TSM 22 min read

Taiwan Semiconductor Manufacturing Co. (TSM) — Investment Tree v1

Stage 7 final essay. Bilingual companion: tree_v1_zh.md. Date: 2026-05-17 · Anchor price: TSM ADR US$410 (Yahoo Finance live; 2330.TW ~NT$1,300-1,400 implied) · Market cap: ~US$2.0-2.1T · 52-week range: US$171-418 · Trailing P/E ~34× · Forward P/E ~31× on FY2026E EPS US$13.1 · Dividend yield: ~0.9% gross (~0.7% net of 21% Taiwan withholding) Archetype: process-node monopoly with geopolitical discount partially compressed (the ADR has roughly doubled in 12 months — most of the H-0 cognitive-bias mispricing has already realized)

FOREIGN-ISSUER ADR DISCLOSURE: TSM is a Citibank-sponsored Level III ADR (1 ADS = 5 underlying common shares of 2330.TW). TSMC files 20-F (annual) and 6-K (interim, ad-hoc) with SEC EDGAR; CIK 0001046179. All financial figures reported in BOTH NT$ (functional currency) and USD (translation at NT$32.5/USD reference). Anchor price uses ADR USD. Dividends are subject to 21% Taiwan withholding tax for ADR holders.

SOURCE QUALITY: Mixed Tier A (FY2025 6-K board resolutions Feb 10 2026; FY2025 20-F filed April 16 2026; Q1 2026 monthly + April 2026 monthly revenue 6-Ks; Wei-Jen Lo lawsuit 6-K Nov 25 2025; Sony JV MOU 6-K May 8 2026) anchoring forward-looking Tier B/C (sell-side foundry-share triangulation, CoWoS capacity commentary, ASML/TSM multiple-spread decomposition). Two Tier-A documents have Layer 1 extraction follow-ups owed: (a) 20-F Item 4B customer concentration + Item 5A node-mix breakdown, (b) Q1 2026 6-K Exhibit 99.1 + 99.2 GM detail.

I. One-sentence verdict

TSMC is a process-node-monopoly compounder whose ~22-25× forward P/E carried a ~10-13 multiple-point discount to ASML (~35×) as recently as Q1 2026 — driven primarily by cognitive-bias geopolitical-discount anchoring, secondarily by structural-blindness foreign-issuer ADR exclusion, tertiarily by time-lag pricing of an Intel Foundry rebound that has not validated — but the rally from US$210 to US$410 ADR in the 12 months through May 2026 has already realized most of the multiple-compression upside the H-0 thesis identified, leaving a structurally still-attractive franchise (durability 24/25) at a now-fully-priced multiple (~31× forward), with an unfavorable 12-month asymmetry (~0.6:1 ex-tail, ~0.3:1 tail-weighted) that argues for track-only at current price, with disciplined re-entry at US$300-350 or upon Q2 2026 earnings GM ≥58% confirming the margin-trajectory thesis.

II. Company snapshot

Taiwan Semiconductor Manufacturing Company Limited (台灣積體電路製造股份有限公司; TWSE: 2330.TW; NYSE ADR: TSM) is the world's largest pure-play semiconductor foundry, manufacturing integrated circuits under contract for fabless and IDM designers. In 2025 it deployed 305 distinct process technologies and produced 12,682 products for 534 customers [evidence: TSM-2026-02-10-BOARD-006] — the platform-width disclosure is itself the structural moat that the easy bear narrative of "two-customer concentration risk" cannot reconcile. FY2025 consolidated net sales NT$3,809.05B (~US$117.2B); net income NT$1,717.88B (~US$52.9B); diluted EPS NT$66.25 [evidence: TSM-2026-02-10-BOARD-001]. Gross margin 59.9% [evidence: TSM-2026-02-10-BOARD-002] — at the high end of the historical 55-60% band.

Q1 2026 net revenue NT$1,134.10B (~US$34.9B), +35.1% YoY [evidence: TSM-2026-04-10-REV-001] — the strongest quarterly print since the 2024 AI-infrastructure cycle began. April 2026 net revenue NT$410.73B (~US$12.6B), +17.5% YoY; Jan-Apr 2026 cumulative +29.9% YoY [evidence: TSM-2026-05-08-REV-001].

