Taiwan Semiconductor Manufacturing Company TSM

Reference table

1. What the business is, and what actually drives it

TSMC is a dedicated semiconductor foundry — the 20-F uses exactly that phrase. It manufactures integrated circuits to customer designs and does not compete with its customers in design. Chief executive C. C. Wei; 65,152 full-time employees per the vendor profile; listed on the NYSE since 1997-10-09 and on the Taiwan Stock Exchange as 2330.

The 20-F states that wafer fabrication accounted for approximately 86% of net revenue in 2025, with the remainder from packaging and testing, mask making, design and royalties. The vendor's seg_prod independently reproduces this exactly:

This table is the single most important data-integrity fact in the dive, and it is in Section 1 rather than Section 8 because everything downstream depends on it. The seg_prod total for FY2025 is NT$3,809,054 million — which is exactly the net revenue the 20-F reports. The vendor's own income statement (inc_a) reports NT$3,848,511 million for the same year. The filing wins, the segment file agrees with the filing, and the income statement is the outlier by NT$39,457 million (1.04%). Full treatment in Section 8; every FY2025 figure in this dive is on the filing basis.

Revenue by technology node — the disclosure that explains the margin. The 20-F's node table did not survive text extraction, but the narrative states the figures precisely:

Sixteen points of advanced-node mix in two years, and 550 basis points of gross margin alongside it. That is the mechanism of this business stated in four numbers. 2-nanometer entered volume production in 2025; 16-angstrom risk production is expected in 2026.

And the growth is not coming from building more. The 20-F: annual capacity in 12-inch equivalent wafers "exceeded 17 million" in 2025 against "approximately 17 million" in 2024, while shipments rose from 13 million to 15 million. Revenue rose 31.6% on a 15% shipment increase — the balance is price and mix, which the filing confirms directly: the increase "was mainly attributed to an increase in ASP due to a higher proportion of advanced technology (7-nanometer and below) revenue and an increase in wafer shipments, partially offset by an appreciation of the NT dollar."

Revenue by platform. The platform table also failed to extract; these figures are derived from the disclosed absolute increases and percentage changes and are labelled as approximations:

High performance computing and smartphone together account for roughly 90% of the entire NT$914,746 million revenue increase in 2025, and high performance computing alone for 78% of it. This is a business whose growth has one dominant driver.

Revenue by geography (customers' headquarters, not shipment destination — the 20-F is explicit about this):

The FY2025 seg_geo block is incomplete — the Taiwan and "Other" lines are absent and NT$369,847 million (9.7% of revenue) is unaccounted for. FY2024 reconciles to the last dollar; FY2025 does not. Flagged in Section 8 and not estimated around. What is safe to read from the complete lines: US-headquartered customers went from 68.8% to at least 74.4% of revenue in one year, while China fell from 11.5% to 8.6% in absolute as well as relative terms (NT$331,673M to NT$327,503M — the only region to shrink).

Customer concentration is high and rising, and the 20-F quantifies it:

The second-largest customer's share has risen 6 points in two years while the largest has fallen 6. The top two together are 36% of revenue. The 20-F names the mechanism: "a growing trend among system companies designing their own semiconductors and working directly with the semiconductor foundries… These shifting business models could lead to significant variations in our sales if the growth of their products and services, particularly in the AI sector, is volatile or not sustainable."

2. The foreign-private-issuer problem — the single most important analytical issue on this name

TSM is a foreign private issuer. There is no 10-K and there is no 10-Q. The filings available are a 20-F filed 2026-04-16 (the annual report, covering FY2025) and three 6-Ks dated 2026-07-13, 2026-07-16 and 2026-07-24. That fact has four consequences, and the first three are valuation-breaking if unhandled.

2.1 The disclosure gap — annual, not quarterly

A US filer of this size would have published two 10-Qs since its last annual report, each with a full balance sheet, a cash-flow statement, segment detail and a management discussion. TSMC has published none. The most recent balance sheet and cash-flow statement in this file are dated 2025-12-31, seven months stale, and there is no quarterly balance sheet at all — bal_a contains three annual periods and nothing else. Quarterly income statements exist and run to 2026-06-30, but with no matching balance sheet, every leverage, liquidity and enterprise-value figure in this dive is anchored to a December year-end. Earnings-call transcripts are not available on this plan either, so management's own commentary on the June quarter is entirely absent. Both are stated as gaps rather than worked around.

2.2 The currency — and it is mixed within the payload

The 20-F is explicit: "Substantially all of our sales are denominated in U.S. dollars while we publish our financial statements in NT dollars." The vendor payload does not honour a single convention. Sorted by what each block actually contains:

km_ttm and ratios_ttm carry no currency field at all. A reader who takes priceToEarningsRatioTTM of 26.89 or bookValuePerShareTTM of 1,248.46 at face value against a $417.17 price is combining two currencies in one ratio. This is precisely the class of error the data contract warns "the renderer cannot detect — only prose catches those."

The conversion rate used throughout this dive is NT$31.11 to US$1.00, which is the FY2025 weighted-average rate disclosed in the 20-F (against NT$32.13 in 2024 — the NT dollar appreciated, which the filing says had an unfavourable effect on reported revenue). It is used because it comes from the filing, and because it reconciles: trailing four-quarter earnings per ADS-equivalent share of NT$434.95 converts to US$13.98, against US$13.81 summed from the four earn_cal USD actuals — a 1.2% agreement. Note the sensitivity honestly: a 5% move in the NT dollar moves every USD-denominated multiple in this dive by 5%, in the opposite direction. The quarter-by-quarter rates implied by dividing TWD revenue by earn_cal USD revenue range from 29.4 to 32.3 and cannot all be right; the filing's disclosed annual average is the only rate with an auditable source.

2.3 The ADS ratio and the premium — where the multiples actually break

The 20-F states: "each ADS represents five (5) common shares," and reports 25,932,524,521 common shares outstanding. Divided by five, that is 5,186,504,904 ADS-equivalent shares — which matches the vendor's weightedAverageShsOutDil of 5,186,504,800 for Q4'25 almost exactly. So the vendor's share count IS ADS-equivalent, and market capitalisation reconciles: $417.17 × 5.1865 billion = $2,163.6 billion, the reported figure. That part is clean.

What is not clean is the price the TWD block is struck against. Vendor km_ttm.marketCap is NT$60,163,168 million. Divided by 5,186.5 million ADS-equivalent shares, that implies a price of NT$11,600 per ADS-equivalent unit, or US$372.87 at NT$31.11. The ADS closed at $417.17. The gap is 11.9% — and against the previous close of $406.11, 8.9%.

The insider block settles it beyond argument. It contains transactions in both securities, and on 2026-07-28 the same officer bought both on the same day:

• Common Shares (2330.TW) at $67.97 → five of them is $339.85 of ownership
• American Depositary Shares (TSM) at $390.00 for the identical five-share claim

That is a same-filer, same-day, directly observed ADS premium of +14.8%.

Therefore every trailing multiple in km_ttm and ratios_ttm is a Taiwan-line multiple, and understates what an ADS buyer pays by roughly 12%. Corrected:

No vendor multiple is used anywhere in this dive. All of Section 6 is rebuilt on the ADS price and filing-sourced conversion.

