Taiwan Semiconductor Manufacturing Company Ltd (NYSE: TSM) — Initiating Coverage POSITIVEScenario Fair Value$475

1 · Summary & Verdict

FY2025 revenue $122.9B (+37%); Q2-2026 posted $40.2B (+36% YoY) at 67.7% GM, and management raised FY26 growth guidance to 'slightly above 40%' with capex to $60-64B. Our Scenario Fair Value of $475 sits ~19% above the $398 price; we initiate Positive.

THE THREE RISKS THAT MATTER

2 · Investment Thesis

1. The N2 monopoly through 2028+ is the structural anchor

TSMC is the only volume producer of leading-edge foundry silicon that global fabless customers trust. N2 debuted at 3% of Q2-26 wafer revenue and ramps toward 15-20% by mid-2027. Samsung SF2 lacks anchor customers; Intel 18A is captive-plus-a-few; SMIC has no EUV access. TSMC entering 2028 with 80-85% of leading-edge foundry share is the base case. The N2 wafer ASP premium (~50% over N3E) and the compounding CoWoS packaging monopoly make every node ramp both a growth event and a margin event.

• N2 mix 3% → 15-20% by mid-2027 (each 100bps ≈ $1.7B annualized revenue)
• A16 (backside-power) late 2026; A14 in R&D for 2028 — no credible competitor at either

2. The Q2-26 raise underwrites a step-function 2026 — with capex to match

Guidance moved from ~30% to 'slightly above 40%' YoY growth (implied ~$175B) and capex jumped from $52-56B to $60-64B — the largest capex step-up since 2022. This is not aspirational: Q2 delivered 36% YoY at a 67.7% GM, Q3 is guided +37% YoY at 65-67% GM, and CoWoS advanced packaging capacity is booked through mid-2028. When a supplier with 100% share of its addressable market raises capex 50%, it is spending against contracted demand.

• FY26E revenue $174.5B (+42%), FY27E $216.4B (+24%) at 65.5% / 63.0% GM
• Q3-26 guide: $44.6-45.8B ± 65-67% GM ± 56-58% OM — record on every dimension

3. The AI cycle re-mixed TSMC toward HPC — the growth and margin implications compound

HPC was 43% of revenue in Q4-2022; it is 66% today, growing +20% QoQ in Q2-26 alone. Every hyperscaler capex dollar that flows into GPUs, custom accelerators, and networking ASICs eventually reaches TSMC's fabs. The mix shift is margin-accretive (HPC customers pay for leading-edge node access at a premium), and the shift is structural — hyperscaler capex is guided at $500-800B cumulatively through 2028 across the major five buyers.

• HPC platform +20% QoQ in Q2-26 — fastest of any segment
• CoWoS capacity booked through mid-2028; 2x expansion planned for 2027

4. The valuation tension is real but the setup is asymmetric to the upside

At $398, TSMC trades at 2.1σ above its own 8-year P/E band on TTM (38.4x vs 24.3x mean) but only ~19-23x forward on FY26E EPS. Perpetuity DCF at WACC 10.0% / g 3.5% gives $374/ADR — the stock is not cheap on cash-flow-multiple discipline. Our $475 fair value is scenario-blended (25/55/20 on $625/$475/$285). Positive means +19% upside vs the base case with a bear-case floor at $285 (-28%) that would require Taiwan escalation, AI digestion, or both. View shifts Neutral above ~$460 (+15% priced in); shifts Cautious above ~$540 (bull case fully priced).

• SFV $475 = 23.1x FY27E ADR EPS $20.57 — reasonable for the monopoly
• DCF cross-check $374 (WACC 10.0%, g 3.5%) sets a hard-value anchor

3 · Financial Analysis

TSMC’s financial profile is uncommon among AI-cycle beneficiaries: high growth, high margin, high capital intensity, high FCF conversion when capex normalizes, all supported by a fortress balance sheet ($82B cash+STI vs $33B debt at YE25, essentially net cash on a functional basis).

