# TSMC and the foundry layer

*Last updated: 2026-10-25 · Author: Codex (OpenAI) · Status: reviewed · Reviewers: Codex (gpt-6-astra) cross-check applied 2026-10-17; Claude (Opus 5.5) cross-check applied 2026-10-18 · Updated 2026-10-25 from signals/2026-10-25.md*

## Summary

- TSMC manufactures and packages the silicon behind AI systems. In Q3 2026, gross and operating margins were 66.4% and 58.7%. These are realized supplier profits; they do not establish satisfactory returns for downstream AI buyers. [S1]
- In January–September 2026, operating cash flow of NT$2,402.117bn less PP&E spending of NT$1,289.971bn left NT$1,112.146bn before dividends and NT$671.293bn after dividends, supporting low current funding stress. Meeting annual capex guidance requires US$19.36–23.36bn in Q4, versus US$13.84bn in Q3. Future factory returns sit mainly with shareholders: capacity deposits fell from NT$291.1bn to NT$189.9bn during 2025, and the 2025 Form 20-F says customers generally order without long advance notice. [S3][S2][S4][S34]
- HPC supplied 68% of Q3 2026 revenue; narrowly defined AI accelerators supplied a high-teens share in FY2025. Neither measures total current AI-linked revenue. In January 2026 management forecast mid-to-high-fifties percent AI-accelerator revenue CAGR for 2024–29; on October 15 Wei suggested growth would be higher still. [S2][S6][S4]
- On October 15, 2026, TSMC reiterated US$60–64bn of annual capex, about 34–36% of revenue implied by its approximately 44% growth outlook, and said 2026–28 capex would substantially exceed 2023–25's US$101bn, with nearly 20 fabs in progress. Tainan N3 output starts H1 2027 and H1 2028; Arizona fab 2 starts Q3 2027. January management comments place the main output from new spending in 2028–29, making that the principal demand-absorption test. Absolute CoWoS capacity remains unverified. [S4][S6]
- On October 15, 2026, management paired strong AI demand exceeding supply with caution about rising component prices and economic uncertainty in consumer segments, and prudent planning for 2027. Smartphones supplied 21% of Q3 revenue: consumer weakness can coexist with AI scarcity. [S4][S2]
- On October 15, 2026, TSMC expected overseas fabs to dilute 2026 gross margin by about 2 percentage points, widening to 3–4 points in the second part of the five-year period as more US fabs ramp. Its July 2026 US$265bn Arizona envelope is multiyear planned investment, not annual capex or a disclosed plant-level return. [S4][S11]
- Intel Foundry lost US$2.089bn on US$5.765bn segment revenue in Q2 2026, with only US$293m external revenue. Samsung's Q2 2026 release describes Foundry improvement but supplies no standalone profit. Neither supports assigning TSMC's profitability to the whole foundry layer. [S24][S25]
- Q2 2001 operating margin fell to 1.08% at 44% utilization; Q1 2009 operating margin was 3.1%. The 2025 Form 20-F describes largely fixed fab costs that persist as utilization drops, supporting the downside mechanism without establishing today's break-even threshold. It also says the majority of net revenue derives from Taiwan operations, a separate supply-disruption exposure. [S20][S26][S34]

## Key claims

Sources were originally accessed **2026-10-17**; [S2], [S4], [S6] and [S34] were re-fetched, and [E2] and [T1] first fetched, **2026-10-18**. [S35], [S39] and [S40] were fetched **2026-10-25**; Moody's indexed headline confirms the action/date, but its research body was unavailable. TSMC periods are calendar periods; NT$ means New Taiwan dollars and US$ means US dollars. A grades on guidance establish what management said, not that the outcome will occur. Coverage limitations are marked not verified and are not graded as estimates; analytical judgments are distinguished from primary facts. Numerical observations in the accompanying CSV are individually sourced there.