2.4 What the 6-Ks give that no US filer gives

The compensation for the thin annual disclosure is real and it is the best thing about this name analytically. TSMC files revenue monthly. The 6-K of 2026-07-13 reports:

> June 2026 consolidated revenue: approximately NT$442.68 billion — an increase of 6.2% from May 2026 and an increase of 67.9% from June 2025. Revenue for January through June 2026 totalled NT$2,404.48 billion, an increase of 35.6% compared to the same period in 2025.

This is more recent than any quarterly statement in the payload and it validates the payload at the same time. The vendor's Q1'26 revenue of NT$1,134,103 million plus Q2'26 of NT$1,270,381 million sums to NT$2,404,484 million — the 6-K's H1 figure to the last reported digit. The quarterly income statements for 2026 are therefore confirmed by an SEC filing, which is more than can be said for the FY2025 annual line.

The 6-K of 2026-07-16 is the Q2 results filing, but its exhibit did not extract — only the 1,106-character cover page is on disk, so the Q2 detail (margin, node mix, guidance) is unavailable and is not used. The 6-K of 2026-07-24 is a month-end administrative report: no share cancellations, no new unsecured bonds, no board-approved capital appropriations — housekeeping, no signal.

Section 3 builds the run-rate discussion on the monthly data, because it is the best evidence available.

3. Growth — and what the monthly data says consensus requires

Filing-basis revenue history (FY2025 corrected to the 20-F figure):

Note FY2023: revenue fell 4.5%. This business is cyclical and has been within the last three years. Anyone underwriting the current growth rate as a permanent condition should look at that row first.

The quarterly path is a straight line up, and the last two points are confirmed by the 6-K:

Gross margin has expanded in seven of the last eight quarters and reached 67.7% in Q2'26 — 780 basis points above the FY2025 full-year figure of 59.9%. Operating margin is 60.3%. These are not foundry economics as historically understood; they are the economics of a supplier with no alternative.

Q2'26 beat by a wide margin: US$4.31 actual against US$3.87 estimated, +11.4%. The four-quarter beat record is unbroken: $2.50 vs $2.38, $2.92 vs $2.63, $3.09 vs $2.90, $3.49 vs $3.31, $4.31 vs $3.87 — five consecutive beats, and the size of the beat is growing.

The monthly run-rate test — what consensus actually requires

This is the analysis the 6-K makes possible and no US filer would permit.

• H1 2026 actual: NT$2,404,484M (+35.6%), a monthly average of NT$400,747M
• June 2026 alone: NT$442,680M — 10.5% above the H1 average, +6.2% month over month, +67.9% year over year
• FY2026 consensus revenue: NT$5,392,870M (21 analysts) — implying an H2 of NT$2,988,386M, a monthly average of NT$498,064M

Which means consensus requires monthly revenue to rise 12.5% above June's level and hold there, on average, for six months. Put the other way: if monthly revenue simply holds flat at June's NT$442,680M through December, FY2026 lands at NT$5,060,564M — +32.9% year over year, and a 6.6% miss against consensus.

That is the cleanest falsifiable statement available on this name, it is checkable every month, and the July report is due within days of this dive. It is not a heroic bar — June already grew 6.2% month over month, and roughly 3% a month compounding gets there — but it is a bar, and consensus is not merely extrapolating.

Consensus forward path (TWD, converted at the filing's NT$31.11):

FY2025's +31.6% accelerating to a consensus +41.6% in FY2026 is the second-derivative signal this house screens for, and it is genuine. Coverage is deep on FY2026 and FY2027 (10 analysts on earnings, 21-22 on revenue) and thin beyond (5 and 4). No conclusion in this dive rests on the FY2028 or FY2029 rows; FY2027 is the primary forward anchor.

4. Balance sheet — and the corrected enterprise value

As at 2025-12-31 (the most recent balance sheet in existence for this filer — see Section 2.1):

• Cash and equivalents NT$2,767,856M · short-term investments NT$360,441M · total NT$3,128,298M (US$100.6B)
• Total debt NT$1,064,583M (short-term NT$136,926M + long-term NT$896,062M + finance leases NT$31,595M — these three sum exactly to the reported total, so there is no double-count of the kind seen on other names)
• Corrected net position: NT$3,128,298M − NT$1,064,583M = NT$2,063,715M NET CASH (US$66.3B)
• Property, plant and equipment NT$3,735,760M · total assets NT$7,932,843M · total equity NT$5,396,219M (minority interest NT$41,181M)
• Debt-to-equity 0.153 · current ratio 2.46 · interest coverage 285x · financial leverage 1.46x

The vendor's netDebt of −NT$1,703,274M omits the entire NT$360,441M of short-term investments — the same defect the data contract flags, though at this scale it is a 17% understatement of net cash rather than a sign-flip.

Corrected enterprise value, and why the vendor's is wrong twice over. km_ttm.enterpriseValueTTM reports NT$58,011,397M (US$1,864.7B). That figure is built on the Taiwan-line market capitalisation and the understated net cash. On the ADS basis an investor actually transacts in:

> EV = market cap US$2,163.6B + debt US$34.2B − cash and short-term investments US$100.6B = US$2,097.3B (NT$65,247,246M)

The vendor understates enterprise value by 11.1%. On trailing EBITDA of NT$3,266,796M (derived from the vendor's 73.4% EBITDA margin on trailing revenue of NT$4,450,379M), corrected EV/EBITDA is 20.0x against the vendor's 17.80x.

A note on the equity base. Book value per ADS-equivalent share is NT$1,248.46 (US$40.13), giving a corrected price-to-book of 10.40x. intangiblesToTotalAssetsTTM is 0 and tangible book equals book — goodwill is NT$5,891M and intangibles NT$19,062M against NT$7.93 trillion of assets, i.e. immaterial. This is a genuinely tangible balance sheet: 47% of total assets is property, plant and equipment. A 10.4x multiple of tangible book is a high price for a capital-intensive manufacturer, and it is one of the few metrics on which this name looks expensive rather than cheap. The vendor's own scoring model agrees: rating gives an overall B+ with a priceToBookScore of 1 and a priceToEarningsScore of 2, against return-on-equity and return-on-assets scores of 5.

5. Cash flow — the capital-intensity problem

The 20-F states capital expenditure of NT$949,817M (2023), NT$956,007M (2024) and NT$1,272,411M (2025), the last being US$40,895M at NT$31.11. The vendor's capitalExpenditure line reports NT$1,285,592M for FY2025 and NT$955,398M for FY2023 — NT$13,181M and NT$5,581M above the filing respectively, while its investmentsInPropertyPlantAndEquipment line matches the filing to the thousand. The vendor's capex figure evidently blends intangible-asset additions into property, plant and equipment. The filing figures are used (Section 8).

Three observations, and the third is the bear case.

First, the free-cash-flow trajectory is genuinely improving in absolute terms — NT$292bn to NT$870bn to NT$1,111bn across three years, as revenue outgrew a capital budget that stayed near a third of sales.

Second, and against that: the 20-F guides 2026 capital expenditure to US$52-56 billion, against US$40.9 billion actually spent in 2025 — an increase of 27% to 37%. This is a management-sourced forward figure and is half-weighted accordingly, but it is a filed disclosure rather than a call remark. If revenue lands on the FY2026 consensus of NT$5,392,870M (US$173.3B), that capital budget is 30-32% of revenue — no relief on intensity — and it converts into depreciation across FY2027-FY2029, which is the most credible route to the margin disappointment Section 6c flags.