Revenue trajectory. FY22 $75.9B → FY23 $69.3B (-9%, the semi digestion) → FY24 $90.1B (+30%) → FY25 $122.9B (+37%) → FY26E $174.5B (+42%, guided “slightly above 40%”) → FY27E $216.4B (+24%) → FY28E $257.5B (+19%) → FY30E $331.7B. The two-year 2024-25 recovery was AI-driven, and the 2026 acceleration is contracted, not modeled: Q2-26 delivered $40.2B (+36% YoY), Q3-26 guided $44.6-45.8B (+37% YoY at midpoint), and the FY26 implied H2 run-rate requires ~$90B — matching H1 exactly and consistent with the +40% guide.

Gross margin structure. FY23 54.4% → FY24 56.1% → FY25 58.0% → FY26E 65.5% (blend of Q1 66.0%, Q2 67.7%, H2 65% mid) → FY27E 63.0% (N2 mix rises 3%→15%, ~2-3ppt dilution) → FY28E 62.0% (overseas fab drag cumulates) → recovering to 63.0% by FY29E → 62.5% FY30E. The market-cap-shattering Q2-26 print of 67.7% GM is well above management’s historical “53%+ long-term goal” guidance — pricing power at N3 and CoWoS is offsetting the N2 ramp dilution + overseas fab drag + FX. Even a normalized 62% GM keeps operating margin above 55% — vs peer semi averages of 30-40% and even ASML’s ~35%.

Operating margin ramp. FY23 42.6% → FY24 45.7% → FY25 49.1% → FY26E 58.1% → FY27E 56.1% → FY28E 55.4% → FY30E 56.5%. Opex leverage is real: R&D grows from $8.0B (FY25) to $14.5B (FY30) but scales as % of revenue from 6.5% → 4.4% given the top-line ramp.

Net income & EPS. FY25 $47.5B ($9.15 ADR EPS) → FY26E $89.7B ($17.30) → FY27E $106.5B ($20.57) → FY30E $164.0B ($31.85). ADR EPS compounds at ~29% CAGR through FY30E on the base case.

Capex intensity and FCF. Capex FY23 $30.5B → FY24 $30.0B → FY25 $40.9B → FY26E $62B (guide $60-64B) → FY27E $65B → FY28E $65B → declining to $55B by FY30E. FCF was suppressed 2022-24 at ~$1-15B/yr; inflected to $34B FY25; ramps to $52B FY26E → $74B FY27E → $102B FY28E → $163B FY30E as revenue outruns capex. Capital return is stable and rising: dividend per ADR $2.65 (FY25) → $3.05 (FY26E) → $5.80 (FY30E), plus modest buybacks $2-6B/yr rising over the plan.

Balance sheet. Net cash of ~$49B at YE25, growing to ~$400B+ by FY30E as FCF stacks. Debt is $33B of Taiwanese local-currency term financing at 1-2% — structurally cheap and stable. WACC computes at 10.0% assuming 95/5 equity/debt weighting, cost of equity 10.4% (Rf 4.4% + 1.15β × 4.75% ERP + 50bps Taiwan risk premium), and after-tax cost of debt 2.6% at a 14% effective tax rate.

H1-2026 actuals (already booked): Q1 $35.8B (GM 66.0%, ADR EPS $3.85); Q2 $40.2B (GM 67.7%, ADR EPS $4.31). H1 ADR EPS $8.16 (+78% YoY) — half of our FY26E $17.30 estimate is already in the books.

4 · Projection Assumptions

Our base case models the AI supercycle sustaining through FY27, decelerating to a still-elevated growth pace through FY28-30, with margins compressing modestly from the Q2-26 record then recovering as N2 productivity matures and overseas fab dilution stabilizes.