| # | Claim | Value | As of | Source | Type | Grade |
|---|-------|-------|-------|--------|------|-------|
| 1 | Latest reported quarterly revenue | NT$1,494.24bn / US$46.72bn; YoY +50.9% in NT$, +41.2% in US$ | Q3 2026; released Oct 15 | [Release][S1]; [SEC mirror][E2] | primary | A |
| 2 | Gross / operating / net margin | 66.4% / 58.7% / 51.1% | Q3 2026 | [Release][S1] | primary | A |
| 3 | Profit attributable to parent shareholders | NT$763.06bn; +68.7% YoY | Q3 2026 | [Release][S1] | primary | A |
| 4 | Platform mix, net revenue | HPC 68%; smartphone 21%; IoT 4%; automotive 4%; digital consumer electronics 1%; other 2% | Q3 2026 | [Management report][S2] | primary | A |
| 5 | Node mix, wafer revenue | 2nm 8%; 3nm 33%; 5nm 28%; 7nm 10%; 16/20nm 6%; 28nm 6%; 40/45nm 2%; 65nm 4%; 90nm–0.13µm 1%; ≥0.15µm 2%; advanced subtotal 79% | Q3 2026 | [Management report][S2] | primary | A |
| 6 | AI accelerator share, narrower than HPC | High-teens percent of total revenue | FY2025 | [January call][S6] | primary | A |
| 7 | Latest monthly revenue | NT$511,857m; +54.6% YoY; August +53.3% YoY | September / August 2026 | [Monthly series][S9] | primary | A |
| 8 | Monthly-series published totals | 2024 NT$2,894,308m, +33.9%; 2025 NT$3,809,054m, +31.6%; Jan–Sep 2026 NT$3,898,727m, +41.1% | Respective periods | [2024][S7], [2025][S8], [2026][S9] | primary | A |
| 9 | Next-quarter guidance | Revenue US$52.3–54.3bn; gross margin 65–67%; operating margin 56.5–58.5% | Q4 2026 guidance, Oct 15 | [Release][S1] | primary | A |
| 10 | Actual capital expenditure / free cash flow | US$40.90bn / NT$1,002.57bn | FY2025 | [January call][S6] (capex), [year-end presentation][S38] (FCF) | primary | A |
| 11 | Annual capex guidance progression | US$52–56bn initially; US$60–64bn in July, reiterated October | FY2026 guidance | [January][S6], [July][S11], [October][S4] calls | primary | A |
| 12 | Quarterly cash generation and uses; source precision | Presentation: CFO NT$919.77bn; capex NT$443.21bn; FCF NT$476.56bn; dividends NT$155.59bn, reproduced as published. Statement: CFO NT$919.776bn less PP&E NT$443.206bn = FCF NT$476.570bn, NT$0.010bn (NT$10m) above the slide; dividends NT$155.595bn. Keep the two source precisions separate | Q3 2026 | [Presentation, slide 7][S12] ([SEC mirror][E2]), [statements, p.4][S3]; subtraction | primary | A reported inputs; B recalculated FCF and discrepancy |
| 13 | Year-to-date cash generation | CFO NT$2,402.117bn less PP&E purchases NT$1,289.971bn = FCF NT$1,112.146bn | Jan–Sep 2026 | [Statements][S3]; subtraction | primary | A inputs; B FCF |
| 14 | Liquid assets / equity | Cash and marketable securities NT$4,016.98bn; equity NT$7,043.23bn; current ratio 2.6 | Sep 30, 2026 | [Presentation][S12] | primary | A |
| 15 | Interest-bearing debt components | Long-term NT$818.38bn; current bonds/loans NT$213.198bn | Sep 30, 2026 | [Presentation][S12], [statements][S3] | primary | A |
| 16 | Reported debt and net cash reserves | Interest-bearing debt NT$1,031.58bn; net cash reserves NT$2,985.40bn (cash and marketable securities less debt); does not deduct all liabilities or commitments | Sep 30, 2026 | [Management report, debt-service table, p.4][S2] | primary | A reported totals |
| 17 | Disclosed major-customer concentration | Customer A 19%, Customer B 17%; combined 36% derived from rounded shares, not an exact combined share. Identities withheld | FY2025 | [Company-hosted audited Form 20-F, major-customer table, F-82][S34] | primary | A individual; B combined |
| 18 | Named commercial relationships | Nvidia, Apple, AMD and Broadcom confirmed; their individual shares of TSMC sales not verified | Company disclosures available at cutoff | [Nvidia][S14], [Apple][S15], [AMD][S16], [Broadcom][S17] | primary | A relationships only |
| 19 | Documented CoWoS expansion baseline | July 2024 management statements: capacity more than doubled in 2024; aimed to double or more again in 2025; partly shared, version-specific tool sets | July 18, 2024 statement | [Call][S18] | primary | A statement; not subsequent actual |
| 20 | Current absolute CoWoS capacity/utilization | No verified monthly wafer-capacity or utilization number in retrieved primary disclosures | Oct 17, 2026 search cutoff | [Current call][S4], [packaging agreement][S19] | primary disclosures inspected; coverage limitation | Not verified; no grade for missing measurement |
| 21 | US interposer expansion | US$2bn agreement; initial five-year term; Malta volume ramp expected H1 2028 | Oct 8, 2026 announcement | [GlobalFoundries][S19] | primary | A announced terms |
| 22 | Overseas gross-margin dilution outlook | Approximately 2 percentage points in 2026; 3–4 points in the second part of the five-year period as more US fabs ramp | Oct 15, 2026 | [Call, prepared remarks and Q&A p.14][S4] | primary | A guidance |
| 23 | Arizona investment envelope | Additional US$100bn; cumulative US$265bn; schedule demand-dependent | July 16, 2026 | [Call][S11] | primary | A announced plan |
| 24 | Leading-edge capacity expansion | N2-family annual capacity expected to more than double in 2027 | Oct 15, 2026 | [Call][S4] | primary | A guidance |
| 25 | Pricing policy | Differentiated by technology, market and location; no verified uniform increase | Oct 15, 2026 | [Call][S4] | primary policy; coverage limitation | A policy; uniform increase not verified (no grade) |
| 26 | Dot-com margin stress | Q2 gross margin 19%, utilization 44%; operating income NT$284m / revenue NT$26,298m = operating margin 1.08% | Q2 2001 | [Financial report][S20] | primary | A inputs; B ratio |
| 27 | Dot-com utilization trough | Q3 utilization 41%, gross margin 25%; Q4 utilization recovered to 50% | Q3 / Q4 2001 | [Q3 report][S21], [Q4 release][S22] | primary | A |
| 28 | Financial-crisis shock | Gross margin 31.3%; operating margin 18.6% | Q4 2008 | [Release][S23] | primary | A |
| 29 | Subsequent financial-crisis trough | Gross margin 18.9%; operating margin 3.1%; utilization below 40% | Q1 2009 | [Release][S26], [call][S27] | primary | A |
| 30 | Recent inventory downturn | FY2023 US$ revenue −8.7%; quarterly gross/operating margins Q1 56.3/45.5%, Q2 54.1/42.0%, Q3 54.3/41.7%, Q4 53.0/41.6%; both minima Q4 | 2023 | [Annual][S28], [Q1][S29], [Q2][S30], [Q3][S31], [Q4][S32] | primary | A figures; B minima |
| 31 | Intel Foundry economics and next release | Revenue US$5.765bn; operating loss US$2.089bn; external revenue US$293m. Q3 results remain scheduled for Oct 29 after market close, call 2pm PDT; no Q3 actual adopted | Q2 2026 actuals; schedule checked Oct 25, 2026 | [10-Q][S24]; [Intel notice][S35] | primary | A actuals and schedule |
| 32 | Samsung Foundry economics | Earnings improved before incentive provisions; standalone revenue/profit not supplied in the inspected Q2 release; 2nm HPC engagements expanding | Q2 2026 | [Samsung results][S25] | primary | A statement; absence bounded to this release |
| 33 | US public support | Up to US$6.6bn direct CHIPS funding awarded, milestone-dependent; not all cash received | Nov 15, 2024 award | [Commerce][S33] | primary | A |
| 34 | Monthly rounding reconciliation | 2024 monthly sum NT$2,894,307m versus published total NT$2,894,308m: NT$1m difference; CSV has 33 reported months through Sep 2026 | Data retrieved Oct 17, 2026 | [S7]–[S9]; arithmetic | primary | B |
| 35 | Investment assessment | Comfortable current profitability; meaningful future overbuild risk; no demonstrated current financing crisis | Oct 17, 2026 | Analysis of [S1]–[S4], [S20]–[S32] | analysis of primary sources | C analytical judgement; not a primary fact |
| 36 | Cash remaining after PP&E and dividends | YTD dividends NT$440.853bn; YTD FCF NT$1,112.146bn less dividends = NT$671.293bn; Q3 statement FCF NT$476.570bn less dividends NT$155.595bn = NT$320.975bn. Residuals are not total changes in cash | Jan–Sep / Q3 2026 | [Cash-flow statement, p.4][S3]; subtraction | primary | A dividends; B residuals |
| 37 | Public-support-related investing inflow | NT$198.74bn of other investing inflows, mainly attributable to US Treasury investment tax credits; neither a separately disclosed tax-credit total nor a CHIPS grant receipt | Q3 2026; reported Oct 15 | [Management report, p.5][S2] | primary | A |
| 38 | Customer financing of reserved capacity | Capacity deposits (temporary receipts) NT$189.858bn; refund or receivable offset by mutual consent when agreed conditions are satisfied; no blanket cancellation protection established | Dec 31, 2025 | [Form 20-F, contractual obligations and temporary receipts, pp.40 / F-57][S34] | primary | A reported balance and terms |
| 39 | Capital and other purchase commitments | NT$1,534.575bn total, including NT$1,220.393bn due within one year; commitments for construction, equipment, raw materials and other property/services not recorded on the balance sheet | Dec 31, 2025 | [Form 20-F, contractual obligations, p.40][S34] | primary | A |