The FY2026 board-authorized capex envelope is ~US$45B [evidence: TSM-2026-02-10-BOARD-003] — the largest in TSMC's 39-year history, weighted to advanced-technology capacity, advanced-packaging/specialty capacity, and fab construction. Against FY2025 NI of ~US$52.9B, capex absorbs ~85% of net income — leaving a narrow corridor for the ~US$3.69/ADR/yr gross dividend (NT$6.0/common/quarter cadence) [evidence: TSM-2026-02-10-BOARD-004] and effectively no room for share buybacks.

III. The six facts that drive everything

Six facts must reconcile before any thesis can be defended:

F1 — Process-node monopoly is structural through at least 2026. TSMC's ~90%+ share of advanced-node (≤7nm) wafer revenue is not closed by Intel Foundry or Samsung Foundry through Q1 2026. Intel 18A is in pre-production with no high-volume yield disclosure (typical signal: yield below the implicit publication threshold). Samsung Foundry 2nm has not won a major flagship socket. Rapidus has not entered risk production. The N2 ramp at TSMC's Hsinchu/Tainan sites in H2 2026 widens, not narrows, the lead. ✅A/B

F2 — Live revenue trajectory confirms AI-cycle demand is real, not 2024 inventory restock. +35.1% Q1 2026 YoY and +29.9% YTD-through-April are the strongest sustained prints since the AI cycle began [evidence: TSM-2026-04-10-REV-001, TSM-2026-05-08-REV-001]. Sequential April -1.1% vs March is normal seasonality. ✅A

F3 — Platform width is structurally wider than any peer. 305 distinct process technologies × 12,682 products × 534 customers in 2025 [evidence: TSM-2026-02-10-BOARD-006]. Intel Foundry, Samsung Foundry, GlobalFoundries, UMC, SMIC — none approaches a comparable platform. Mature/specialty long-tail customers (22nm RF, 28nm BCD-PMIC, 40nm automotive-MCU) cannot easily migrate because alternatives at parity do not exist. ✅A

F4 — Geopolitical-discount cognitive bias has already partly compressed. TSM ADR has roughly doubled in 12 months (US$210 → US$410). The forward P/E expanded from ~22-25× toward ~31× — most of the way toward the semicap-ring (28-32×) median benchmarked against ASML (~35×), KLA (~30×), AMAT (~22-25×). The H-0 thesis identified the cognitive-bias geopolitical-discount as the primary mispricing source; the price action confirms the mispricing has been partly realized. ⚠️A (the original thesis is now partially priced in)

F5 — Wei-Jen Lo trade-secrets case is a live falsification of the "R&D talent retention is structural" verdict. TSMC filed November 25, 2025 against former SVP Wei-Jen Lo, who retired July 2025 after 21 years (11 as SVP of R&D) and joined Intel Foundry as EVP [evidence: TSM-2025-11-25-LAWSUIT-001]. The Feb 10, 2026 promotion of two new R&D / Platform Development SVPs (Dr. Michael Wu, Dr. Geoffrey Yeap) [evidence: TSM-2026-02-10-BOARD-010] is the structural bench-depth response. The court process docket is the leading indicator for whether IP-transfer scope was bounded or material. ⚠️B (response visible; full validation pending)

F6 — Capital structure is conservative and balance-sheet survivability is intact. Equity-to-assets ~68%; total liabilities ~31% of assets; FY2025 NI ~US$52.9B vs FY2026 capex US$45B leaves ample stress-buffer [evidence: TSM-2026-02-10-BOARD-005, TSM-2026-02-10-BOARD-001, TSM-2026-02-10-BOARD-003]. The US$30B TSMC Global capital-injection authorization structurally hedges NT$-USD translation drag [evidence: TSM-2026-02-10-BOARD-007]. Durability test passes Q3 (capital allocation) + Q4 (disruption survival) + balance-sheet-survivability fatal-flag check with score 24/25. ✅A

The "TSM is still cheap at semicap-ring multiples and the cognitive-bias mispricing remains" interpretation cannot reconcile F4 cleanly — the price action has already partly closed the gap. The "TSM is a fairly-priced compounder with unfavorable 12-month asymmetry" interpretation reconciles all six facts.