Revenue path. FY25A $122.9B → FY26E $174.5B (+42%) → FY27E $216.4B (+24%) → FY28E $257.5B (+19%) → FY29E $296.1B (+15%) → FY30E $331.7B (+12%). The step-function 2026 growth is Q2-guided; the 2027 deceleration to +24% reflects a normal industry maturation curve on top of a much higher base. 2028-30 tapering assumes AI capex intensity normalizes and TSMC’s growth becomes a share-preservation + ASP-lift + volume-modest story rather than the current all-in ramp.

Gross margin walk. FY25A 58.0% → FY26E 65.5% (Q2 print 67.7%, guided range 65-67% for Q3, H2 dilution from N2 initial ramp) → FY27E 63.0% (N2 mix +12ppt) → FY28E 62.0% (overseas fab drag $8-12B cumulative through 2028) → recovering to 63.0% FY29E → 62.5% FY30E. Even the FY28 trough of 62.0% remains ~9ppt above the FY23 trough of 54.4% — the structural margin re-basing from AI-cycle pricing power is durable.

Operating margin, net income, EPS. OM 49.1% FY25A → 58.1% FY26E → 55-56% steady FY27-30E; net margin 38.7% FY25A → 51.4% FY26E on tax discipline (14% Taiwan effective rate) and interest income from a growing cash pile. ADR EPS: $9.15 FY25A → $17.30 FY26E → $20.57 FY27E → $24.19 FY28E → $28.52 FY29E → $31.85 FY30E — a 28% CAGR through FY30E on the base case.

Cash flow. CFO FY26E $114B, capex $62B → FCF $52B (30% FCF margin). CFO FY27E $139B, capex $65B → FCF $74B. FCF ramps to $102B FY28E, $136B FY29E, $163B FY30E as capex plateaus and CFO scales.

Capex. FY26E $60-64B (guided; $62B midpoint) is a +52% YoY step-up on the FY25A $40.9B base — the largest capex increase since 2022. FY27E holds ~$65B as N2 volume ramps and Arizona Fab 21 Phase 2 comes online. FY28E $65B, then normalizing to $55B by FY30E as the AI-cycle capacity buildout matures. 70-80% of capex remains advanced process, 10% specialty, 10-20% advanced packaging + mask/interconnect.

Key modeling assumptions. (1) N2 mix ramps 3% (Q2-26) → 8% (Q4-26E) → 15% (mid-2027E) → 25% (2028E); (2) HPC platform mix rises to 68% (FY27E) → 72% (FY30E); (3) Advanced packaging revenue $12B (FY25A) → $22B (FY26E) → $56B (FY30E), doubling roughly every 18 months through 2027 then decelerating; (4) FX at NT$/US$ 31.5 (2026E) rising to 32 (2030E — modest NT$ depreciation baseline); (5) US corporate tax rate stable at 14% effective; (6) Diluted ADR count roughly stable at ~5,180M through FY30E (buybacks offset SBC).

5 · Scenario Analysis

We model three explicit five-year paths through FY2030E, weighted 25% bull / 55% base / 20% bear. The blend produces a Scenario-Weighted Fair Value of ~$474 which we publish as a $475 Scenario Fair Value (the base-case anchor). The spread is wide — $285 to $625 — because the AI capex cycle can extend or digest inside four quarters, and Taiwan geopolitical risk carries a genuine step-function tail.

Base case (55%, $475). The guided 2026 lands at $175B; 2027 grows 24% on the N2 mix ramp plus committed CoWoS capacity; 2028-30 decelerates to +19%/+15%/+12% as the AI capex intensity normalizes at a still-elevated level. Gross margin walks 65.5% → 63% (N2 dilution) → 62.5% (steady). ADR EPS $17.30 → $31.85 (28% CAGR). $475 equates to 23.1x FY27E ADR EPS or 19.5x FY28E — reasonable for a monopoly with 25%+ EPS growth. The base case is management’s contracted plan minus the 2028 optionality on CoWoS expansion + A16 pull-forward.