| 40 | Actual capex and implied remaining annual spending | Q3 US$13.84bn; YTD US$40.64bn. US$60–64bn guidance less YTD spending implies Q4 US$19.36–23.36bn, 39.88–68.79% above Q3, if guidance is met; this is not separate Q4 guidance | Actuals through Sep 30, 2026; guidance reiterated Oct 15 | [Management report, p.5][S2], [October call][S4]; subtraction/division | primary | A actuals and annual guidance; B implications |
| 41 | Capex guidance growth, derived | July midpoint US$62bn versus January US$54bn: +14.81%; current US$60–64bn range versus FY2025 US$40.90bn: +46.70–56.48% | FY2026 guidance; Oct 17, 2026 calculation | [January][S6], [July][S11], [October][S4] calls; division | primary | B |
| 42 | Dated fab commissioning plans | Additional Tainan N3 volume production in H1 2027 and H1 2028; Arizona fab 2 volume production in Q3 2027 | Oct 15, 2026 plan | [October call, p.5][S4] | primary | A guidance |
| 43 | Equipment supply constraint and expansion | ASML expects planned 2027–2028 supply to fall short of customer demand; low-NA EUV and immersion DUV capacity expansion planned at 30% in 2027, with 2028 expansion being examined beyond the previously indicated further 30% | Oct 14, 2026 | [ASML Q3 release, outlook and footnote 2][S37] | primary | A management outlook, not independently verified demand |
| 44 | Contingent US expansion | Texas evaluation and Terafab discussions remain exploratory; no final arrangement announced; TSMC requires fab operational control. Excluded from committed capacity | Oct 15, 2026 | [October call, p.9][S4] | primary | A statement |
| 45 | Full-year revenue outlook | Approximately 44% growth in US-dollar revenue | FY2026 guidance, Oct 15, 2026 | [Presentation, slide 9][S12] | primary | A guidance |
| 46 | Through-cycle profitability targets | Gross margin at least 56%; ROE in the high twenties; management expectations, not guaranteed floors | Jan 15, 2026 | [January call, p.5][S6] | primary | A management targets |
| 47 | Forward AI-accelerator and company revenue outlook | Narrowly defined AI-accelerator revenue: mid-to-high-fifties percent CAGR, 2024–2029; overall company US$ revenue: approximately 25% CAGR over the same period. In October Wei suggested AI growth would be higher still, without issuing a replacement number | Jan 15 / Oct 15, 2026 | [January call, pp.5–6][S6]; [October call, pp.6–7][S4] | primary | A management forecasts and statement, not outcomes |
| 48 | Scale of multiyear investment and fab construction | 2026–28 capex expected to be significantly higher than US$101bn in 2023–25; nearly 20 fabs in progress: 13 leading-edge / advanced-packaging fabs in Taiwan and five to six overseas | Jan 15 / Oct 15, 2026 | [January call, pp.5 / 18][S6]; [October call, pp.4–5][S4] | primary | A reported baseline and management plans |
| 49 | Lag between capex and output | New fab construction takes two to three years; January's new spending contributes almost no output in 2026, a little in 2027, with principal supply additions in 2028–29 | Jan 15, 2026 | [January call, pp.14 / 18][S6] | primary | A management timing assessment |
| 50 | Consumer demand caution alongside AI scarcity | Consumer and price-sensitive segments challenged by rising component prices and economic uncertainty; prudent 2027 planning despite AI demand exceeding supply; smartphone share 21% of Q3 revenue (claim 4) | Oct 15, 2026 | [October call, pp.3–4 / 6][S4] | primary | A management assessment and reported mix |
| 51 | Depreciation and amortization, versus depreciation guidance | Q3 D&A NT$220.20bn versus Q2 NT$198.54bn and Q3 2025 NT$162.79bn: +35.3% YoY, +10.9% QoQ, 14.7% of Q3 revenue. January guided FY2026 depreciation alone to grow by a high-teens percentage | Q3 2026 actuals; Jan 15, 2026 guidance | [Management report, p.5][S2]; [January call, p.4][S6]; division | primary | A inputs and guidance; B growth rates and revenue share |
| 52 | Capacity-deposit trend and order horizon | NT$291.102bn at end-2024 to NT$189.858bn at end-2025: −34.8%; ending balance equals approximately 5.0% of FY2025 revenue. Customers generally do not place purchase orders far in advance; settlement remains conditional refund / receivable offset (claim 38) | Dec 31, 2024 / 2025; filing Apr 16, 2026 | [Form 20-F, Item 4 / F-57][S34]; division | primary | A balances and terms; B decline and revenue comparison |
| 53 | Anonymous customer concentration trend | Customer A revenue NT$352.2712bn (12%) in 2024 to NT$726.9743bn (19%) in 2025: +106.4%. Customer B share 25% / 22% / 17% in 2023 / 2024 / 2025; identities not established | FY2023–2025 | [Form 20-F, F-82][S34]; division | primary | A amounts and shares; B growth |
| 54 | Revenue path implied by rounded reported results and forecasts | FY2025 US$122bn, +35.9%, implies FY2024 approximately US$89.8bn; FY2026 at +44% implies US$175.7bn. A 25% CAGR from 2024 reaches approximately US$175.3bn in 2027 and US$274.0bn in 2029; reaching the latter from implied FY2026 requires approximately 16.0% CAGR in 2027–29 | FY2025 actual; Jan / Oct 2026 forecasts; calculated Oct 18, 2026 | [January call, pp.3 / 6][S6]; [October call, p.4][S4]; arithmetic | primary | A reported inputs and forecasts; B conditional calculations, not new guidance |
| 55 | Capital intensity and scale of annual spending | FY2026 US$60–64bn / implied US$175.68bn revenue = 34.2–36.4%, versus FY2025 US$40.9bn / US$122bn = 33.5%; FY2026 capex equals 59.4–63.4% of the entire US$101bn 2023–25 total | FY2026 guidance; calculated Oct 18, 2026 | [January][S6] and [October][S4] calls; division | primary | B conditional ratios from primary inputs |
| 56 | Scale of public-support-related investing inflow | NT$198.74bn equals 21.6% of Q3 CFO; capex NT$443.21bn / US$13.84bn implies NT$32.02/US$, converting the inflow to approximately US$6.2bn, comparable to US$6.6bn CHIPS grant ceiling. Entire inflow is not a separately identified tax-credit receipt | Q3 2026; calculated Oct 18, 2026 | [Management report, p.5][S2]; [award][S33]; division | primary | A inflow and award ceiling; B scale comparisons and approximate conversion |
| 57 | Approved dividend increase and illustrative cash requirement | Q2 2026 dividend NT$7.00/share, distribution Jan 7, 2027; Q3 2026 dividends paid NT$155.595bn imply approximately NT$6.00/share using 25.932bn shares. At that unchanged share count, NT$7 requires approximately NT$181.5bn per payment; later quarterly declarations are not established here | Board approval Aug 11, 2026; Q3 report Oct 15 | [6-K, presentation slides 6 / 10][E2]; [cash-flow statement][S3]; multiplication/division | primary | A declared amount, date, shares and cash paid; B per-share inference and illustrative total |
| 58 | Additional technology commissioning plans | Japan fab 2: N3 volume production in 2028; A14: risk production in 2027, volume production in 2028 | Oct 15, 2026 | [October call, p.5][S4] | primary | A guidance |
| 59 | Cost inflation in the build | CFO acknowledged inflation in tools, materials and labor, while maintaining that most capex is driven by long-term demand and expecting revenue growth to outpace capex growth if execution succeeds | Oct 15, 2026 | [October call, p.14][S4] | primary | A management statement, not independently measured inflation |
| 60 | Reported 2027 pricing changes | Baseline increases of 5–10% on advanced nodes from the beginning of 2027, varying by node/customer, with reported additional 10–15% premiums for overflow HPC orders; reported, unverified. October management response reiterated differentiated pricing without confirming the figures | Report Jul 21 / management response Oct 15, 2026 | [Tom's Hardware, citing Nikkei][T1]; [October call, pp.17–18][S4] | secondary report; primary response | C reported figures; A management policy response |
| 61 | Fixed-cost exposure and Taiwan concentration | Operating fab costs largely fixed and generally do not decline with demand/utilization; majority of net revenue derived from Taiwan operations | FY2025 Form 20-F, filed Apr 16, 2026 | [Form 20-F, Item 3][S34] | primary | A disclosed risk factors |
| 62 | Power and financing in management's demand assessment | Asked about datacenter power and funding constraints, Wei said capacity planning considers CSP capex, datacenter construction, financing, power and other component supply | Oct 15, 2026 | [October call, p.6][S4] | primary | A management statement, not proof of downstream funding adequacy |
| 63 | Sequential cash-flow improvement and lower capex | Presentation Q3 FCF NT$476.56bn versus Q2 NT$287.36bn; CFO NT$919.77bn versus NT$783.36bn; capex NT$443.21bn versus NT$496.00bn. Management attributes improvement to both higher CFO and lower capex | Q2 / Q3 2026 | [Management report, p.5][S2]; [6-K, presentation slide 7][E2] | primary | A reported figures and explanation |
| 64 | Credit rating / financial strength | Moody's upgraded TSMC Aa3→Aa2, outlook positive→stable; action also covers TSMC-guaranteed senior unsecured debt of TSMC Arizona and TSMC Global. Agency growth/cash-flow projections are expectations, not realized results | Oct 21, 2026 action; reported Oct 22 | [Moody's indexed research headline; body unavailable][S39]; [Taipei Times][S40] | primary indexed headline; reputable secondary action details | B action/details; agency body restricted |