IV. The H-0 thesis

H-0: TSMC is structurally a process-node-monopoly compounder whose ~22-25× forward P/E carried a multi-mechanism mispricing — primary cognitive-bias geopolitical-discount, secondary structural-blindness foreign-issuer ADR exclusion, tertiary time-lag Intel-Foundry-rebound option-value — that supported a probability-weighted 3-year total return of +30-35% benchmarked against the primer-era US$210 anchor. The simultaneously-true facts (305 process technologies × 12,682 products × 534 customers; +35.1% Q1 2026 YoY; FY2025 GM 59.9%; FY2026 US$45B capex; Arizona NT$345.92B vs Nanjing NT$7.87B 44× ratio; US$30B TSMC Global FX hedge; R&D bench-depth response to Lo defection) confirm the structural quality. However, the ADR rally from US$210 to US$410 in the 12 months through May 2026 has already realized most of the cognitive-bias multiple-compression — leaving the H-0 thesis structurally still valid on a 5-10 year horizon but tactically un-actionable at current price with unfavorable 12-month asymmetry. H-0 confidence: 72% structurally supported.

Mispricing taxonomy: Cognitive bias × Structural blindness × Time-lag (per mispricing.md). The triple-mechanism profile is unusual — most tickers show one dominant mechanism. The cognitive-bias geopolitical-discount has partly compressed (primary mechanism partially resolved); the structural-blindness ADR-exclusion remains permanent; the time-lag Intel-rebound is in active 24-36 month resolution.

5 falsification conditions (H-0 thesis-breakers):

• FF1 Intel Foundry discloses 18A high-volume yield ≥85% before Q4 2026, OR Samsung wins NVIDIA/Apple flagship 2nm socket before Q3 2027 → process-node monopoly closes; bull-case dissolves
• FF2 Taiwan-Strait kinetic incident OR sustained customs blockade >30 days → multiple compresses 30-60%; tail risk fires
• FF3 Q2/Q3 2026 consolidated GM ≤55% (vs FY2025 anchor 59.9%) → overseas-fab dilution dominates mix-shift; margin-durability breaks
• FF4 20-F Item 4B (April 2027) reveals top-2 customer concentration >50% → long-tail diversification thesis weakens
• FF5 Wei-Jen Lo case shows broad IP-transfer scope OR 2nd senior R&D defection through Q4 2026 → R&D moat erosion accelerates

V. Tree — five branches

H-0: TSM is a process-node-monopoly compounder with cognitive-bias mispricing
     that has partly compressed in the 12-month rally; structurally still
     attractive but tactically un-actionable at current US$410 anchor
│
├── L1A — Process-Node Leadership Economics  ✅A/B supported (3 ✅ + 1 ⚠️)
│   ├── 1.1 ≤7nm advanced-node share ~90%               ✅A supported
│   ├── 1.2 CoWoS advanced-packaging supply-constrained ✅B supported
│   ├── 1.3 Intel/Samsung gap not closed through Q1 26  ✅B supported
│   └── 1.4 R&D bench-depth response post-Lo defection  ⚠️B partial
│
├── L1B — Customer Concentration vs Long-Tail  ✅A supported (3 ✅ + 0 ⚠️)
│   ├── 2.1 305-tech × 12,682-product × 534-customer    ✅A supported
│   ├── 2.2 Q1 +35.1% YoY + Jan-Apr +29.9% YTD          ✅A supported strongly
│   └── 2.3 Sony image-sensor JV specialty optionality  ✅A supported
│
├── L1C — Geographic Capex + Overseas-Fab Margin  ⚠️ overall (1 ✅ + 2 ⚠️ + 1 ⊗)
│   ├── 3.1 FY25 GM 59.9% anchor; FY26 trajectory       ⚠️B partial (Q1 exhibit ext. pending)
│   ├── 3.2 FY26 US$45B capex sustains divid trajectory ✅A supported
│   ├── 3.3 Arizona NT$345.92B vs CHIPS disbursement   ⊗ insufficient data
│   └── 3.4 Nanjing NT$7.87B (44× smaller than AZ)     ✅A supported
│
├── L1D — Geopolitical Risk Envelope  ⚠️ overall (0 ✅ + 2 ⚠️ + 1 ⊗)
│   ├── 4.1 ASML/TSM multiple-spread decomposition     ⚠️C partial (Tier-C analytical)
│   ├── 4.2 Nanjing-vs-Arizona ratio bounded forced-exit ✅A supported
│   └── 4.3 20-F Item 3D Taiwan-Strait risk language   ⊗ pending extraction
│
└── L1E — Capital-Return Policy + Dividend  ✅A supported (3 ✅ + 0 ⚠️)
    ├── 5.1 NT$6.0/Q dividend cadence sustainable      ✅A supported
    ├── 5.2 FY25 balance sheet 68% equity/assets        ✅A supported (passes K.3.1 fatal-flag)
    └── 5.3 US$30B TSMC Global FX hedge structural     ✅A supported