Third, and this is the number that most argues against paying up: trailing free-cash-flow yield on the ADS is roughly 1.65%. FY2025 free cash flow of NT$1,110,765M is US$35.7B against a market capitalisation of US$2,163.6B. capexToOperatingCashFlowTTM is 58.3%. A shareholder is buying a business that converts a 45% net margin into a sub-2% cash yield, because the cash is being reinvested in fabrication plants. That is defensible while the returns on that capital are 27.1% (returnOnInvestedCapitalTTM) — but it means the equity return must come from earnings growth and multiple, not from distributions.

Distributions are small. FY2025 dividends paid were NT$471,614M, roughly NT$90.94 per ADS-equivalent share (US$2.92); the trailing declared figure is NT$110 (US$3.54). On the ADS at $417.17 that is a yield of 0.70% to 0.85% — not the 0.95% the vendor reports, which is struck on the Taiwan line. Share repurchase was zero in FY2025 and NT$3,089M in FY2024 — immaterial. The 2026-07-24 6-K confirms no cancellation of common shares in June.

6. Valuation — priced in or room?

At $417.17 per ADS (market cap US$2,163.6B; 5.1865 billion ADS-equivalent shares; 1 ADS = 5 common shares). All multiples below are on the ADS basis, converted at the 20-F's NT$31.11 — see Section 2.3 for why the vendor's figures differ.

PEG on the FY2027 line is 0.61 (18.7x against 30.4% consensus earnings growth). By that single metric the stock is inexpensive. By price-to-book at 10.4x on a business whose assets are 47% factories, it is not. Both are true, and the tension between them is the valuation question.

Peer note. The vendor peer set is mostly adjacent rather than comparable: ASML ($659.8B), AMAT ($434.0B) and LRCX ($397.4B) are equipment suppliers (TSMC's vendors, not its competitors); AVGO ($1,989.4B), MU ($1,008.2B) and DIOD ($4.1B) are chip designers or memory makers; AMKR ($13.9B), ASX ($85.1B) and TSEM ($26.8B) are packaging, testing or specialty foundries an order of magnitude smaller; META ($1,497.8B) is not a semiconductor company at all and its presence in the set is a vendor error. There is no pure leading-edge foundry peer, because there is no pure leading-edge foundry peer — which is the investment case. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn.

6a. What today's price assumes (the inversion)

At $417.17 — 24.4x FY2026 and 18.7x FY2027 consensus earnings per ADS — the price embeds roughly the following, each falsifiable and each dated:

• Monthly revenue keeps climbing from the June base. (Consensus-derived, from the 6-K of 2026-07-13 and FY2026 consensus of NT$5,392,870M.) June 2026 printed NT$442,680M; consensus requires a second-half monthly average of NT$498,064M, 12.5% above June. A flat-at-June second half produces NT$5,060,564M, a 6.6% miss. Checkable monthly — the July report is due around 2026-08-10. This is the most immediately testable assumption in the price.
• Gross margin holds near the Q2'26 level of 67.7% while the overseas fabrication plants ramp. (Our reading. FY2025 whole-year gross margin was 59.9%; Q2'26 was 67.7%.) The 20-F states that gross margin "would be negatively impacted in the year when a new technology is introduced," discloses Arizona, Kumamoto and Dresden projects, and does not quantify their margin effect anywhere. The most recent independent knowledge-base reading on this is negative — eurodollar_university, 2026-07-30, conviction 65, on record capex and overcapacity at a capital-cycle peak.
• Earnings per share reach roughly US$22.32 by FY2027 and US$28.13 by FY2028. (Consensus, 10 and 5 analysts. Dispersion on FY2027 is wide: US$17.98 low to US$25.57 high — a 42% spread.) At the FY2027 low the stock is on 23.2x today, not 18.7x.
• The market keeps paying roughly 19-24x forward earnings despite an unquantifiable concentration risk. (Our reading.) This is the most fragile assumption in the price, because it is the one with no financial anchor at all — the discount the market applies to that risk is a sentiment variable that has ranged widely and can reset in a day without a single number changing.
• The ADS premium of roughly 15% to the Taiwan-listed ordinary share persists. (Our number, derived from the same-day insider transactions of 2026-07-28: ADS at $390.00 against common at $67.97, i.e. $339.85 for the identical five-share claim.) A US buyer at $417.17 is paying about 15% more per unit of the same company than a Taipei buyer. Nothing in the fundamentals supports or refutes that spread; it is pure demand for the accessible listing, and it can compress independently of the business.

6b. The return bridge (why the multiple moves)

Expected return over 12-24 months decomposes as: earnings-per-share growth (+30.4% consensus FY2027, +26.0% FY2028) + multiple drift + shareholder yield (0.70-0.85% dividend, no meaningful buyback).

Our $470 base assumes modest multiple EXPANSION — from 18.7x to 21.1x FY2027 — and we name that as the fragile leg, but the far more important point is how little of the return depends on it. If the multiple simply holds at today's 18.7x and FY2027 consensus earnings of US$22.32 are delivered, the price is $417 — exactly where it is, because 18.7x FY2027 is what is being paid today. The return over a 12-month horizon therefore comes from the market rolling its anchor forward: at an unchanged 18.7x on FY2028 consensus of US$28.13, the price is $526 (+26%). That rolldown, not a re-rating, is the engine.

We do not underwrite the full $526, for two reasons stated plainly. First, the FY2028 line carries only five analysts and this dive does not rest conclusions on it. Second, the rolldown assumes no widening of the risk discount — and on a name whose dominant risk is unsizeable, assuming the discount stays constant is an assumption, not a neutral base case. Our $470 therefore sits between the no-return case ($417, multiple holds on FY2027) and the full-rolldown case ($526), and represents 21.1x FY2027 or 16.7x FY2028.

The bull case at $610 requires 27.3x FY2027 — roughly where the street already is (its $596 consensus target is 26.7x FY2027) — and that leg genuinely needs the crowd to stop discounting the concentration risk. It is not needed for the base case.

Note the mechanical trap this name sets, per the addendum's guardrail: 24.4x FY2026 becoming 18.7x FY2027 at a constant price is rolldown, not compression. The only honest re-rating statement is relative to what the market pays at each date, and on that basis our base assumes a modest expansion and the bear assumes a sharp compression to 12.1x.