## Analysis

**Current earnings and financing.** Currency matters: NT$ and US$ revenue growth differ. Platform mix uses net revenue; node mix uses wafer revenue. [S1][S2]

Reported operating cash flow exceeds gross PP&E purchases: after dividends, the statement-based residual is NT$671.293bn for January–September 2026 and NT$320.975bn for Q3. These residuals support observed self-funding; they are not total changes in cash. The presentation reports Q3 FCF of NT$476.56bn, while statement inputs yield NT$476.570bn; the NT$10m discrepancy is retained explicitly rather than mixing source precisions. The year-to-date FCF calculation is CFO minus gross PP&E purchases; the after-dividend residual then deducts dividends. Neither calculation adds asset-sale proceeds. The liquidity cushion uses current and long-term interest-bearing debt, not merely the long-term bond balance. This is not a liquidation estimate or a deduction of all liabilities. These observations support a low current refinancing-pressure assessment. [S3][S12]

TSMC identifies operating cash and corporate bonds as funding sources for production investment. [S10] Customer deposits are a small and shrinking offset: NT$189.858bn at December 31, 2025, down from NT$291.102bn a year earlier (−34.8%) and approximately 5% of FY2025 revenue, with refund or receivable-offset settlement by mutual consent under agreed conditions. The Form 20-F says customers generally do not place purchase orders far in advance. Capital and other purchase obligations were NT$1,534.575bn, including NT$1,220.393bn due within one year. These commitments constrain cash flexibility, but the older balances cannot simply be subtracted from September 2026 liquidity; future operating inflows also matter. They establish neither current deposit balances nor universal take-or-pay protection. [S34] If demand disappoints, the immediate mechanisms are lower factory absorption, weaker cash generation and reduced returns on invested equity. Contractors and equipment suppliers face order deferrals; creditors bear losses only if deterioration ultimately overwhelms the company’s resources. Government grants transfer some construction risk to taxpayers, subject to award conditions. [S33] Public support does not guarantee plant profitability.

Q3 2026 depreciation and amortization was NT$220.20bn, +35.3% YoY and +10.9% QoQ, or 14.7% of revenue. January guidance was for a high-teens percentage increase in full-year depreciation; D&A includes amortization, so these are different measures and periods, not a demonstrated full-year guidance miss. The expanding fixed-cost base increases exposure to weak loading, the mechanism behind the 2001 and 2009 margin troughs. Q3 FCF also benefited from capex declining to NT$443.21bn from NT$496.00bn in Q2, alongside higher CFO; Q3 cash generation should not be extrapolated without the Q4 spending step-up. [S2][S6][S34]

The approved Q2 2026 dividend rises to NT$7.00 per share, payable January 7, 2027, versus approximately NT$6.00 per share implied by Q3 cash dividends paid. At September's unchanged share count it would require approximately NT$181.5bn per payment, compared with Q3 presentation FCF of NT$476.56bn. That is an illustrative cash requirement; future quarterly declarations and cash generation remain to be observed. [E2][S3]

**Demand, customer concentration and pricing.** Customer relationships are documented independently of the anonymous concentration table: Nvidia describes Blackwell manufacturing at TSMC; Apple describes its Arizona chip supply; AMD confirms its manufacturing partnership; Broadcom identifies TSMC as an outsourced foundry. [S14]–[S17] The filings do not authorize assigning the anonymous shares to particular names. Customer A revenue rose from NT$352.2712bn (12%) in 2024 to NT$726.9743bn (19%) in 2025, +106.4%; Customer B's share fell from 25% in 2023 to 22% in 2024 and 17% in 2025. Exposure to the largest anonymous customer is rising, although no named-customer share is established. [S34] AMD and Broadcom may also transmit demand from overlapping cloud customers, so diversity of chip designers does not establish diversity of ultimate funding.

TSMC can benefit from either merchant GPUs or custom accelerators using its processes and packaging. A broad AI investment retreat would still hurt demand. The October 15 outlook for approximately 44% FY2026 US-dollar revenue growth is counterevidence to an already collapsing market, while leaving future investment returns unresolved. [S12] The FY2025 accelerator share and Q3 2026 HPC mix do not reconcile total current AI exposure; AI-linked CPUs and networking can sit outside the narrow accelerator category. [S4][S6] Management's forward AI outlook is much stronger than the backward share: January's mid-to-high-fifties percent accelerator revenue CAGR for 2024–29, with Wei suggesting in October that growth could be higher. This is the principal bullish demand assumption underwriting capacity, not a realized result. [S6][S4] Apple-related consumer demand offers some diversification but cannot be assumed to fill freed AI capacity immediately: process qualifications, designs and packaging requirements differ. On October 15 management flagged rising component prices and economic uncertainty hurting consumer and price-sensitive markets, and prudent planning for 2027. With smartphones at 21% of Q3 sales, this is material counterevidence to an assumption of uniformly strong end markets. Component inflation provides a possible cross-layer constraint, but the statement does not quantify how much AI investment causes consumer weakness. [S2][S4]