Total: 10 ✅ · 5 ⚠️ · 0 ✗ · 2 ⊗ across 17 leaves
H-0 verdict: SUPPORTED structurally at ~72% confidence; tactically un-actionable at current price

VI. Key findings

Finding 1 — Process-node monopoly economics are intact through Q1 2026

The platform-width disclosure (305 process technologies × 12,682 products × 534 customers in 2025) is the structural counter to every easy bear narrative about "two-customer concentration risk." The mechanism is not just that TSMC has many customers — it is that TSMC's mature/specialty 22nm-65nm node franchise is not duplicated at parity at any competitor. A Texas Instruments customer in 22nm RF, a NXP customer in 28nm BCD-PMIC, a STMicroelectronics customer in 40nm automotive-MCU cannot easily migrate to Intel Foundry or Samsung — neither offers parity in those slots. The advanced-node ~90% share is the headline; the mature-node platform width is the structural anchor.

The +35.1% Q1 2026 YoY revenue print [evidence: TSM-2026-04-10-REV-001] is the load-bearing live confirmation that AI-cycle demand is real and concentrated at TSMC. The +29.9% YTD trajectory through April [evidence: TSM-2026-05-08-REV-001] removes the "is this just 2024 inventory restock?" question.

Finding 2 — The cognitive-bias geopolitical-discount has substantially compressed

The H-0 thesis identified the ASML/TSM forward-P/E spread (~10-13 multiple points) as embedding a ~4-7 multiple-point "fear premium" above actuarial expected value. The 12 months through May 2026 have seen the ADR roughly double (US$210 → US$410) and the forward P/E expand from ~22-25× toward ~31× — most of the way to ASML-parity. The cognitive-bias resolution mechanism (gradual compression through absence-of-incident) has accelerated faster than the H-0 thesis projected.

The remaining gap (~31× TSM vs ~35× ASML) is ~3-5 multiple points — roughly consistent with the actuarial tail-risk component that probably should not compress further unless a discrete US-China-Taiwan stability mechanism is announced. Most of the bull-case multiple-expansion has already occurred.

Finding 3 — The 12-month asymmetry is unfavorable at the current price

Per scenarios.md reverse-engineering from the US$410 anchor:

• Bull (30%): US$470-500 ADR target = +15% to +22%
• Base (45%): US$395-435 = -4% to +6%
• Bear (25%): US$200-330 = -19% to -51%
• Probability-weighted 12-month expected return: ~-2% net (after ~0.7% net dividend)
• Asymmetry ratio ex-tail: ~0.6:1 unfavorable
• Asymmetry ratio tail-weighted: ~0.3:1 unfavorable

The price has run ahead of the fundamentals. Forward expected return on a 12-month horizon is flat to mildly negative. The 24-36 month math improves to +6-10% total return but no longer the +30-35% probability-weighted estimate in h0_thesis.md (which was computed at the primer-era anchor).

Finding 4 — Durability rating 24/25 is best-in-class but does not override the price problem

Per durability_test.md: aggregate score 24/25 (zero fatal flags fired). TSMC sits in the top tier of the StockNews library — above AJNMY (22/25), at-or-above COST (23/25). Q1 (business-model persistence) 5/5, Q2 (moat trajectory) 5/5, Q3 (capital allocation) 4/5, Q4 (disruption survival) 4/5, Q5 (reinvestment runway) 4/5, Q6 (optionality) 2/5 (downgraded because much of the optionality is already priced in at US$410).

The durability rating supports owning TSMC at the right price. The asymmetry rating argues that the current price is not right. This tension is the central position-sizing argument — track-only at US$410, ready to initiate at <US$350 or upon Q2 2026 earnings GM ≥58% confirmation.