6c. Variant perception (where we differ, what would surprise)

We are more cautious than the entire covering street, and that is unusual enough to justify. Consensus target is $596, median $600, low $500, high $700, with 18 buys, 7 holds and zero sells. Our $470 base is 21% below the consensus target and 6% below the street's LOWEST published target. Four specific departures:

• We differ on what security is being valued. Every vendor multiple in this file, and quite possibly some street models, is struck on the Taiwan-listed line. The ADS trades roughly 15% above it — proven from a same-day, same-filer insider transaction on 2026-07-28. A target set on the local shares translates to an ADS target 15% lower than it appears. We do not know how many of the 25 covering analysts model which line, and neither does the vendor file, but the correction runs one way only: it makes the ADS more expensive, never less. Watchable: the ADS-to-local spread, observable in any future dual-security insider filing.
• We differ on cash conversion. The street's enthusiasm is anchored to a 45% net margin and 39.3% return on equity. The trailing free-cash-flow yield is 1.65% and capital expenditure absorbs 58.3% of operating cash flow, guided to US$52-56 billion in 2026 against US$40.9 billion in 2025. A business reinvesting at 27% returns on invested capital should reinvest — but the equity holder's return is then entirely earnings growth and multiple, with no cash cushion if either disappoints. Watchable number: FY2026 capital expenditure against the guided range, and the depreciation line in FY2027.
• We differ on the risk discount, and we think the market's is too small rather than too large. The bull framing on this is well argued — gavin_baker (2026-01-05, conviction 62) holds that "TSMC national-security risk is overstated (world could roll back ~2 nodes)" and raoul_pal (2025-05-31, conviction 70) argues that US and Taiwanese interests are "so interwoven… that outright conflict becomes unthinkable." We do not dismiss that. But the countervailing claims are more specific about consequences than the bull claims are about probability: dylan_patel (2026-05-13, conviction 65) sizes a hard takeover at "GDP down ~20%, and a global depression lasting 5-8+ years," and all_in (2026-07-15, bearish) notes Taiwan holds "under 3 weeks of energy reserves" against a blockade requiring a "90-day restart." When one side of an argument can size the outcome and neither side can size the probability, the honest response is a wider discount, not a narrower one.
• We do NOT differ on the growth, and this is worth saying clearly. The +31.6% FY2025, the +35.6% first half, the +67.9% June and the 74% advanced-node mix are all from filings and all point the same way. Our caution is entirely about what multiple to pay, not about whether the business is performing. Anyone whose disagreement with us is "the growth is real" is agreeing with us.
• Positive surprise that would force us up: a July or August monthly revenue print above NT$500,000 million (which would put the second half ahead of the consensus run-rate immediately rather than requiring a ramp), or a Q3 print on 2026-10-15 above US$4.45 with gross margin holding near 67% and explicit evidence that overseas-plant dilution is smaller than feared. Watchable: the monthly 6-K, roughly the 10th of each month.
• Negative surprise that would break the thesis: the capital cycle turning, which is the freshest independent bearish reading available — eurodollar_university, 2026-07-30, conviction 65: "Chips are at the peak of a capital cycle — record profits are driving record capex and overcapacity near the top; MSCI World Semiconductor Index fell 16% in July, its worst month since 2022." That claim does not name TSM and sits in an adjacent lane, but it names the exact mechanism (record capex at the top) that the 20-F's US$52-56 billion guidance instantiates. Watchable number: monthly revenue growth decelerating below +30% year over year for two consecutive months.

Synthos fair values

All three anchors are multiples of FY2027 consensus earnings per ADS of US$22.32 (10 analysts, the best-covered forward line), cross-checked against the FY2027 dispersion and against FY2026 and FY2028.

• Bear ~$270 — 12.1x FY2027E. Cross-check: 15x the FY2027 estimate LOW of US$17.98 gives $270, the same number from the other direction. Also 15.8x FY2026E and 19.3x trailing. The scenario: the capital cycle rolls over as eurodollar_university argues, the US$52-56 billion 2026 capital budget lands into softening advanced-node demand through 2027, gross margin retreats from 67.7% toward the mid-fifties on overseas-plant dilution and lower utilisation, the ~15% ADS premium compresses toward zero, and the risk discount widens. Note this is not a catastrophe case — it contains no disruption event, only a normal semiconductor down-cycle of the kind FY2023's −4.5% revenue year already demonstrated. −35.3%.
• Base ~$470 — 21.1x FY2027E. Cross-checks: 27.5x FY2026E, 16.7x FY2028E, and it sits between the no-return case ($417, today's 18.7x held on FY2027) and the full-rolldown case ($526, 18.7x on FY2028). The scenario: monthly revenue continues the ramp the 6-K shows, FY2026 lands near the NT$5,392,870M consensus, advanced-node mix carries margin, and the market's risk discount neither widens nor narrows. +12.7%, plus a 0.70-0.85% dividend.
• Bull ~$610 — 27.3x FY2027E. Cross-checks: 24x the FY2027 estimate HIGH of US$25.57 gives $614; 21.7x FY2028E; and it brackets the street's $596 consensus target (26.7x FY2027) while staying below the street high of $700. The scenario: the second-half monthly ramp beats consensus, 2-nanometer and 16-angstrom sustain pricing through 2027, capital expenditure lands at the low end of guidance, and the concentration discount narrows on continued geographic diversification. +46.2%.

Base is 12.7% above spot; asymmetry is roughly 1.31:1 to the upside (35% down, 46% up). That is positive but not compelling, and it is why the verdict is Buy — Tactical rather than Buy — Core: the fundamentals justify ownership, the asymmetry does not justify size, and the entry does not justify haste.

7. Knowledge base — broad, deep, and not as independent as the headline count

Raw hits naming TSM, TSMC or Taiwan Semiconductor: 132, across 29 distinct sources, spanning 2017-11-17 to 2026-07-23. Headline stance: 87 bullish, 37 neutral, 8 bearish. That headline overstates the case, and here is the correction.

Thirty-five percent of this lane is management, and 38 of those 46 claims are jensen_huang — the chief executive of TSMC's largest customer. That is not a neutral observer; it is a commercially interested party whose own business depends on the narrative that leading-edge supply will scale. The remaining management claims are sam_altman_mgmt (3), lisa_su_mgmt (3) and eli_lilly_ceo (2). The management lane contains zero bearish claims out of 46. It is half-weighted throughout and is not counted toward breadth.

Two quarantine_misattributed claims (2026-05-04 and 2026-02-14) are excluded entirely per house policy and are not quoted.

The independent lane — 84 claims, 24 voices — is the one that carries weight, and it splits 48 bullish / 28 neutral / 8 bearish, i.e. 57% bullish and 10% bearish. Twenty-six of the 84 are from 2026.

A further caution on the raw count: the lane contains near-duplicates. Four jensen_huang claims dated 2026-01-21 say substantively the same thing about new plant construction; three sam_altman_mgmt claims dated 2026-03-11/12 are near-verbatim repetitions. The effective distinct-claim count is meaningfully below 132, which is one more reason the count is not the conviction.

The moat lane — the strongest part, and genuinely independent

> 2026-05-11 · jordi_visser · bullish · conviction 85

> "Only three foundries—Taiwan Semi, Samsung, Intel—control semiconductor production and can't make more; demand came out of nowhere and supply can't catch up."

> 2023-04-21 · business_breakdowns · bullish · conviction 85

> "Near-monopoly/duopoly at leading edge yields record margins and cash flow; can sustain that edge at least through this decade."

> 2026-04-07 · dylan_patel · bullish · conviction 70

> "AI chips are buying up all TSMC 3nm and 2nm capacity, squeezing Apple, Qualcomm and MediaTek off leading-edge nodes and toward Intel foundry."

The last of those is directly corroborated by the 20-F, which shows 3-nanometer at 24% of wafer revenue (up from 18%) and 2-nanometer entering volume production in 2025 — and by the customer concentration table, where the second-largest customer's share rose from 12% to 17% while the largest fell from 22% to 19%. A knowledge-base claim that the filing independently confirms is worth more than five that it cannot touch.

The supply-discipline lane — the reason the exponential score is 7 and not 9

> 2026-05-15 · gavin_baker · bullish · conviction 68

> "Wafer shortage persists for a long time because TSMC's flinty 70-something leaders guard Morris Chang's legacy and expand ~5% a year despite Jensen wanting them to double or triple."