Bull case (25%, $625). The AI capex cycle extends deep into 2028: Rubin ramps outperform, custom hyperscaler ASIC volume beats, CoWoS capacity doubles again in 2027 as planned and again in 2028, N2/A16 pricing sustains at 50%+ premium, Samsung/Intel fail entirely at the leading edge. Revenue grows +48%/+32%/+24%/+18%/+15%, hitting $500B by FY30E with gross margin holding 66-67%. ADR EPS reaches $42 by FY30E on the bull path. $625 is ~26x bull FY27E ADR EPS $23.66 — demanding, but this is the scenario in which TSMC posts three consecutive +25% years and the market treats the AI-cycle re-basing as permanent. Q4-26 CoWoS capacity announcement + strong initial 2027 guide are the first confirmations.

Bear case (20%, $285). The 2027 AI capex digestion arrives: hyperscaler capex plateaus, NVIDIA order revisions cascade through the ASIC customers, HPC revenue growth compresses to single digits. Revenue: +35%/+8%/-5%/+8%/+10% — 2028 posts an outright decline as the industry digests the 2026-27 capacity build. Gross margin troughs at 55% as under-utilized N2 lines carry high depreciation and pricing concessions emerge. ADR EPS bottoms near $12.70 in FY28E before recovering to $16.31 FY30E. Overlay: incremental Taiwan risk premium expansion adds another 100-200bps to WACC. $285 is ~22x recovered FY30E bear EPS — bear-case multiples stay reasonable because the monopoly survives every downcycle; it is the entry price that determines whether the holder does. Note the asymmetry at a $398 entry: -28% to the bear anchor vs +57% to the bull.

What moves us between scenarios. Watch, in order: (1) the Q3-26 print and Q4-26 guide (bull needs Q4 revenue to hold >$45B run-rate); (2) mid-Jan 2027 FY-26 results and initial 2027 guide (bull needs +25%+ revenue growth guide, capex range $65-75B); (3) NVIDIA/AMD/hyperscaler capex guidance revisions — the ultimate upstream signal; (4) N2 mix trajectory (bull path requires 15%+ mix by mid-2027); (5) Taiwan geopolitical headlines in either direction. We would re-weight bull above 25% on any two of the first four confirming.

6 · Valuation — and the Lab

We anchor our Scenario Fair Value of $475 on the base case of an explicitly modelled bull/base/bear blend (25/55/20 weights compute to $474), cross-checked against an intrinsic DCF that deliberately builds in Taiwan geopolitical risk and trading comparables that place TSMC inside the AI-infrastructure semi complex on forward earnings. Precedent transactions are not a meaningful method — nothing at TSMC’s scale and strategic profile has ever traded, nor could it under any regulatory framework.

DCF — perpetuity on normalized FCF ($374/ADR). We discount four years of explicit unlevered FCF (FY27-30E: $70B / $85B / $110B / $135B) at a 10.0% WACC (cost of equity 10.4% = 4.4% risk-free + 1.15β × 4.75% ERP + 50bps Taiwan risk premium; ~95/5 equity/debt weighting), then apply Gordon growth on a normalized $145B FCF at g = 3.5% — a terminal growth rate consistent with global semi CAGR expectations. PV of explicit FCF: $308B. PV of terminal value: $1,577B (83.6% of EV — high, honestly disclosed: this is a duration asset). Enterprise value $1,886B, plus net cash $55B, over 5,186M diluted ADRs → $374 per ADR.

DCF sensitivity — WACC + growth grid. Across the full 5×5 WACC × g grid (9.0-11.0% × 2.5-4.5%), per-ADR value spans roughly $310 to $500 (the interactive heatmap in the Valuation Lab below maps the full envelope). The base case at 10.0% / 3.5% sits at $374 — meaningfully below the $398 spot. Reducing the Taiwan geopolitical premium to 25bps (WACC 9.75%) and raising terminal growth to 4.0% pushes DCF to ~$430 — near the spot but not the $475 SFV. The DCF is the honest hard-value floor of this call; the $475 SFV is a scenario/multiple-based number that assumes AI capex sustains.