Citi explicitly asked about datacenter power and funding constraints; Wei said TSMC considers CSP capex, construction, financing, power and component availability. This connects foundry demand to hyperscaler spending and the financing of downstream AI deployment. Management's diligence does not establish those buyers' eventual economic returns. [S4]

Pricing power is supported by high realized margins and scarcity, but neither proves unlimited price discretion. Revenue mix, currency, yield and utilization also influence profitability. The management report attributes the sequential gross-margin decline primarily to the initial N2 ramp, demonstrating that technological progress itself can impose a near-term cost. [S2] No customer-level contractual price schedule or blanket increase is established here. A July 21 Nikkei-sourced report describes 5–10% base increases on advanced nodes from the beginning of 2027 and additional 10–15% premiums for overflow HPC orders (reported, unverified; C). Asked about price-increase reports on October 15, Wei reiterated differentiated pricing by segment, location and technology without confirming or denying the reported figures. If implemented, price would contribute to revenue growth, so sales growth would not measure chip-volume demand alone. [T1][S4]

**Capacity and overbuild.** The July capex midpoint rose 14.81% from January, and the current range implies 46.70–56.48% growth over FY2025. With US$40.64bn spent through September, meeting the annual range requires US$19.36–23.36bn in Q4, 39.88–68.79% more than Q3. Observed cash coverage therefore needs a forward spending check. [S2][S4][S6][S11] The October plan dates Tainan N3 additions to H1 2027 and H1 2028, Arizona fab 2 to Q3 2027 and Japan fab 2 N3 to 2028; A14 risk production is planned for 2027 and volume production for 2028. These are useful early loading checkpoints. The broader exposure runs into 2028–29: in January Wei explained the two-to-three-year fab construction lag, with almost no same-year contribution and only a little in 2027 from the new spending. October's plans envisage nearly 20 fabs in progress, including 13 in Taiwan and five to six overseas, and 2026–28 capex substantially above the US$101bn spent in 2023–25. Demand absorption in 2028–29 is therefore the principal test of this investment wave. Texas evaluation and Terafab talks remain exploratory and are excluded from committed capacity. [S4][S6]

Revenue is already approximately a year ahead of January's long-term growth path. FY2025 US$122bn at +35.9% implies FY2024 of about US$89.8bn; October's approximately 44% FY2026 growth outlook implies US$175.7bn. A 25% CAGR from 2024 reaches about US$175.3bn in 2027 and US$274.0bn in 2029. Holding that 2029 endpoint would imply approximately 16% annual growth during 2027–29. These are conditional calculations from rounded primary inputs, not new guidance. FY2026 capital intensity is approximately 34.2–36.4%, versus 33.5% in 2025; annual capex alone equals 59.4–63.4% of the whole 2023–25 total. [S6][S4]

This combination raises the burden on later sales, loading and productivity. An upward long-term outlook revision would support the bullish demand case; unchanged guidance alongside faster investment would increase concern about returns, but would not by itself prove deterioration, since yields, mix, pricing and productivity also matter. Watch the January 2027 long-term outlook. The CFO acknowledged tool, material and labor inflation in October while still expecting revenue to outgrow capex if execution succeeds: more capex dollars do not translate one-for-one into more physical capacity. [S4]

CoWoS integrates logic and memory into advanced packages. Its economic constraint is useful packaged output, not just the number of wafer starts. TSMC’s earlier explanation emphasized differences among CoWoS versions and their equipment. [S18] Consequently, a capacity headline can overstate flexibility: tools serving one package architecture may not immediately relieve another bottleneck. Larger packages, yield losses and processing intensity also make a single wafer-count comparison an incomplete measure of economic capacity.

The GlobalFoundries agreement establishes an additional US interposer supply route, not an already operational expansion of completed CoWoS packages. [S19] Other packaging steps and customer qualification must also be available. The announced ramp date illustrates the lag between demand signals and usable output. It does not establish a take-or-pay obligation for the agreement’s full headline value; termination and minimum-volume protections were not disclosed in the announcement.

ASML’s October 14 outlook says planned 2027–2028 supply cannot fully meet expected customer demand even with expansion. Tool scarcity can delay usable fab capacity, tempering the speed of a potential surplus; the same outlook also shows correlated AI optimism reaching equipment suppliers. Neither supplier’s forecast proves profitable end-user demand. [S37]

An overbuild mechanism is synchronized expansion across foundries, packaging partners and downstream customers on correlated demand forecasts. Scarcity-driven reservations may weaken once availability improves. The Codex cross-check correctly identified customer funding, but the 34.8% deposit decline during 2025 and short order horizon limit the protection that can be inferred: deposits do not establish universal take-or-pay commitments. Shareholders carry the principal utilization and fixed-cost exposure; exact cancellation rights remain contract-dependent. [S34] Even if AI usage keeps increasing, cheaper inference, improved utilization of existing accelerators or slower monetization could reduce incremental chip purchases. Conversely, efficiency could expand affordable use sufficiently to sustain demand. Neither direction is settled by supplier revenue alone. This is an analytical risk assessment, not evidence that current orders are fictitious.

**Overseas economics and Taiwan.** Arizona’s cumulative investment envelope is a multiyear plan; adding it to annual global capex would double-count investment. [S11] Public assistance offsets some capital cost, but award ceilings are not current cash receipts. [S33] Actual receipts are partly visible: Q3 2026 other investing inflows of NT$198.74bn were mainly attributable to US Treasury investment tax credits. This is neither an exact tax-credit total nor a disclosed CHIPS grant receipt. The investing classification helps explain cash accumulation without entering CFO-minus-gross-capex FCF. [S2] At the capex-implied exchange rate of approximately NT$32.0/US$, the entire NT$198.74bn inflow is about US$6.2bn in one quarter, or 21.6% of Q3 CFO, comparable in scale to the US$6.6bn CHIPS grant ceiling. These are scale comparisons: the inflow is outside CFO, is not all identified as tax credits, and is not proof of cumulative tax-credit support exceeding grants. [S2][S33] Local labor, supplier networks, construction and ramp efficiency determine whether geographic diversification earns acceptable returns. Overseas group-margin dilution is not the same thing as a disclosed Arizona standalone loss. October guidance put 2026 dilution near 2 percentage points; the CFO clarified in Q&A that 3–4 points refers to the second part of the five-year period as more US fabs ramp, rather than automatically to 2027. [S4]

The annual filing locates most principal production facilities and revenue-generating operations in Taiwan and warns about military conflict, transportation disruption and trade restrictions. [S34] A blockade or prolonged infrastructure interruption could prevent production or delivery even if customer demand remains strong. Overseas capacity reduces exposure gradually; it cannot instantly reproduce the interconnected Taiwan manufacturing, supplier and engineering ecosystem. No invasion probability or assumption of normal operations during a disruption is supported here.