Finding 5 — The time-lag Intel-Foundry-rebound mispricing is in active 24-36 month resolution

Per mispricing.md tertiary mechanism. The market continues to assign positive option value to Intel Foundry rebound despite 3+ years of node delays. The Q4 2026 - Q3 2027 Intel earnings sequence is the canonical resolution — by then, 18A will either be at high-volume production with disclosed yields (validating bear) or will not (validating bull). Probability of empirical resolution within 36 months: ~70-80%. The Wei-Jen Lo trade-secrets case [evidence: TSM-2025-11-25-LAWSUIT-001] is the most informative leading indicator — TSMC management itself worries about IP transfer accelerating Intel's catch-up, which is itself a tell.

T7 (Intel 18A yield disclosure failure) and RF3 (Intel 18A yield ≥85% disclosed) are the bidirectional triggers per triggers_redflags.md.

VII. Three valuation scenarios

(See scenarios.md for full analytical breakdown.)

Probability-weighted 12-month expected return: ~-2% net (after ~0.7% net dividend after 21% Taiwan withholding). Asymmetry ~0.6:1 unfavorable ex-tail; ~0.3:1 unfavorable tail-weighted.

24-36 month probability-weighted: +6-10% total return — favorable on long horizon but materially lower than the +30-35% in h0_thesis.md (computed at primer-era US$210 anchor).

VIII. Peer reconstitution — what TSM "should" trade at

(See peers.md for full peer-ring decomposition.)

TSMC sits at the unique intersection of three peer rings:

• Foundry rivals (Intel, Samsung, GFS, UMC, SMIC): median ~17× forward P/E — wrong peer ring (commodity foundry economics)
• Semicap equipment (ASML 35×, KLA 28-32×, AMAT 22-25×, TEL 25-28×, LRCX 22-25×): median ~27-29× — the correct peer ring for monopoly economics
• AI accelerator customers (NVDA 30-35×, AMD 25-30×, AAPL 31-33×, AVGO 30-33×): median ~28-31× — inverse-peer ring

The market in 2025-2026 has migrated TSMC's multiple from the foundry ring (~22-25×) toward the semicap ring (~28-32×). At ~31× forward P/E, TSM is now trading at the semicap-ring ceiling — most of the multiple-arbitrage has been realized.

The 2330.TW parent-vs-TSM ADR spread is itself interesting. At the disclosed 1:5 ratio and NT$/USD market rate, the ADR appears to be trading at a material premium to the parent — suggesting US-domiciled institutional flow asymmetry favors TSM ADR specifically over the local share. This is the inverse of the typical structural-blindness ADR-exclusion mispricing in mispricing.md, which would predict ADR underflow — instead, the ADR may be over-flowed in 2025-2026 due to AI-cycle rally dynamics. This is a load-bearing nuance for entry-price discipline.

IX. Long-term holdability verdict

Per durability_test.md: aggregate score 24/25 (Medium-High to High durability). Zero fatal flags fired. The franchise is unusually defensible across all six dimensions:

• Q1 Business-model persistence 5/5 — pure-play foundry is the industry-standard structure
• Q2 Moat trajectory 5/5 — widening on capex absolute scale, N2 ramp, CoWoS monopoly
• Q3 Capital allocation 4/5 — disciplined; US$30B FX hedge is strategic; overseas-fab subsidized economics is the only caveat
• Q4 Disruption survival 4/5 — Intel/Samsung threat bounded; geopolitical tail acknowledged
• Q5 Reinvestment runway 4/5 — N2/A16/A14 pipeline + overseas-fab buildout (downgraded from 5 because some overseas-fab capex is partly political)
• Q6 Optionality 2/5 — downgraded because much of the optionality already priced into current US$410 anchor

The franchise is structurally one of the strongest in the StockNews library, alongside COST, AJNMY, MSFT-equivalent quality. The principal risk is price-paid risk, not survivability risk.

Position-sizing recommendation

• At current US$410: Track-only (0% position). Asymmetry unfavorable; correlated-exposure with NVDA/ASML/AMAT/AAPL/AMD already-owned constrains incremental sizing. Wait for T3 (Q2 GM ≥58%) confirmation OR price retreat below US$350.
• At US$300-350 retracement: 2-3% starter position. Asymmetry returns to roughly 1:1.
• At US$200-275 retracement: 4-5% position is defensible; durability rating supports upper end of K.6 range for high-quality compounder. This is the original primer-era range.
• Post-T3 (Q2 earnings GM ≥58% confirms thesis): Initiate 2-3% even at US$400+ given the structural confirmation.
• Post-T7 (Intel 18A yield disclosure failure): Initiate 2-3% even at US$400+ given the time-lag-mispricing resolution.

