TSM: a stronger cash quarter still needs a valuation test

The 60-second read

Five things to know

1. The headline understates the operating comparison. Q2 contained a NT$63.2bn VIS disposal and remeasurement gain. Q3 non-operating income returned to NT$34.59bn while revenue rose 17.6% sequentially.
2. Cash improved materially. Quarterly operating cash exceeded attributable profit, and free cash flow more than tripled from the year-earlier period. It still pays for a capital-intensive expansion.
3. Mix is getting more concentrated. HPC reached 68% of revenue. This is broader than AI accelerators; the company does not publish profit by platform.
4. A good business can be fully valued. The normalized cash model gives about $318, versus $489 from a forward earnings multiple. We publish both rather than hide the disagreement.
5. Reporting bases matter. Quarterly figures use Taiwan-IFRS. IASB annual tax and attributable earnings differ. We keep those bases separate and never label the difference a company non-GAAP add-back.

Three scenarios, one probability-weighted number

Analyst scenarios for the next twelve months; probabilities are judgments, not frequencies. The anchor and probability-weighted outcome are separate.

Signposts: what would change our mind

Thresholds are our monitoring rules. A lower share price does not override broken demand, export restrictions, governance problems or a weaker cash outlook. These are the first signposts in this packet series; no prior TSM packet score is invented.

The tape

An intraday reaction is a timestamped observation, not a closing verdict.

At 8:35 a.m. Chicago time on October 15, the U.S. ADS traded at $411.27, down 4.36% from the previous $430 close. That is an early-session snapshot. The release was available before the U.S. open, but the exact issuer publication clock was not independently established. The SEC acceptance timestamp is not substituted for that clock. The price could change materially before publication of this packet or the end of trading.

The first question is not whether the price fell after good numbers. It is what the price already required. A business can beat its own guidance while investors reduce the multiple, react to capital spending, hedge an existing position or respond to unrelated market news. Our reaction history measures close-to-close returns on earlier release dates; it does not isolate the earnings surprise from those other influences. The current point is deliberately separated from that completed-session history.

The quote vendor’s capitalization is close to our share-count calculation, but the packet uses a transparent denominator: 25.932 billion ordinary shares divided by five equals 5.1864 billion ADS equivalents. This is not the number of receipts actually deposited in the U.S. facility. It is the denominator that makes consolidated earnings and an ADS price comparable. Spot FX is used for valuation translation; quarter-average FX is retained for historical operating comparisons.

The print

Fast growth, a real guidance beat, and a harder fourth-quarter hurdle.

Third-quarter revenue reached NT$1,494.24bn, up 50.9% in the reporting currency and 41.2% in U.S. dollars. The difference is meaningful: the quarter-average exchange rate was NT$31.98 per dollar versus NT$29.91 a year earlier. Dollar growth better approximates the commercial demand comparison for a business whose sales are substantially dollar-linked. Neither growth rate is wrong; they answer different questions. We retain both instead of presenting the larger number without its currency basis.

Revenue of US$46.72bn exceeded the prior US$44.6–45.8bn guide, finishing 3.36% above its midpoint and above the top end. Gross margin of 66.4% was inside the 65–67% range. Operating margin of 58.7% was above the 56–58% range. This comparison is against management’s own pre-quarter range, not a reconstructed analyst consensus. The quarterly consensus endpoint was unavailable, so no claimed consensus beat is manufactured from annual estimates.

For Q4, management guides to US$52.3–54.3bn revenue, 65–67% gross margin and 56.5–58.5% operating margin using NT$31.8 per dollar. The revenue midpoint would put the year at US$176.12bn when added to the first three reported quarters. That agrees directionally with management’s wording of around 44% full-year dollar growth. This is a guidance-based bridge, not an audited full-year result or a claim that every order is irrevocable.

The midpoint also implies further sequential revenue growth while operating margin is lower than Q3’s achieved level. That combination prevents a simple extrapolation of the best margin quarter. New-node ramps, overseas capacity, mix and exchange rates can all change the conversion from revenue to profit. We do not assign exact percentage-point contributions without a verified issuer bridge. The useful test is whether the company delivers the revenue range while keeping the guided margin discipline.

Platforms and capacity

A single operating segment, with increasingly concentrated demand.

TSMC reports one operating segment. Its platform presentation separates revenue into high-performance computing, smartphones, Internet of Things, automotive, digital consumer electronics and other applications. These are demand categories, not separately reported profit centers. The packet therefore shows platform revenue and consolidated operating profit side by side, but does not invent an HPC margin or allocate corporate research spending to individual customers.

HPC represented 68% of Q3 revenue, compared with 66% in Q2 and 57% a year earlier. Smartphones represented 21%, with IoT and automotive at 4% each, digital consumer electronics at 1% and other applications at 2%. Multiplying rounded mix percentages by total revenue gives approximate platform revenue, not an audited segment table. HPC includes several forms of computing; treating the entire category as AI accelerator revenue would overstate what the disclosure establishes.