> 2025-12-14 · jordi_visser · bullish · conviction 72

> "Compute shortage persists because TSMC refuses to over-invest; demand stays ahead of supply, so chip/compute scarcity is a durable feature."

The 20-F confirms the mechanism exactly: capacity was approximately 17 million wafers in 2024 and "exceeded 17 million" in 2025. The panel is describing a real, disclosed, deliberate policy. It is bullish for margin and bearish for slope, and this dive scores it as both.

The bearish and cautionary lane — eight independent claims, and they do not agree with each other

> 2026-07-15 · all_in · bearish · conviction 55

> "Taiwan has under 3 weeks of energy reserves; a blockade browns out fabs (90-day restart), an economic hit worse than the Great Depression — China has blockaded the straits seven times in four years."

> 2026-01-19 · niall_ferguson · bearish · conviction 80

> "Taiwan produces all AI GPUs; a China blockade/semiconductor crisis is the defining tail risk of the Trump term and grows as TSMC's centrality rises."

> 2026-05-13 · dylan_patel · neutral · conviction 65

> "Taiwan makes 95%+ of leading-edge and ~half of older chips; a hard takeover would freeze the supply chain, sending markets into free fall, GDP down ~20%, and a global depression lasting 5-8+ years."

> 2025-12-04 · we_study_billionaires · neutral · conviction 45

> "Semiconductors change fastest (Innovator's Dilemma); cutting-edge leaders like Nvidia and TSMC compete only on performance with no margin of safety, so advantages can erode quickly once someone else breaks through."

> 2023-09-01 · marko_papic · bearish · conviction 55

> "Skeptical CHIPS Act reshoring becomes a meaningful trend — costs and business environment aren't competitive; TSMC already delayed its Arizona plant."

> 2025-06-10 · chinatalk · neutral · conviction 70

> "TSMC Arizona is starved of talent because critical tacit manufacturing knowledge sits inside Taiwan and is hard to relocate."

The bearish lane is almost entirely about concentration and reshoring economics, not about the business. Not one bearish claim disputes the margin, the node leadership or the demand. That is a specific and useful finding: the panel's disagreement is about where the fabs are, not about how good they are. The lone valuation-adjacent bear reading (brent_johnson, 2025-06-08, conviction 65, a technical overbought call) is fourteen months old and is not weighted.

The adjacent claim that matters most right now

eurodollar_university has no entity-tagged TSM claim, so it does not appear in the 132. Its 2026-07-30 reading is nonetheless the most recent relevant independent bearish view in the knowledge base and is quoted here with that limitation stated explicitly:

> 2026-07-30 · eurodollar_university · bearish · conviction 65 · NOT entity-tagged to TSM

> "Chips are at the peak of a capital cycle — record profits are driving record capex and overcapacity near the top; MSCI World Semiconductor Index fell 16% in July, its worst month since 2022."

This is the direct counterweight to today's +2.72%. A 16% one-month sector drawdown followed by a market-wide melt-up is the signature of a positioning unwind and re-cover, not of a fundamental re-rating — and it argues for staging an entry rather than establishing one in a single session.

One claim the filing refutes

> 2025-04-17 · dwarkesh · neutral · conviction 60

> "AI chips are still a small share (~5%) of even TSMC leading-edge capacity; most goes to mobile — AI compute has large runway but from a small base."

The 20-F puts high performance computing at roughly 58% of FY2025 revenue against smartphone at roughly 28%, with high performance computing contributing 78% of the entire year's revenue increase. Whether "AI chips" is narrower than the high-performance-computing platform is arguable, but the directional claim — that mobile dominates — is no longer supported by the disclosure. Flagged because a knowledge base that is never checked against filings is a rumour mill.

Why Moderate-High rather than High

Twenty-four independent voices across nine years is genuine breadth, the moat claims are unusually consistent across unrelated sources, and several are corroborated by the 20-F. Against that: 35% of the lane is management, 38 claims come from a single customer's chief executive, that lane has zero bearish entries, the raw count is inflated by near-duplicates, and the most recent adjacent independent reading is negative. A knowledge-base lane whose bullish half is significantly voiced by the customer is not independent conviction, and it is scored as Moderate-High rather than High for exactly that reason.

8. Data integrity — what we rejected from the vendor file and why

Nine material findings. This is the longest such list in the batch, and the first two are of a kind the data contract explicitly says the renderer cannot catch.

1. km_ttm and ratios_ttm are denominated in NEW TAIWAN DOLLARS, carry NO currency field, and are struck against the TAIWAN-LISTED ORDINARY SHARE — every trailing multiple in them is therefore wrong for an ADS buyer. ALL REJECTED. km_ttm.marketCap is NT$60,163,168M, which over 5,186.5M ADS-equivalent shares implies NT$11,600 per ADS-equivalent unit = US$372.87 at the filing's NT$31.11 — against an actual ADS close of $417.17, a gap of 11.9%. The insider block proves the mechanism directly: on 2026-07-28 the same officer bought Common Shares (2330.TW) at $67.97 and American Depositary Shares (TSM) at $390.00, i.e. $339.85 versus $390.00 for the identical five-share claim — a same-day ADS premium of 14.8%. Corrected on the ADS basis: price/earnings 29.8x (not 26.89x), price/book 10.40x (not 9.35x), price/sales 15.16x (not 13.55x), EV/EBITDA 20.0x (not 17.80x), dividend yield 0.85% (not 0.95%), free-cash-flow yield 1.65% (not 1.86%). This is the share-class/partial-unit error class the data contract names; only prose catches it.

2. inc_a FY2025 revenue and net income CONTRADICT the 20-F, and the vendor's own seg_prod sides with the filing. FILING USED THROUGHOUT. The 20-F states net revenue of NT$3,809,054 million and net income attributable to parent shareholders of NT$1,697,604 million for FY2025. inc_a reports NT$3,848,511 million and NT$1,735,678 million — overstatements of NT$39,457M (1.04%) and NT$38,074M (2.24%). Three independent checks favour the filing: (a) seg_prod FY2025 sums to NT$3,809,054.3M, matching the filing exactly; (b) the filing states the increase over 2024 was NT$914,746M, and NT$2,894,308M + NT$914,746M = NT$3,809,054M exactly; (c) the filing's stated +31.6% growth is reproduced only by the filing's figure, not the vendor's (+33.0%). The net-income gap is consistent with inc_a reporting total net income including non-controlling interests rather than the attributable-to-parent figure — FY2024 shows the vendor correctly separating the two (NT$1,158,380M in inc_a against NT$1,173,268M in cf_a) while FY2025 shows both blocks reporting the identical NT$1,735,678M. Corrected FY2025 earnings per ADS-equivalent share: NT$327.31 (US$10.52), not the vendor's NT$333.05. Note that est.epsAvg for FY2025 (NT$323.34) is within 1.2% of the filing figure and 3% below the vendor's — the estimate block agrees with the filing, which is a fourth check.

3. capitalExpenditure in cf_a exceeds the 20-F's stated capital expenditure. FILING USED. The 20-F states NT$949,817M (2023), NT$956,007M (2024) and NT$1,272,411M (2025). cf_a reports NT$955,398M, NT$956,007M and NT$1,285,592M — NT$5,581M and NT$13,181M above the filing in 2023 and 2025. The vendor's investmentsInPropertyPlantAndEquipment line matches the filing to the thousand in every year, so the capitalExpenditure line evidently folds in intangible-asset additions. Free cash flow is computed on the filing figure: FY2025 NT$1,110,765M, not NT$1,097,584M.