Trading comparables. The natural peer set: foundry peers (GFS + INTC + Samsung), the customer-anchor NVDA + AVGO, and tool-monopoly ASML + AMAT. Forward P/E on ETL-pulled multiples: GFS 23x, ASML 31x, AMAT 37x, NVDA 16x (base year), AVGO 42x, INTC 60x. Median ~30x — TSMC at ~19-23x forward on FY26E ADR EPS trades at a discount to the AI-infrastructure semi complex despite the highest-quality moat and 40%+ near-term growth. Applying 18-25x to our FY27E ADR EPS $20.57 brackets $370-$514; EV/EBITDA on our $175B FY27E EBITDA at 10-15x (bridged over $55B net cash) brackets $340-$520.

Football field & the blend. Method ranges: DCF perpetuity $330-$500 (base $374); Forward P/E comps $370-$514; EV/EBITDA comps $340-$520; scenario range $285-$625. The scenario blend (25% × $625 + 55% × $475 + 20% × $285 = $474) is our published anchor: $475 Scenario Fair Value, sitting near the top of the DCF grid, mid-range of both comps methods, and at the base scenario by construction.

Street context. Post-Q2-26 print sell-side consensus clusters near $470 (range roughly $385-$550, with the top-end raised within days of the July 16 guidance revision). Our $475 sits at consensus. The differentiation of this report is not the number but the explicit accounting of which future the number buys — a Positive rating grounded in scenario probability weighting rather than DCF-only discipline.

Rating logic. Kaamos View bands anchor on scenario fair value vs price: Positive above +15% upside, Neutral between -5% and +15%, Cautious below -5%. At +19% ($475 vs $398.37), TSM is Positive. The View shifts Neutral above ~$460 (upside falls below 15%) and Cautious above ~$540 (price exceeds our SFV; bull case pre-paid).

Which multiples to actually trust for TSM

P/E (TTM) — primary. The street’s anchor for a foundry monopoly with durable AI-cycle earnings and rising node ASPs. Right now it reads 38.4x TTM against a ~24.3x 8-year mean (+2.1σ) — well above the historical band. The honest caveat: TTM earnings lag a violent inflection; on our FY26E ADR EPS $17.30 the same price is 23x, on FY27E it’s 19x. The z-score is telling you how much of the AI re-basing is pre-paid, not that earnings are about to disappoint.

EV/EBITDA (TTM) — primary. Controls for TSMC’s ~$50B net cash + capital structure and reads the operating inflection fastest. 18.2x vs a 12.7x mean (+1.5σ) is the least-alarmed multiple on the board — because EBITDA is already re-basing on the H1-26 step-up. The multiple most consistent with a Positive view: elevated but explainable by the underlying earnings power.

P/Sales (TTM) — secondary. The purest read on how much monopoly pricing power is being capitalized, immune to margin flattery: 18.0x vs a 9.4x mean (+2.9σ) is the honest “expensive” signal on the tape. Useful precisely because gross-margin expansion cannot flatter it. Some P/S expansion is deserved as GM structurally re-rated from 54% to 62%+; +2.9σ suggests more than “some” is priced.

Excluded. P/B (13x, +3.0σ): a capital-intensive foundry where book value tracks PP&E accounting and the AI-era stock re-rating outruns the book denominator — not a valid anchor. P/CF and P/FCF: customer prepayments and CoWoS booking swings whipsaw cash flow at quarterly granularity, and the FY26 capex step-up (+50% YoY) suppresses trailing FCF at a moment when the business is inflecting; our DCF normalizes through-cycle FCF instead. Dividend yield (0.65% vs 1.72% mean): the AI-year stock re-rating (+62% YoY) mechanically compresses yield below the historical mean; TSMC’s dividend policy is on the rise (per-ADR growing at ~15%/yr), but yield is a payout-vs-price artifact, not a valuation signal at the current price.