**What downturns demonstrate.** The 2025 Form 20-F says operating fabs are largely fixed-cost assets: a significant portion of costs generally persists when demand or utilization drops. This directly supports the historical margin mechanism. [S34] During the dot-com contraction, the low point for utilization followed the low point for gross margin. Better mix and higher NT-dollar average selling prices helped margins recover while factory loading still deteriorated; this does not isolate contractual price increases. [S20][S21] TSMC also deferred expansion. Historical accounting scope and technology mix differ from today, so the episode demonstrates fixed-cost sensitivity rather than a reusable numerical stress-test formula.

The financial-crisis shock began in the late-2008 results, but the deeper margin and utilization trough came in early 2009. Calling the later figures “2008 utilization” would misdate the evidence. [S23][S26][S27] In the more recent inventory correction, TSMC retained much higher margins; its annual report nevertheless attributes pressure partly to lower utilization and the new-node ramp. [S28] The historical conclusion is survival with widely varying profitability, not an invariant margin floor. Today’s favorable product mix may cushion a contraction, but historical resilience does not guarantee that newly installed capacity earns its cost of capital. Management’s January targets of at least 56% through-cycle gross margin and high-twenties ROE provide an opposing profitability case; future loading and realized returns must test those targets rather than treating old troughs or management ambitions as fixed floors. [S6]

**Other foundries.** Intel’s segment revenue is overwhelmingly internal manufacturing, so it is not comparable to TSMC’s external-customer revenue. Its filing also says external growth benefited from Altera’s reclassification following deconsolidation, while reduced period charges helped the operating-loss comparison. A revenue increase therefore does not by itself demonstrate competitive external foundry traction. [S24] Losses must be financed within Intel’s wider corporate funding structure; the disclosed segment has not demonstrated self-funding economics.

Samsung reports improving orders and expanding advanced-node engagements, which contradict any assumption that alternative suppliers cannot gain business. But its disclosure does not establish standalone Foundry profitability. [S25] Group or memory earnings cannot fill that gap. These alternatives could constrain TSMC pricing or attract geographically motivated demand, yet technology availability is insufficient: manufacturing yield, reliable delivery, packaging and customer qualification determine commercial substitutability.

## Conflicts and open questions

- **Management confidence versus overinvestment risk.** In October, Wei said demand was “still far above our capability today to supply.” [S4] In January he said, “I'm also very nervous about it,” and added, “If we did not do it carefully, that will be a big disaster to TSMC for sure,” while explaining checks with customers and cloud providers. [S6] The Claude review describes this warning once as an October comment; the fetched transcript dates it to January 15, 2026. Management also reports inspecting US datacenter construction and power readiness (primary, A). [S4] Such diligence supports capacity planning, but construction progress does not establish profitable end-user demand.
- **Funding interpretation: Codex versus Claude cross-checks.** Codex (October 17) emphasized customer deposits, tax-credit-related inflows and purchase commitments as additions to the original operating-cash/bond/grant funding account; it argued that contracts and assistance affect the sharing of losses. Claude (October 18) accepts those facts but finds the protection overstated: deposits fell from NT$291.102bn to NT$189.858bn in 2025 (−34.8%), about 5% of annual revenue at year-end, while orders generally are not placed far ahead. Both support low current funding stress. Retain Codex's contractual qualifications, with Claude's emphasis that shareholders bear the principal factory-utilization risk; neither source reveals all cancellation terms. [S2][S3][S34]
- **Timing of overbuild: Codex versus Claude cross-checks.** Codex used the Tainan H1 2027 / H1 2028 and Arizona Q3 2027 commissioning dates as concrete investment tests. Claude shifts the principal window to 2028–29, based on January's two-to-three-year construction lag and October's 2026–28 spending wave. Both are retained: individual ramps are early checkpoints, while the broader capex wave's output must be absorbed in 2028–29. [S4][S6]
- **Overseas dilution timing: Codex versus Claude cross-checks.** Codex accepted 3–4 points in the following few years from October's prepared remarks. Claude points to the same call's Q&A clarification: the second part of the five-year period. Claim 22 follows the more specific clarification; neither wording supplies a precise 2027 dilution forecast. [S4]
- **Long-term path versus returns.** Claude argues that revenue running a year ahead of the 25% path requires either higher guidance or deteriorating investment returns. The conditional growth arithmetic is retained (claim 54), but that is not an exhaustive causal choice: pricing, yields, productivity and the phasing of investment/output also affect returns. October's expectation that revenue growth can outpace capex is management's opposing case. [S4][S6]
- **Risk emphasis: original note versus cross-check.** The original note called sustained AI demand the biggest uncertainty and described Taiwan disruption as potentially much more severe. The review accepts both risk mechanisms but finds no measured probability or severity ranking. Dated commissioning and spending checkpoints make demand risk testable; the note does not quantify or rank it against Taiwan disruption. [S4][S20][S26][S34]
- **HPC versus AI.** TSMC excludes CPUs from its narrow accelerator definition because their ultimate application cannot reliably be separated. [S4] The disclosed accelerator share and HPC platform mix are different measurements and periods. A current, reconciled AI revenue bridge is missing. The original note used the narrow historical share to caution against equating HPC with AI; the cross-check agrees but warns that this can understate current AI sensitivity by omitting AI-linked CPUs and networking. Neither disclosed share measures total current AI exposure. [S2][S4][S6]
- **Scarcity versus future surplus.** The original note argued that present scarcity supports investment but invoked an unquantified backlog. The cross-check instead emphasizes commissioning dates and the remaining spending requirement; no contractually protected backlog total is established here. TSMC’s approximately 44% FY2026 revenue-growth outlook is evidence against an already collapsed market. [S12] ASML on October 14 expected planned 2027–2028 supply to fall short of demand, with 30% capacity expansion planned for 2027 and a 2028 increase under study beyond the previously indicated further 30% for low-NA EUV and immersion DUV. This can delay effective foundry supply, while expansion still creates later surplus risk. [S37] Exact installed CoWoS capacity, utilization by version and blanket cancellation protection remain unverified.
- **Profitability targets versus historical stress.** January management targets of at least 56% through-cycle gross margin and high-twenties ROE support a stronger resilience case than the original note emphasized. [S6] The note’s historical stress evidence shows that profitability has varied sharply; the cross-check asks that both cases be tested, with management targets treated as expectations rather than guaranteed floors. [S20][S26]
- **Pricing and overseas costs.** Expansion costs and competitors limit margin extrapolation. The original note described subsidy receipts as insufficiently disclosed; the cross-check identifies NT$198.74bn of Q3 other investing inflows mainly attributable to investment tax credits. [S2] Customer-specific US premiums, plant-level returns and an exact tax-credit/CHIPS receipt split remain unverified.
- **Concentration and historical comparability.** Named customer revenue percentages, exact late-2008 utilization and the quarterly utilization series for 2023 were not verified. Anonymous customer letters can change ranking between annual presentations. A current break-even utilization threshold cannot responsibly be extrapolated from old troughs.
- **Scope limits.** Intel’s next results were scheduled after this cutoff; Samsung’s October preliminary announcement did not disclose Foundry economics. [S35][S36] The peer comparison therefore uses their latest detailed quarterly disclosures. This note assesses business and capacity risk, not whether any current share price already discounts those risks.