4. km_ttm.enterpriseValueTTM (NT$58,011,397M) is understated by 11.1% — two errors compounding. It uses (a) the Taiwan-line market capitalisation per finding 1, and (b) a netDebt of −NT$1,703,274M that omits the entire NT$360,441M of short-term investments. Corrected net cash: NT$2,063,715M (US$66.3B). Corrected enterprise value on the ADS basis: NT$65,247,246M (US$2,097.3B). evToSalesTTM (13.06x), evToEBITDATTM (17.80x), netDebtToEBITDATTM (−0.660x), enterpriseValueMultipleTTM (17.80x), evToOperatingCashFlowTTM (21.63x) and evToFreeCashFlowTTM (51.87x) are all rejected as published; corrected EV/EBITDA of 20.0x is used.

5. seg_geo FY2025 is INCOMPLETE — the Taiwan and "Other" lines are missing and 9.7% of revenue is unaccounted for. FY2025 sums to NT$3,439,207M against NT$3,809,054M of revenue, a NT$369,847M gap. FY2024 and every earlier year reconcile exactly. The four disclosed FY2025 regions are used with their shares stated against the full revenue base and the gap named; the missing lines are not estimated. Separately, the label "TAIWAN, PROVINCE OF CHINA" in the FY2024 block is an ISO country-list artifact of the vendor's mapping, not a disclosure by the company; the 20-F uses "Taiwan" and "R.O.C."

6. The 20-F's node, platform, geography and employee TABLES did not survive text extraction — only surrounding narrative did. The node percentages (3nm 24%, 5nm 36%, 7nm 14%, advanced 74%) and the platform increases (high performance computing +NT$716,040M/+48%; smartphone +NT$105,686M/+11%) are quoted from the narrative and are exact. The platform ABSOLUTE figures in Section 1 are DERIVED from those increases and percentages and are labelled as approximations; deriving FY2024 high-performance-computing revenue from the 2025 disclosure (NT$1,491,750M) versus from the 2024 disclosure (NT$1,476,815M) differs by 1.0%, which is the rounding of the stated percentages. No conclusion rests on the derived absolutes; the disclosed percentages and increases carry the argument.

7. earn_cal USD figures cannot be reconciled to the TWD statements at any single exchange rate. Dividing TWD quarterly revenue by earn_cal USD revenue gives implied rates of 29.41 (Q2'25), 30.59 (Q3'25), 31.87 (Q4'25), 31.52 (Q1'26) and 32.28 (Q2'26) — a 10% range — against the 20-F's disclosed annual weighted averages of NT$32.13 (2024) and NT$31.11 (2025). The filing's NT$31.11 is used for all conversions because it is the only rate with an auditable source, and the resulting trailing USD earnings of $13.98 agrees with the earn_cal sum of $13.81 to within 1.2%. The FX sensitivity is stated in Section 2.2 rather than smoothed over.

8. quote and tech disagree on the 52-week range. quote reports a high of $479.00 and a low of $223.70; tech reports $477.57 and $227.33. The tech figures are used (they are computed from closing prices per the data contract, while quote reflects intraday extremes). The difference is immaterial to any conclusion but is noted rather than silently reconciled.

9. The 6-K of 2026-07-16 — the Q2'26 results filing — did NOT extract its exhibit. Only 1,106 characters of cover page are on disk. The Q2 earnings release detail (segment margin, node mix, forward guidance) is therefore unavailable and is not used anywhere in this dive. Combined with the absence of earnings-call transcripts on this plan, there is no management commentary of any kind on the June quarter in this analysis. Stated as a gap.

Not defects, correctly reported and verified: est.ebitdaAvg for FY2025 (NT$2,662,235M) checks out against operating income plus depreciation and amortisation (NT$1,955,696M + NT$695,224M = NT$2,650,920M, a 0.4% agreement) — this is one of the roughly 30% of files where the estimate EBITDA block is NOT corrupt, though EBIT (NT$1,813,252M against NT$1,955,696M actual) is 7.3% low and is not used. shortTermDebt + longTermDebt + capitalLeaseObligations sum exactly to totalDebt with no double-count. The price-target set shows genuine dispersion ($500 low / $600 median / $596 consensus / $700 high), unlike the single-value sets seen elsewhere. stockBasedCompensationToRevenueTTM of 0.006% is consistent with the cash-flow statement's near-zero stock compensation in every year.

Non-equity tripwire: NOT triggered. TSM is common equity in ADS form — beta 1.369, a 52-week range of $227-$478 (a 110% band), 13.7 million shares of daily volume and a variable dividend. Nothing about it resembles a preferred share or baby bond.

9. Technicals — a bounce inside a drawdown

• Price $417.17, up 2.72% today, on volume of 13.67M against a 15.42M average — the move came on BELOW-average volume, which is not the signature of conviction accumulation.
• −12.6% from the 52-week high of $477.57, but +83.5% above the 52-week low of $227.33. Maximum drawdown from peak over the trailing year is −12.6%, i.e. the stock is close to its worst point of the year and has nearly doubled from its best entry.
• BELOW the 50-day moving average of $425.62 by 2.0%; ABOVE the 200-day of $358.40 by 16.4%. A stock below its short-term average and well above its long-term one is mid-correction within an intact uptrend, not broken and not confirmed.
• RSI 49.2 — precisely neutral. MACD −6.89 — NEGATIVE. Even after a 2.72% day the momentum oscillator has not crossed back.
• Relative performance, and the split is the whole story: 12-month +77.4% against SPY +24.3% and QQQ +30.8% (a 47-to-53 point outperformance); 6-month +22.2% against SPY +11.1% (+11 points); but 3-month +3.9% against SPY +7.6% (−3.7 points). The leadership stopped roughly a quarter ago.
• The context, from the adjacent knowledge base: the MSCI World Semiconductor Index fell 16% in July, its worst month since 2022 (eurodollar_university, 2026-07-30). Today's move was market-wide across semiconductors, not TSM-specific.
• Practical read for entry. Today is a sector-beta bounce off a one-month sector drawdown, on below-average volume, into an unreclaimed 50-day average, with MACD still negative. Buying $417.17 after a +2.72% session is chasing. The disciplined approach is a staged entry: an initial tranche now to establish presence, a second on any retest of the 200-day at $358.40 (−14%), and the balance held against a monthly revenue print that confirms the second-half ramp. Reclaiming and holding the 50-day at $425.62 would justify accelerating that schedule; a failure back below $390 on company-specific news would justify pausing it.

10. Insiders — small, one-directional, and evidentially useful for something else

There is no insider selling in this file. The only disposition is the 5,000,000-share "gift" of 2026-07-31, which is matched by an identical 5,000,000-share acquisition on the same date by the same director at zero price, moving from direct to indirect ownership — a reclassification of holding form, not a sale, and it should not be read as one.

The rest is one officer buying steadily through the July drawdown — 1,000 common shares on 2026-07-21 (twice), 2026-07-28 and 2026-08-03, plus 15 ADSs on 2026-07-28. These are small (roughly $73,000 on 2026-08-03) and represent one vice-president, not the board. They should not be over-read as a signal.