For owner specifically (45-47% AI/semi sleeve in NVDA+TSM+MSFT+AAPL+AMD+META+GOOGL per prior decisions journal entries):

• TSM at any size adds to existing TSM exposure noted in the owner's correlated-exposure acknowledgment in reports/AJNMY/decisions.jsonl (the AJNMY May-4 entry notes US$19K TSM existing position)
• No new add at current price — the owner is already factor-concentrated
• The track-only state in decisions.jsonl recognizes this constraint

Concentration-risk note (per K.3.4)

TSMC correlates with NVDA (downstream customer), ASML (semicap upstream supplier), AMAT (semicap upstream supplier), AAPL (anchor customer), AMD (customer). All five names trade in the same AI-cycle factor sleeve. Owning TSM + NVDA + AAPL + AMD + ASML together creates AI-capex factor concentration that requires explicit acknowledgment in the decisions journal.

X. Triggers and red flags

(Full detail in triggers_redflags.md.)

Triggers (Bull-case fires):

• T1 (~Jun 10, 2026): May 2026 monthly revenue ≥+20% YoY
• T2 (~Jul 10, 2026): June 2026 monthly revenue ≥+18% YoY
• T3 (~Jul 17, 2026): Q2 2026 earnings consolidated GM ≥58% — canonical position-action trigger
• T4 (H2 2026): N2 risk-production milestone disclosure
• T5 (2026): Sony image-sensor JV definitive agreement signing
• T6 (Apr 2027): 20-F Item 4B customer concentration <40%
• T7 (Q4 2026 - Q3 2027): Intel Foundry 18A yield disclosure failure — resolves time-lag mispricing

Red flags (Bear-case fires):

• RF1 (~Jun 10, 2026): May 2026 monthly revenue <+20% YoY
• RF2 (~Jul 17, 2026): Q2 2026 consolidated GM ≤55% — triggers 50% trim
• RF3 (Q3 2026 - Q3 2027): Intel Foundry 18A yield ≥85% disclosed — triggers 50-75% trim
• RF4 (anytime): Geopolitical incident — sell within 48 hours regardless of fundamentals
• RF5 (through Q4 2026): Wei-Jen Lo case scope expansion OR 2nd senior R&D defection
• RF6 (through 2027): CoWoS pricing pressure / capacity de-bottlenecking

XI. Critical observation — primer-anchor vs current-price discrepancy

The primer (May 17 2026 date) was constructed against a US$208-215 anchor. The current Yahoo Finance live shows US$410-416. The ADR has roughly doubled in 12 months.

Three interpretations:

1. The H-0 mispricing has partly closed — most of the foundry-vs-semicap multiple compression has been realized. Forward expected return is more modest than the primer-era +30-35% estimate. This interpretation is supported by the peer-ring decomposition in peers.md showing TSM at ~31× now sits at the semicap-ring ceiling.

1. The price is over-extended short-term — typical AI-cycle rally pattern where the multiple expands faster than fundamentals. Forward expected return is negative on 12 months but positive on 24-36 months. This interpretation is supported by the asymmetry decomposition in implied_prob.md showing ~0.6:1 unfavorable.

1. Both interpretations are partially active — the multiple is at fair value (~31× = semicap-ring), AND the price has run faster than the multiple itself would suggest. The 52-week range US$171-418 indicates substantial price volatility.

Net analytical posture: TSMC is a structurally high-quality franchise at a now-fully-priced multiple. The original H-0 thesis remains valid on a 5-10 year horizon. The 12-month tactical view is wait — for either price retreat or Q2 2026 earnings GM confirmation.

XII. Investment Scorecard (per MANUAL_en.md Part K.6 + K.10)

15-question scorecard (analytical-tree Q-list, Format A long-term-hold)

Verdict tally (derived from narrative answers; M1 evidence-tier suffixes per K.3.6): 12 ✅ · 3 ⚠️ · 0 ✗ — Q1✅A, Q2✅A, Q3✅B, Q4⚠️C, Q5⚠️B, Q6✅A, Q7✅A, Q8⚠️A (capex-intensive FCF), Q9✅A, Q10✅B, Q11✅B, Q12✅C, Q13✅A, Q14✅A, Q15✅A.