## Signals to watch

| Signal | Current value | As of | Source | Why it matters |
|--------|---------------|-------|--------|----------------|
| Monthly sales growth | September +54.6% YoY; August +53.3% | Sep 2026 | [S9] | Persistent deceleration is more informative than a seasonal monthly dip; check currency effects. |
| HPC platform mix | 68% | Q3 2026 | [S2] | Indicates computing exposure, without treating all HPC as AI. |
| Leading-edge mix | Advanced nodes 79% of wafer revenue | Q3 2026 | [S2] | Tests whether new capacity supports high-value output. |
| Capex revisions and remaining spend | US$60–64bn annual guidance; implied Q4 US$19.36–23.36bn versus Q3 US$13.84bn | Oct 15, 2026 | [S2][S4] | Check execution and cash coverage; a cut alongside weaker sales/loading strengthens an overbuild warning. |
| Fab commissioning and capex lag | Tainan N3 H1 2027 / H1 2028; Arizona fab 2 Q3 2027; main new-spending output in 2028–29 | Jan 15 / Oct 15, 2026 plans | [S4][S6] | Early ramp checks lead into the principal 2028–29 demand-absorption test. |
| Equipment availability | ASML expects planned 2027–2028 supply below demand | Oct 14, 2026 | [S37] | Delays can slow effective foundry supply growth. |
| Exploratory US expansion | Texas evaluation / Terafab talks; no final arrangement; TSMC operational control required | Oct 15, 2026 | [S4] | Count capacity only after a concrete commitment. |
| CoWoS supply announcements | GF interposer agreement; ramp expected H1 2028 | Oct 2026 | [S19] | Track commissioning and qualification, not just committed dollars. |
| Utilization commentary | Demand exceeds supply; numerical node/CoWoS rates unverified | Oct 2026 | [S4] | Watch for allocation relief, idle tools and order postponements. |
| Margins | Gross 66.4%; operating 58.7% | Q3 2026 | [S1] | Separate loading deterioration from predictable ramp costs. |
| Cash versus investment | Quarterly FCF NT$476.56bn | Q3 2026 | [S12] | Sustained deterioration would weaken the self-funding conclusion. |
| Overseas cost absorption | About 2 points in 2026; 3–4 in the second part of the five-year period | Oct 15, 2026 guidance | [S4] | Tests whether diversification remains economically manageable; no automatic 2027 step to 3–4 points. |
| Competitive profitability | Intel Foundry operating loss US$2.089bn; Samsung standalone profit absent from inspected release. Intel Q3 due Oct 29 after close, call 2pm PDT; not yet released | Q2 2026 actuals; schedule checked Oct 25, 2026 | [S24][S25][S35] | Distinguishes alternative capacity from commercially sustainable competition; retain Q2 economics until new results. |
| Credit / financial strength | Moody's Aa2 stable, upgraded from Aa3 positive on Oct 21; includes guaranteed subsidiary debt; B because agency body unavailable | Oct 21, 2026 action; checked Oct 25 | [S39][S40] | Supports low current funding stress; monitor ratings alongside cash generation, capex and Taiwan concentration. |
| Long-term outlook revision | AI-accelerator CAGR mid-to-high-fifties percent; company approximately 25%, both 2024–29; October suggested higher AI growth | Jan 15 / Oct 15, 2026 | [S6][S4] | An upward revision supports the demand case; holding the company endpoint implies approximately 16% growth in 2027–29 against higher investment. |
| Customer capacity deposits | NT$189.9bn, approximately −35% YoY | Dec 31, 2025 | [S34] | Higher deposits indicate more customer funding; falling balances leave more capacity exposure with TSMC, subject to settlement terms. |

## Sources

Full URLs are the link targets below. Sources were originally accessed **2026-10-17**; **S2, S4, S6 and S34 re-fetched 2026-10-18; E2 and T1 first fetched 2026-10-18; S35 re-fetched and S39/S40 first fetched 2026-10-25**. Other sources retain their original access date. E2 was fetched with the generic User-Agent `research-bot contact@example.com`. Company-hosted call transcripts are primary evidence of management statements. The [monthly CSV](/data/tsmc-monthly-revenue.csv) preserves reported rounded values and omits unreported future months.

1. [S1 — TSMC, Q3 2026 earnings release][S1], published 2026-10-15.
2. [S2 — TSMC, Q3 2026 management report][S2], published 2026-10-15.
3. [S3 — TSMC, Q3 2026 condensed financial statements, unaudited][S3], published 2026-10-15.
4. [S4 — TSMC/LSEG, Q3 2026 earnings-call transcript][S4], call 2026-10-15.
5. [S5 — TSMC, Q4 2025 management report][S5], published 2026-01-15.
6. [S6 — TSMC/LSEG, Q4 2025 earnings-call transcript][S6], call 2026-01-15.
7. [S7 — TSMC, 2024 monthly revenue][S7], rolling monthly releases; completed January 2025.
8. [S8 — TSMC, 2025 monthly revenue][S8], rolling monthly releases; completed January 2026.
9. [S9 — TSMC, 2026 monthly revenue][S9], rolling monthly releases through 2026-10-08.
10. [S10 — TSMC, 2025 annual report, cash flow and investment sections][S10], published 2026; precise day not verified.
11. [S11 — TSMC/LSEG, Q2 2026 earnings-call transcript][S11], call 2026-07-16.
12. [S12 — TSMC, Q3 2026 earnings presentation][S12], published 2026-10-15.
13. [S13 — TSMC, 2025 consolidated financial report, major-customer note, SEC-hosted][S13], published 2026; precise filing day not verified.
14. [S14 — Nvidia, first US-produced Blackwell wafer][S14], published 2025-10-17.
15. [S15 — Apple, US manufacturing commitment and Arizona supply relationships][S15], published 2025-08-06.
16. [S16 — AMD, first TSMC N2 product-silicon milestone][S16], published 2025-04-14.
17. [S17 — Broadcom, FY2025 Form 10-K][S17], fiscal year ended 2025-11-02; filed December 2025.
18. [S18 — TSMC/LSEG, Q2 2024 earnings-call transcript][S18], call 2024-07-18.
19. [S19 — GlobalFoundries, US silicon-interposer manufacturing agreement][S19], published 2026-10-08.
20. [S20 — TSMC, Q2 2001 financial report][S20], published 2001-07-26.
21. [S21 — TSMC, Q3 2001 financial report][S21], published 2001-10-26.
22. [S22 — TSMC, Q4 2001 results][S22], published 2002-01-25.
23. [S23 — TSMC, Q4 2008 results][S23], published 2009-01-22.
24. [S24 — Intel, Q2 2026 Form 10-Q][S24], filed 2026-07-24; quarter ended 2026-06-27.
25. [S25 — Samsung Electronics, Q2 2026 results][S25], published 2026-07-30.
26. [S26 — TSMC, Q1 2009 results][S26], published 2009-04-30.
27. [S27 — TSMC, Q1 2009 earnings-call transcript][S27], call 2009-04-30.
28. [S28 — TSMC, 2023 annual report][S28], published 2024; precise day not verified.
29. [S29 — TSMC, Q1 2023 results][S29], published 2023-04-20.
30. [S30 — TSMC, Q2 2023 results][S30], published 2023-07-20.
31. [S31 — TSMC, Q3 2023 results][S31], published 2023-10-19.
32. [S32 — TSMC, Q4 2023 results][S32], published 2024-01-18.
33. [S33 — US Department of Commerce, final TSMC Arizona CHIPS award][S33], published 2024-11-15.
34. [S34 — TSMC, 2025 Form 20-F, risk factors, contractual obligations, customer deposits and concentration][S34], filed 2026-04-16.
35. [S35 — Intel, notice of Q3 2026 reporting date][S35], published 2026-10-06.
36. [S36 — Samsung Electronics, Q3 2026 preliminary guidance][S36], published 2026-10-08.
37. [S37 — ASML, Q3 2026 results, equipment-capacity outlook and footnote 2][S37], published 2026-10-14.
38. [S38 — TSMC, Q4 2025 earnings presentation, SEC-hosted][S38], published 2026-01-15.
39. [E2 — TSMC, Q3 2026 Form 6-K with earnings release and presentation exhibits, SEC mirror of S1/S12][E2], filed 2026-10-15; accessed 2026-10-18.
40. [T1 — Anton Shilov, Tom's Hardware, reported TSMC 2027 chip-production price increases, citing Nikkei][T1], published 2026-07-21; accessed 2026-10-18; reported, unverified.
41. [S39 — Moody's Ratings, research feed, indexed TSMC Aa2/stable upgrade headline][S39], action 2026-10-21; accessed 2026-10-25; research body unavailable, so action details retain B grade.
42. [S40 — Crystal Hsu, Taipei Times, Moody's upgrades TSMC credit rating][S40], published 2026-10-22; accessed 2026-10-25.