Their real value here is evidential. The 2026-07-28 entries are the only place in the dataset where both securities are transacted on the same day by the same person, and they establish the ADS premium at +14.8% ($390.00 per ADS against $339.85 for the five common shares it represents). That single pair of rows corrects every valuation multiple in Section 6 — a good illustration of why the insider block is worth reading even when the transactions themselves are trivial.

11. Verdict, kill-criteria and flip conditions

Buy — Tactical.

What is genuinely, verifiably good. Net revenue grew 31.6% in FY2025 to NT$3,809,054 million per the 20-F, with gross margin rising 380 basis points to 59.9% and advanced technologies at 7-nanometer and below rising from 69% to 74% of wafer revenue — on capacity that did not grow. The 6-K of 2026-07-13 shows the acceleration continuing: first-half 2026 revenue +35.6%, June alone +67.9% year over year. Q2'26 gross margin reached 67.7%, operating margin 60.3%, and the company has beaten the estimate five consecutive quarters by a widening margin. The balance sheet holds US$66.3 billion of net cash with 285x interest coverage. Consensus compounds earnings per share 30.4% and 26.0% in FY2027 and FY2028 against an 18.7x FY2027 multiple — a PEG of 0.61. And 24 independent knowledge-base voices converge on a moat thesis the filing corroborates on the specifics.

What you are accepting. A dominant risk that cannot be sized from any statement, binary and unhedgeable. A price 15% above the same company's Taiwan listing, a spread with no fundamental support that can compress on sentiment alone. A 1.65% free-cash-flow yield on a business guiding capital expenditure up 27-37% to US$52-56 billion. Customer concentration at 78% for the top ten and 36% for the top two. A seven-month-old balance sheet, no quarterly filing, no earnings-call transcript and an unextracted Q2 exhibit — the thinnest interim disclosure of any name at this market capitalisation. A capital-cycle warning that is the freshest independent bearish reading in the data. And an entry price that rose 2.72% today on below-average volume inside a sector-wide melt-up, with the 50-day average unreclaimed and MACD negative.

Why Tactical rather than Core. The base case is +12.7% against a bear case of −35.3% — an asymmetry of 1.31:1 that supports ownership but not size. Stage it: an initial tranche, a second near the 200-day average at $358.40, and the balance released by a monthly revenue print that confirms the second-half ramp. Do not establish the full position on a day the whole sector moved together.

Pre-registered KILL criteria — what would break the thesis:

• Two consecutive monthly revenue 6-Ks showing year-over-year growth below +30%. June printed +67.9%; a fall below +30% would mean the second-half ramp consensus requires is not happening, and it would be visible in a filing within weeks rather than at the quarterly print.
• Gross margin falling below 60% in a reported quarter — retracing the entire FY2025 gain and confirming that overseas-plant dilution or utilisation loss is real. Q2'26 was 67.7%; FY2025 was 59.9%.
• 2026 capital expenditure guided or tracking ABOVE the US$52-56 billion range without a matching revenue upgrade — the signature of the capital-cycle top eurodollar_university describes, and directly destructive to an already-thin free-cash-flow yield.
• The second-largest customer's revenue share reversing sharply (it went 11% to 12% to 17% across 2023-2025). Concentration cuts both ways and this is the most likely single source of a revenue air-pocket.
• Credible evidence of second-source qualification at the leading edge, which is the knowledge base's own competitive bear case: "cutting-edge leaders… compete only on performance with no margin of safety."
• A decisive break below the 200-day moving average of $358.40 on company-specific news rather than sector beta.
• Any material change in the concentration risk picture. This is unfalsifiable in advance and therefore not a number — but it is the criterion that would matter most, and the position is sized on the assumption that it cannot be forecast.

Pre-registered FLIP TO CORE (upsize):

• A monthly revenue print above NT$500,000 million, which would put the second half at or above the consensus run-rate immediately rather than requiring a ramp — the single cleanest upside confirmation available.
• The Q3'26 print on 2026-10-15 above US$4.45 per ADS with gross margin held near 67%, demonstrating that the margin step is structural rather than a utilisation peak.
• 2026 capital expenditure landing at the LOW end of the US$52-56 billion guide with revenue on or above consensus — the combination that turns the free-cash-flow yield from a liability into an asset.
• Compression of the ADS premium toward the Taiwan line below roughly 5%, which would remove one of the two reasons our base sits below the street's lowest target and would be worth about 10% of fair value on its own.
• A price below approximately $360 (the 200-day average, and 16.1x FY2027 consensus) absent a fundamental deterioration — at which point this becomes a size decision rather than an entry decision.

Where TSM fits in the Synthos Framework Portfolio. The semiconductor and compute-infrastructure sleeve, as a 2-2.5% position, built over 4-6 weeks in three tranches. The weight is deliberately below what the growth score alone would justify, and the reason is single-point concentration: this is not a position that can be risk-managed by diversifying within the sleeve, because most of the sleeve depends on the same island. Anyone holding equipment suppliers, fabless designers or memory alongside this name should treat the aggregate as one exposure, not several. Logged as a tracked Synthos call (Buy — Tactical) as of 2026-08-04 at $417.17.

Single biggest risk: the geographic concentration of the world's leading-edge logic capacity. The 20-F names it — the majority of principal facilities and revenue derive from R.O.C. operations, and military conflict, blockade or trade action "could have an adverse impact on our business and results of operations as well as the market price and the liquidity of our ADSs and common shares." It cannot be sized from the financial statements, because no financial statement contains an input that bears on it. The knowledge base sizes the consequence (a global depression lasting 5-8 years, GDP down 20%) but no source sizes the probability, and the panel's own members disagree sharply about it — gavin_baker calls the concern "overstated," niall_ferguson calls it "the defining tail risk." We treat an unquantifiable binary as a reason for a wider valuation discount and a smaller position, not as a reason to avoid the name and not as something to model. That is why the risk score is 6 on a company with US$66 billion of net cash and 285x interest coverage.

Most fragile assumption in the price: that the market continues to apply only a modest discount for that concentration while paying 19-24x forward earnings. Every other assumption in the price has a checkable number attached — the monthly revenue ramp, the margin, the capital budget, the consensus earnings path. This one does not. It is a sentiment variable, it has no floor, and it can reset without a single operating number changing. If the risk discount widens to where the bear case implies, the stock is worth $270 on unchanged fundamentals.