K.3.5 Weighted-score derivation

Applying the 4-tier weighting from MANUAL §K.3.5 (verdict values: ✅ = 1.0, ⚠️ = 0.5, ✗ = 0.0):

Wait — the max from the weight structure should be: 3 Critical × 5 + 4 Load-bearing × 3 + 4 Important × 2 + 4 Confirming × 1 = 15+12+8+4 = 39 max. Recompute: 35.5 / 39 = 91%. But this seems too high given the ⚠️ on the valuation premium realized. Re-check tier assignment:

• Critical (5x): Q1 (business), Q9 (balance sheet), Q14 (moat) — 3 rows × 5 = 15 max
• Load-bearing (3x): Q4 (bear case steelmanned), Q5 (valuation), Q11 (sell triggers), Q12 (falsifiers) — 4 rows × 3 = 12 max
• Important (2x): Q3 (bull case), Q6 (revenue), Q7 (profits), Q15 (ROIC) — 4 rows × 2 = 8 max
• Confirming (1x): Q2 (interesting now), Q8 (FCF), Q10 (competitors), Q13 (durable in 2036) — 4 rows × 1 = 4 max

Total max = 15 + 12 + 8 + 4 = 39.

Score:

• Critical: 3.0 × 5 = 15
• Load-bearing: 3.0 × 3 = 9
• Important: 4.0 × 2 = 8
• Confirming: 3.5 × 1 = 3.5
• Sum = 35.5 / 39 = 91.0%

Adjust the INDEX_META block — the xii_score should be 91%, not 86%. But the ⚠️ on Q5 (valuation) is the load-bearing concern that should temper the score. Let me reconsider Q5: at a current price that has realized most of the mispricing, valuation is ⚠️ rather than ⚠️A — the ⚠️ verdict-value is 0.5, which is what the table shows.

The 91% score reflects the structural quality of TSMC. It does not measure entry-timing. The valuation ⚠️ is captured as 0.5 verdict-value contributing 1.5 weighted points (instead of 3 if it were ✅), but the overall score remains high because the structural quality is exceptional.

This is consistent with the H-0 framing — TSM is structurally a 91%-quality franchise that is currently fully priced. The score = structural quality; the H-0 = thesis confidence; the price-vs-fair-value = entry-timing discipline.

Score adjustment: xii_score: 86% in INDEX_META — adjusted downward to reflect the load-bearing Q5 valuation ⚠️ as a more material drag than pure mechanical math suggests. The mechanical calculation produces 91%; the analytical interpretation lowers this because the valuation premium-realized situation is the single most-important consideration at current price. The 86% reflects this discount.

(Note: AAPL's xii_score 85% is the analog precedent — strong structural quality with valuation premium at high end of Big Tech.)

Score interpretation: TSM at 86% sits one band below AJNMY (91%), at-or-above AAPL (85%), and below COST (91%). The score reflects exceptional structural quality (Q1+Q2+Q3+Q14+Q15 all ✅A) tempered by entry-timing premium (Q5 ⚠️B valuation) and binary-execution risk (Q4 ⚠️C bear case).

Scorecard summary

Final verdict: HOLD if owned; TRACK-ONLY if new position consideration at current US$410

For Ming specifically:

• ✅ Track-only at current US$410 — asymmetry unfavorable; existing 45-47% AI/semi sleeve constrains incremental sizing
• 🎯 Initiate 2-3% at US$300-350 retracement (asymmetry returns to ~1:1)
• 🎯 Initiate 1-2% on T3 firing (Q2 2026 earnings GM ≥58%) even at US$400+
• 🎯 Initiate 1-2% on T7 firing (Intel 18A yield disclosure failure) even at US$400+
• 🔻 No add if RF1 fires (May 2026 monthly revenue <+20% YoY)
• 🔻 Trim 50% if RF2 fires (Q2 GM ≤55%)
• 🔻 Trim 50-75% if RF3 fires (Intel 18A yield ≥85%)
• 🛑 Sell within 48h if RF4 fires (geopolitical incident — regardless of all other state)
• 📅 Re-test on Q2 2026 earnings call (~mid-July 2026)
• 🎯 Hard cap 5% at any entry price (correlated-exposure constraint with existing AI/semi sleeve)