[S1]: https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-10/526b65af7305df2dde6b9ece6812b1fdf1e49f9b/3Q26%20EarningsRelease.pdf
[S2]: https://investor.tsmc.com/chinese/encrypt/files/encrypt_file/reports/2026-10/f83780000ddc0c3b48e57a87eab6814c4ab35058/3Q26%20ManagementReport.pdf
[S3]: https://investor.tsmc.com/chinese/encrypt/files/encrypt_file/reports/2026-10/3a94d1034b9e6b22446b674f7aaf3d088a12a193/FS.pdf
[S4]: https://investor.tsmc.com/japanese/encrypt/files/encrypt_file/reports/2026-10/eba4fac72873bf546b669bd2b5d973df5c93215d/TSMC%203Q26%20Transcript.pdf
[S5]: https://investor.tsmc.com/english/encrypt/files/encrypt_file/qr/phase4_reports/2026-01/a920e9c2e3879f198dc700d041344229e7ffb438/4Q25%20Management%20Report.pdf
[S6]: https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-01/51d09df96cd89ac19d65af39032b038dc2896a24/TSMC%204Q25%20Transcript.pdf
[S7]: https://investor.tsmc.com/chinese/monthly-revenue/2024
[S8]: https://investor.tsmc.com/english/monthly-revenue/2025
[S9]: https://investor.tsmc.com/english/monthly-revenue/2026
[S10]: https://investor.tsmc.com/sites/ir/annual-report/2025/2025%20Annual%20Report.E.pdf
[S11]: https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-08/3e494f0c14dd0890f897aa044415e21d93486cc4/TSMC%202Q26%20Transcript.pdf
[S12]: https://investor.tsmc.com/english/encrypt/files/encrypt_file/reports/2026-10/9f58506f6f114f91cefbf86820cc63297b39b130/3Q26%20Presentation%20%28E%29.pdf
[S13]: https://www.sec.gov/Archives/edgar/data/1046179/000104617926000024/a2025q4consolidatedreport-.htm
[S14]: https://blogs.nvidia.com/blog/tsmc-blackwell-manufacturing/
[S15]: https://www.apple.com/newsroom/2025/08/apple-increases-us-commitment-to-600-billion-usd-announces-ambitious-program/
[S16]: https://ir.amd.com/news-events/press-releases/detail/1245/amd-achieves-first-tsmc-n2-product-silicon-milestone
[S17]: https://www.sec.gov/Archives/edgar/data/1730168/000173016825000121/avgo-20251102.htm
[S18]: https://investor.tsmc.com/chinese/encrypt/files/encrypt_file/reports/2024-08/5122725a56670882d777a8e8bfe0ed247cc55330/TSMC%202Q24%20Transcript.pdf
[S19]: https://gf.com/news-and-events/news/globalfoundries-reaches-agreement-to-establish-us-based-supply-of-silicon-interposers-for-advanced-ai-packaging/
[S20]: https://investor.tsmc.com/english/encrypt/files/encrypt_file/english/2001/Q2/2q01fr.pdf
[S21]: https://investor.tsmc.com/english/encrypt/files/encrypt_file/english/2001/Q3/3q01fr.pdf
[S22]: https://pr.tsmc.com/english/news/2404
[S23]: https://pr.tsmc.com/english/news/1532
[S24]: https://www.sec.gov/Archives/edgar/data/50863/000005086326000157/intc-20260627.htm
[S25]: https://news.samsung.com/global/samsung-electronics-announces-second-quarter-2026-results
[S26]: https://pr.tsmc.com/english/news/1548
[S27]: https://investor.tsmc.com/schinese/encrypt/files/encrypt_file/english/2009/Q1/09Q1%20Conf%20Call-Transcript.pdf
[S28]: https://investor.tsmc.com/static/annualReports/2023/english/index.html
[S29]: https://pr.tsmc.com/schinese/news/3013
[S30]: https://investor.tsmc.com/schinese/quarterly-results/2023/q2
[S31]: https://pr.tsmc.com/english/news/3076
[S32]: https://pr.tsmc.com/english/news/3101
[S33]: https://www.commerce.gov/news/press-releases/2024/11/biden-harris-administration-announces-chips-incentives-award-tsmc
[S34]: https://investor.tsmc.com/sites/ir/sec-filings/2025_20F%20Report.pdf
[S35]: https://www.intel.com/content/www/us/en/newsroom/news/corporate/intel-reports-third-quarter-2026-financial-results.html
[S36]: https://news.samsung.com/global/samsung-electronics-announces-earnings-guidance-for-third-quarter-2026

[S37]: https://investor.asml.com/news-releases/news-release-details/asml-reports-eu11.7-billion-total-net-sales-and-eu5.0-billion-net-income-q3-2026
[S38]: https://www.sec.gov/Archives/edgar/data/1046179/000104617926000008/a4q25presentatione.htm

[E2]: https://www.sec.gov/Archives/edgar/data/1046179/000104617926000726/0001046179-26-000726.txt
[T1]: https://www.tomshardware.com/tech-industry/semiconductors/tsmc-eyes-price-hikes-of-up-to-25-percent-on-chip-production-services-in-2027-report-claims-plans-to-raise-baseline-prices-by-5-percent-to-10-percent-on-advanced-nodes
[S39]: https://ratings.moodys.com/research
[S40]: https://www.taipeitimes.com/News/biz/archives/2026/10/22/2003866384