Provenance & disclosures

• Traceability: 132 tagged knowledge-base claims name TSM, TSMC or Taiwan Semiconductor across 29 distinct sources, spanning 2017-11-17 to 2026-07-23, searched across 51,928 records via exact case-insensitive entity match on the aliases TSM, TSMC and Taiwan Semiconductor. Raw hits 132. Excluded: 2 quarantine_misattributed claims (2026-05-04, 2026-02-14), per house policy. Half-weighted: 46 management-class claims from 4 slugs — jensen_huang (38), sam_altman_mgmt (3), lisa_su_mgmt (3), eli_lilly_ceo (2) — of which 38 are bullish, 8 neutral and ZERO bearish. jensen_huang is the chief executive of TSMC's largest customer, a commercially interested party, and a lane 29% voiced by him is not independent conviction. Independent lane: 84 claims from 24 distinct voices — 48 bullish, 28 neutral, 8 bearish (57% / 33% / 10%), with 26 of the 84 dated 2026. The raw count is further inflated by near-duplicates (four jensen_huang claims dated 2026-01-21; three sam_altman_mgmt claims dated 2026-03-11/12). One additional claim is quoted from eurodollar_university (2026-07-30) and is explicitly labelled as NOT entity-tagged to TSM — it appears in an adjacent sector lane and is included because it is the most recent independent bearish reading available. All quotes are verbatim from the stored claim text. kb_breadth in the front matter is set to 24, the independent-voice count, not the 29 raw sources.
• Data as-of: fundamentals — income statement 2026-06-30 (Q2'26, from inc_q, corroborated against the 6-K of 2026-07-13); annual income statement, balance sheet and cash flow 2025-12-31 (FY2025, per the 20-F filed 2026-04-16; there is NO quarterly balance sheet or cash-flow statement in existence for this filer) · estimates 2026-08-04 · prices 2026-08-04 ($417.17, +2.72%; 50-DMA $425.62; 200-DMA $358.40; RSI 49.2; MACD −6.89) · knowledge-base claims 2026-08-04. All figures come from the Synthos vendor data file for TSM or from the SEC filings on disk; no figure comes from memory, recall or external retrieval.
• Filings used: 20-F filed 2026-04-16 (FY2025 annual report — the 10-K equivalent; there is no 10-K and no 10-Q for a foreign private issuer), 6-K 2026-07-13 (June 2026 monthly revenue), 6-K 2026-07-16 (Q2'26 results — exhibit did not extract; only the cover page is on disk and no Q2 detail is used), 6-K 2026-07-24 (June month-end administrative report).
• Data rejections (detailed in Section 8): km_ttm and ratios_ttm rejected in full — TWD-denominated, unlabelled, and struck on the Taiwan-listed ordinary share, understating ADS multiples by roughly 12% (corrected: P/E 29.8x not 26.89x; P/B 10.40x not 9.35x; P/S 15.16x not 13.55x; EV/EBITDA 20.0x not 17.80x; dividend yield 0.85% not 0.95%; free-cash-flow yield 1.65% not 1.86%), with the ADS premium of +14.8% established from same-day dual-security insider transactions on 2026-07-28; inc_a FY2025 revenue (NT$3,848,511M) and net income (NT$1,735,678M) rejected in favour of the 20-F's NT$3,809,054M and NT$1,697,604M, a 1.04% and 2.24% vendor overstatement confirmed by three independent checks including the vendor's own seg_prod; cf_a.capitalExpenditure rejected in favour of the 20-F's NT$1,272,411M (FY2025) and NT$949,817M (FY2023); km_ttm.enterpriseValueTTM corrected from NT$58,011,397M to NT$65,247,246M (US$2,097.3B), an 11.1% understatement arising from both the wrong share line and a netDebt that omits NT$360,441M of short-term investments (corrected net cash NT$2,063,715M / US$66.3B); seg_geo FY2025 flagged as incomplete — Taiwan and "Other" lines missing, 9.7% of revenue unaccounted for; est.ebitAvg FY2025 rejected as 7.3% below actual operating income (est.ebitdaAvg verified and accepted, agreeing within 0.4%); earn_cal USD figures flagged as unreconcilable to the TWD statements at any single rate (implied rates ranging 29.41 to 32.28 against the filing's disclosed 31.11); the quote/tech 52-week range disagreement resolved in favour of tech.
• Currency basis note — the most important methodological statement in this dive: TSMC reports in New Taiwan dollars and its ADS trades in US dollars. All USD conversions use NT$31.11 to US$1.00, the FY2025 weighted-average rate disclosed in the 20-F, chosen because it is the only rate with an auditable source and because it reconciles trailing TWD earnings per ADS-equivalent share (NT$434.95 → US$13.98) with the sum of the four earn_cal USD actuals (US$13.81) to within 1.2%. A 5% move in the NT dollar shifts every USD multiple in this dive by 5% in the opposite direction. The 20-F confirms that "substantially all of our sales are denominated in U.S. dollars while we publish our financial statements in NT dollars," and that the NT dollar appreciated from NT$32.13 (2024) to NT$31.11 (2025), which had an unfavourable effect on reported revenue.
• Share-class note: 1 ADS = 5 common shares (20-F). Common shares outstanding 25,932,524,521, i.e. 5,186,504,904 ADS-equivalent shares, which matches the vendor share count and reconciles market capitalisation to price × shares ($417.17 × 5.1865B = $2,163.6B). Per-share metrics in the vendor file ARE on the ADS-equivalent basis — but denominated in TWD and priced off the local line, which is the defect, not the share count.
• Disclosure-gap note: as a foreign private issuer TSMC files a 20-F annually and 6-Ks as needed — there is no 10-Q, no quarterly balance sheet, no quarterly cash-flow statement and no quarterly segment disclosure. Earnings-call transcripts are not available on this plan, and the Q2'26 results exhibit did not extract. There is therefore NO management commentary of any kind on the June quarter in this analysis, and no interim balance sheet more recent than 2025-12-31. Against that, TSMC's monthly revenue 6-Ks provide a higher-frequency read than any US filer offers, and Section 3's run-rate analysis rests on them.
• Estimate coverage: deep on the near years and thin beyond — 10 analysts on FY2026 and FY2027 earnings, 5 on FY2028, 4 on FY2029; 21, 22, 10 and 15 on the revenue lines. FY2027 dispersion is wide (NT$559.25 low to NT$795.53 high per ADS-equivalent share, US$17.98 to US$25.57, a 42% spread). FY2027 is the primary forward anchor; no conclusion rests on FY2028 or FY2029.
• Peer note: the vendor peer set contains no true comparable. ASML, AMAT and LRCX are equipment suppliers (TSMC's vendors); AVGO, MU and DIOD are designers or memory makers; AMKR, ASX and TSEM are packaging, testing or specialty foundries far smaller; META is not a semiconductor company and its inclusion is a vendor error. No peer multiples are supplied and no peer-multiple comparison is drawn.
• Fair-value caveat: the $270 / $470 / $610 anchors are multiples (12.1x / 21.1x / 27.3x) of FY2027 consensus earnings per ADS of US$22.32, cross-checked against the FY2027 estimate low (15x × US$17.98 = $270) and high (24x × US$25.57 = $614), and against FY2026 and FY2028. Stated arithmetic, not a discounted cash flow. The base assumes a modest multiple expansion from 18.7x to 21.1x and is bracketed by the no-return case ($417, today's multiple held on FY2027) and the full-rolldown case ($526, today's multiple on FY2028). Our base sits 21% below the street consensus target of $596 and 6% below the street's lowest published target of $500, for the reasons given in Section 6c: the ADS-versus-local-line correction, the 1.65% free-cash-flow yield, and a wider discount for an unquantifiable concentration risk.
• Timing: Q2'26 was reported 2026-07-16, nineteen days before this dive, and beat at US$4.31 against US$3.87. The next print is 2026-10-15, 72 days away — no earnings banner applies (the five-day rule is not triggered). The July 2026 monthly revenue 6-K is due around 2026-08-10, within days of publication, and is the near-term event most likely to move the analysis. The price used is the 2026-08-04 16:00 ET close of $417.17, which was up 2.72% on a market-wide semiconductor rally; that timing is treated as an entry problem in Sections 9 and 11 rather than ignored.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person.
• Version: 2026-08-04-full.