Net read: TSMC prices between +1.5σ (EV/EBITDA) and +2.9σ (P/S) above its own history, with the earnings-based anchor at +2.1σ. This is meaningfully cheaper on cycle-adjusted multiples than ASML (+2.5σ / +0.5σ / +2.9σ was the ASML read) despite comparable monopoly economics and 40%+ near-term growth. The gap is the Taiwan geopolitical discount — real, unhedgeable, and priced into every reasonable framework. That is precisely a Positive setup: right company, discounted price, tail risk explicit.

7 · Company

Business. Taiwan Semiconductor Manufacturing Company Ltd (Hsinchu, Taiwan; ~83,000 FTEs) is the world’s largest and most technically advanced dedicated foundry — the pure-play contract manufacturer that fabricates leading-edge silicon for every major fabless chip designer on earth. TSMC has effectively 100% share of the sub-3nm foundry market and ~65% of the total foundry market. Revenue splits along two axes: by technology generation — leading edge (7nm and below, 77% of Q2-26 wafer revenue), middle nodes (16nm/28nm, ~14%), and mature/specialty (40nm+, ~9%); and by platform end-market — High-Performance Computing (66% in Q2-26), Smartphone (22%), Automotive + IoT (~10%), and DCE + specialty (~2%). Advanced packaging (CoWoS/SoIC/InFO), booked through mid-2028, is the AI cycle’s binding capacity constraint. Customers include Apple, NVIDIA, AMD, Qualcomm, Broadcom, MediaTek, Marvell — the entire leading-edge fabless universe.

History. Founded 1987 by Morris Chang (ex-TI, backed by the Taiwan government and Philips) — the invention of the pure-play foundry model that decoupled chip design from manufacturing. IPO’d Taipei 1994, NYSE ADR 1997 (1 ADR = 5 common). TSMC has never lost a leading-edge process node race: 130 → 90 → 65 → 45 → 28 → 16 → 7 → 5 → 3 → 2nm all shipped on schedule. FY2025 closed at $122.9B revenue (+37%) and $47.5B net income on $40.9B capex. Q2-2026 (reported July 16, 2026) delivered $40.2B revenue (+36% YoY) at a 67.7% gross margin — a company record on both — and management raised full-year 2026 revenue growth guidance from ~30% to ‘slightly above 40%’ and lifted 2026 capex to $60-64B. C.C. Wei (Chairman & CEO since June 2024) owns the node cadence, the capex program, and the +$8B July raise. CFO Wendell Huang architects the gross-margin bridge — the print at 67.7% GM re-anchors the multi-year margin frame well above the historical “53%+ long-term” guidance.

Business mix & divisional economics

8 · Risks to the Target

TSMC’s risk stack is dominated by one tail (Taiwan geopolitics) and one cycle (AI capex sustainability); the operational risks (customer concentration, execution, overseas fab drag) are second-order and manageable.

1. Taiwan geopolitical risk (severity: extreme, undiversifiable — the defining tail). 85-90% of leading-edge capacity remains in Taiwan through 2028 even with Arizona, Kumamoto, and Dresden ramping. A cross-strait escalation would be an existential-scale event for the global semiconductor supply chain — and TSMC’s stock is the single most direct financial expression of that risk. This risk is unhedgeable, undiversifiable, and the perennial reason TSMC trades at a structural discount to what its fundamentals would otherwise justify. Our WACC includes an explicit ~50-100bps premium for this; the market prices some but not all of it.

2. AI capex cycle digestion in 2027 (high). The +40% FY26 guide assumes hyperscaler AI capex sustains at current levels. Hyperscaler capex guides $500-800B cumulatively through 2028 across Microsoft, Meta, Alphabet, Amazon; NVIDIA’s outlook, AMD’s MI400 ramp, and hyperscaler custom silicon roadmaps are unprecedented. If 2027 orders pause (analogous to 2023’s semi digestion), leading-edge utilization slips from 90-95% to 70-80% and gross margin compresses 400-600bps. The stock has re-rated 60% on this cycle; a pause compresses the multiple as fast as it expanded — the 2024 Nvidia GB200 delay (-15% in days) is a precedent for how fast that plays.

