been the single highest-share customer for a decade; concentration is the cost of being the only credible leading-edge foundry.
Third, the entire moat now leans on AI demand. HPC reached 61% of revenue in Q1 2026 and management's FY24–FY29 25% revenue CAGR target assumes hyperscaler AI capex stays where it is or grows. If aggregate hyperscaler capex flattens or rolls over and TSMC has pre-built capacity at the leading edge, the cycle returns — and a 33% capex/revenue ratio drops earnings disproportionately as utilisation falls.
Fourth, the moat is recreated, not transferred, off-island. The 2–3 pp Arizona margin dilution proves that TSMC's Taiwan cluster — engineer supply, materials, OSAT, utility, permitting — contributes part of the economics. Mandatory geographic diversification (whether driven by customer demand or US/EU policy) is a slow erosion of the location-specific component of the cost advantage.
The fragile assumption underlying the wide-moat call: Samsung Foundry continues to under-deliver on 2nm yield. If Samsung wins one large HPC tape-out at 2nm (Nvidia, AMD, Broadcom, or Qualcomm) within 2026–27 and ramps it to volume, the customer-lock-in evidence weakens immediately, even if TSMC retains 70%+ share for years. This is the single technical variable that could downgrade the moat read from wide to narrow.
5. Moat vs Competitors
The peer panel below uses the same set the Competition page identified, scored on the same moat sources. Two of the five rivals (Samsung, Intel) hold structurally narrower advantages in adjacent territory; three (UMC, GFS, SMIC) are not competing for the same revenue.
The shape of the peer table is the moat read: TSMC owns the largest profit pool in the industry on the strongest set of sources; rivals own specific corners (China demand, specialty mix, capital-light returns) that do not threaten the leading-edge engine. The one peer who could theoretically take share at the leading edge (Samsung) is currently shrinking on the same axis; the one who could theoretically be politically privileged (Intel) is currently subsidising losses to do it.
6. Durability Under Stress
A wide-moat conclusion only stands if the advantage survives stress. The stress table tests the moat against six historically-relevant or plausibly-near scenarios. Five of six pass; one (Taiwan strait disruption) is acknowledged as binary tail risk where moat assessment is the wrong framework.
The honest stress read: four of six scenarios leave the moat intact with some margin or share cost; one (Samsung 2nm yield catch-up) downgrades it from wide to narrow; one (Taiwan strait) is outside the moat framework and should be sized as tail risk separately. The cyclical scenario (AI correction) compresses earnings but does not threaten the structural premium, because the leading-edge node base has nowhere else to go for current designs.
7. Where TSMC Fits — Which Segment Actually Carries the Moat
The moat is not uniform across the business. It is concentrated in the leading edge and dilutes sharply at older nodes — the same fab footprint earns very different economics on different process technologies. This matters because the moat conclusion changes depending on which segment a reader is underwriting.
Two-thirds of FY2025 revenue (3nm + 5nm = 61% Q1 26) sit in the wide-moat segment. About 14% of revenue (16/20nm + 28nm) is narrow-moat — TSMC competes on specialty mix and customer relationships, not technology lead. About 12% (40nm and older) is effectively no-moat — pure-commodity foundry pricing pressured by China-domestic capacity. The valuation premium attaches to the first bucket; the second bucket is what UMC and GFS economics look like; the third bucket is the SMIC pricing problem in microcosm. The wide-moat read is unconditional only for the 61% that sits on the leading edge.
The chart makes the asymmetry visible. The wide-moat segment is 61% of revenue and earns an estimated 65%+ gross margin; the no-moat tail is 12% of revenue at ~30% gross margin. As the leading-edge mix continues to climb (N2 from 2H 2025; A16 risk production 2026), the moat-weighted average revenue share is increasing, not decreasing. This is the clearest argument for why TSMC's moat is widening rather than holding.
8. What to Watch
Seven signals — each quarterly-observable in public disclosures — that tell an investor whether the moat is widening, holding, or fragmenting. All can be tracked without insider channels or paid databases.
The first moat signal to watch is Samsung Foundry segment revenue growth. If Samsung's foundry revenue returns to sustained double-digit growth with at least one large HPC customer publicly named, the wide-moat call should be downgraded to narrow within a quarter — before it ever shows up in TSMC's own margins.