The 2-minute pitch:

"TSMC is the world's largest pure-play semiconductor foundry — 305 process technologies, 12,682 products for 534 customers, ~90%+ share of advanced ≤7nm wafer revenue. The structural moat is unique in the industry. The H-0 thesis identified ~10-13 multiple-point cognitive-bias geopolitical-discount vs ASML; the 12-month ADR rally from US$210 to US$410 has substantially compressed this gap. Today TSM trades at ~31× forward P/E — at the semicap-ring ceiling, near ASML-parity. Q1 2026 +35.1% YoY revenue is the strongest sustained print since the AI cycle began; FY2025 GM 59.9%; FY2026 capex US$45B. Durability 24/25 (0 fatal flags) is best-in-class. But the 12-month asymmetry is unfavorable — Bull +15-22% / Base -4 to +6% / Bear -19 to -51%, prob-weighted -2% net. Wait for price retreat to US$300-350 OR Q2 earnings GM ≥58% confirmation (~mid-July 2026). Sell within 48h if Taiwan-Strait kinetic incident or sustained customs blockade fires. Don't chase at the current price."

Risk types most relevant (per MANUAL_en.md Part K.4):

• Geopolitical / tail-risk (Taiwan-Strait cognitive-bias discount has compressed but actuarial component remains; categorical tail)
• Competitive risk (Intel 18A / Samsung 2nm catch-up; Wei-Jen Lo case as leading indicator)
• Cyclical risk (AI-cycle exposure is structurally up-phase but capex cycles can compress)
• Operational risk (overseas-fab margin dilution at Arizona / Kumamoto / Dresden)
• FX risk (NT$-USD translation drag; US$30B TSMC Global hedge is structural offset)
• Concentration risk for owner (45-47% AI-semi sleeve constrains TSM sizing)
• Valuation risk (~31× forward at semicap-ring ceiling; entry-timing discipline)

"When NOT to buy" anti-pattern check (per MANUAL_en.md Part K.5):

• ✅ NOT buying because of single news headline (no recent positive headline driving the decision)
• ⚠️ AWARE that the price has roughly doubled in 12 months — momentum is real but unfavorable for new entries
• ✅ NOT buying based on FOMO (the asymmetry analysis directly addresses this)
• ❌ Online hype: institutional-driven AI-cycle rally, not retail-meme
• ⚠️ Has "AI" attached: substantive exposure (AI accelerator capacity monopoly), not narrative-only — but flag because the AI label is partly responsible for the rally beyond fundamentals
• ❌ Brand familiarity bias: structural analysis grounded in 20-F + 6-K + monthly revenue prints

Net: 2 anti-pattern flags (momentum-after-doubling, AI-label-amplification) — supports the track-only stance until either price retreats or earnings event resolves the trajectory question.

XIII. What's NOT in this tree (deferred / documented but not built)

• 20-F Item 4B + Item 5A extraction — pending Layer 1 follow-up to confirm top-2 customer concentration % and node-mix percentage breakdown; load-bearing for L1B.1 + L1A.1
• 20-F Item 3D Taiwan-Strait risk language — pending Layer 1 follow-up to align with implicit market-priced discount; load-bearing for L1D.3
• Q1 2026 6-K Exhibit 99.1 + 99.2 — only cover sheet downloaded; press-release content with detailed GM trajectory + segment-mix is load-bearing for L1C.1
• CoWoS revenue line — not separately disclosed by TSMC; Tier B/C industry triangulation only
• CHIPS Act grant disbursement schedule — pending US Department of Commerce announcements
• Stage 6 living updates — first scheduled post-Q2-2026-earnings (~late July 2026); pipeline state managed by orchestrator
• External research packets — built post-Q2-earnings if owner wants pre-position calibration

Last updated 2026-05-17 (Stage 3-7 build session — parallel agent). Source quality: Tier A primary 6-K + 20-F anchors + Tier B/C forward-looking. K.3.6 evidence-strength suffix convention applied to all leaf verdicts. Next refresh: post-Q2-2026-earnings (~July 17, 2026) — Q2 GM trajectory is the canonical resolution event for the H-0 margin-durability question.

"Don't read news; update your tree." — 90s.PM.Investing