3. Customer concentration + Apple/NVIDIA revenue dependence (high, structural). Apple + NVIDIA together are ~40% of revenue; top-10 customers ~70%. An NVIDIA order revision or an Apple insourcing surprise (perennially theorized, never delivered) would move consensus by 5-10%. Countervailing: switching cost at the leading edge is effectively infinite — an N2 tape-out is a 12-18 month, $500M+ commitment.

4. US export controls and China revenue (moderate). China is 9% of revenue (down from 22% in 2020). Further tightening on mature-node exports would hit ~$10-15B of low-margin revenue — meaningful but not existential. Conversely, retaliatory Chinese IP or rare-earth restrictions on TSMC could hit supply chain (some rare earths still Chinese-sourced).

5. Overseas fab margin drag (moderate). Arizona/Kumamoto/Dresden add strategic optionality but ~$8-12B of cumulative gross margin drag through 2028 vs an all-Taiwan cost base. CHIPS Act subsidies (received/pending: ~$6.6B) offset some, but the operational cost delta is real. Political enforcement (subsidy clawbacks, forced-ownership requirements) mitigates but does not eliminate.

6. N2/A16 ramp execution (moderate). Every node ramp has execution risk; TSMC has an unbroken track record but N2 (first GAA/nanosheet), A16 (backside power delivery), and A14 (novel materials) are each first-of-kind. Yield below plan would compress FY27-28 GM by 100-300bps. Q3-26 GM guide of 65-67% (down from Q2-26 67.7%) is the first read on N2 dilution — a meaningful undershoot would send the stock down.

7. Advanced packaging bottleneck (moderate — this is TSMC’s binding constraint). CoWoS capacity is booked through mid-2028; if TSMC cannot double capacity again in 2027 as planned, NVIDIA Rubin shipments push out — capping TSMC’s own upside on HPC platform revenue.

8. FX / NT$ strength (low-moderate). ~100% of revenue is NT$-invoiced (USD-translated for reporting); a 5% NT$ appreciation costs ~2ppt of reported gross margin. NT$ has strengthened from 32 → 30 vs USD across 2024-26 — some of this is now in the base.

9. Regulatory / antitrust (low). No material antitrust exposure. Taiwan-Netherlands-Japan-US export-control coordination is the closest thing to regulatory risk.

10. Valuation / expectations (moderate market risk, not business risk). At $398 the stock trades 2.1σ above the 8-year TTM P/E band; forward P/E is a more benign 19-23x. Multi-year forward valuation is defensible but leaves less cushion than a 2σ read implies. Any earnings miss or negative capex-guidance revision would compress the multiple fast.

Catalysts to watch

• Q3 2026 earnings (mid-Oct) — first quarter at the $45B run-rate; a beat validates the >40% FY guide and pushes consensus to the top end
• FY 2026 results + 2027 initial guide (mid-Jan 2027) — the single most important date on the calendar; initial 2027 growth guide and capex range set consensus
• N2 revenue mix ramp — 3% Q2-26 → 5-8% by Q4-26 → 15-20% by mid-2027; each 100bps ≈ $1.7B annualized revenue
• CoWoS capacity announcements — each incremental capacity print unblocks Nvidia/AMD/hyperscaler ASIC shipments and lifts the HPC platform ceiling
• Apple N2 launch (Sept 2026 iPhone 18) — consumer-visible validation of the leading-edge node
• Geopolitical developments — either direction: de-escalation compresses the risk premium; escalation is a step-function drawdown

Upcoming events

• 2026-10-16 — Q3 2026 earnings (approx.) (First print at the $45B run-rate; Q4-26 guide is where the FY26 top-end debate resolves)
• 2027-01-15 — Q4 + FY 2026 results (approx.) (Initial 2027 revenue growth guide + 2027 capex range — the single biggest catalyst on the calendar)
• 2026-09-08 — Apple iPhone 18 launch (approx.) (First mass-market N2 product; consumer-visible validation of the node ramp)