The technology chart gives a second concentration lens: 2-nanometer contributes 8%, 3-nanometer 33%, 5-nanometer 28% and 7-nanometer 10%. Together, 7-nanometer and more advanced nodes represent 79% of wafer revenue. This is not the same denominator as platform demand. A technology share describes manufacturing processes; an HPC share describes applications. Adding the two percentages or treating them as separate growth pools would double count activity.

Shipments rose 9.6% year over year while reporting-currency revenue rose much faster. More valuable process mix, packaging and currency matter to that gap. We cannot infer an exact like-for-like wafer price increase from the aggregate figures. The practical risk is that future capacity must earn attractive returns after the construction and equipment bills arrive. Strong demand today supports investment, but it does not eliminate utilization risk several years into a fab’s life.

Customer concentration compounds that capital-cycle exposure. In the 2025 annual report, the top ten customers represented 78% of revenue; the largest two represented 19% and 17%. The company does not identify those customers in that disclosure, and this packet does not attach names to the percentages. A concentrated customer base can support close technology planning, but it also means that a few customers’ product schedules, inventory decisions or export permissions can influence the consolidated outcome.

The profit bridge

The prior investment gain changes the sequential interpretation.

Attributable net income increased 8.0% sequentially to NT$763.06bn, slower than revenue. The first explanation to inspect is below the operating line. Q2 non-operating income was NT$95.83bn, including NT$63.202bn from the completed disposal and remeasurement of the VIS investment. The remaining stake became a fair-value-through-other-comprehensive-income investment after TSMC lost significant influence. Part of the recognized gain was a noncash remeasurement. It should not be treated as repeatable foundry earnings.

In Q3, total non-operating income was NT$34.59bn. That normalization reduces the apparent sequential earnings growth even though operating profit rose. The selected release does not yet provide every current-quarter interest, FX, investment, tax and minority-interest line. We therefore show an approximate operating bridge using the rounded 58.7% operating margin and a combined tax-and-NCI residual. That residual is not an effective tax rate, and it must not be described as a tax charge alone.

The annual comparison deserves its own guardrail. The Taiwan-IFRS annual filing reports FY2025 tax expense of NT$326.27bn and attributable earnings of NT$1,717.88bn. The IASB 20-F reports NT$346.53bn and NT$1,697.60bn. The tax policies include different recognition treatment for unappropriated earnings, with small associate differences elsewhere. We preserve the two bases rather than splice the 20-F earnings into the quarterly series and create a false growth or EPS trend.

The economic lesson is more useful than a mechanically adjusted EPS number. The core operation is growing strongly; a prior investment gain makes sequential net growth look weaker; and taxes and ownership allocation still affect what belongs to the ADS holder. We do not apply a speculative tax rate to the VIS gain and call the result company-adjusted earnings. Our forward earnings route instead uses an explicit normalized net-margin assumption, with the cash route acting as a separate check.

Earnings quality

Strong cash evidence, with disclosure gaps kept visible.

Operating cash of NT$919.77bn equals 120.5% of attributable profit, compared with a weaker year-earlier conversion. This is encouraging, but a single quarter is sensitive to collection dates, tax payments, bonuses, supplier terms and customer-related balances. Receivable days improved sequentially to 28; inventory days improved to 82. Both remain above the year-earlier levels of 25 and 74. The combination is a favorable sequential trend, not proof that working capital cannot reverse.

Three-year cash-flow history gives a useful counterweight to the latest quarter. Operating cash rose from NT$1,241.97bn in 2023 to NT$1,826.18bn in 2024 and NT$2,274.98bn in 2025. Capital spending also rose, reaching NT$1,272.41bn in 2025. The difference was NT$1,002.57bn of issuer-defined free cash flow. These statements place interest received in investing and interest paid in financing, which is one reason the simple OCF-minus-capex measure should not be mislabeled as a fully reconciled equity cash flow.

SBC is small in the available annual cash-flow adjustment, but that does not mean people costs are immaterial. The annual report describes substantial cash profit sharing and a growing workforce, reaching 90,557 full-time employees at the end of 2025. Cash bonuses already affect operating expense and cash timing. Adding them back because they fluctuate with profit would remove a real cost of the business. We also avoid adding depreciation back in valuation without charging for recurring replacement and expansion investment.

The current detailed interim filing remains the next verification point. It should allow a proper separation of taxes, noncash items, minorities and working-capital movements. Until then, the packet keeps unavailable cells visibly unavailable. A missing granular disclosure is neither a zero nor automatically a red flag. The existing cash and margin disclosures are strong enough to form a provisional valuation judgment, while the signposts specify what additional evidence could change it.

Balance sheet and cash

A large liquidity cushion is still small beside the equity price.

At September 30, cash and equivalents were NT$3,621.89bn, while cash plus marketable securities were NT$4,016.98bn. Those are different measures. The latter includes investments and is not an unrestricted cash balance. The selected release also lists NT$818.38bn of long-term interest-bearing debt, but that is not total debt because current maturities and other obligations need separate treatment. Subtracting only that long-term line from the latest cash would overstate a clean net-cash claim.

For valuation, we use the fully detailed June 30 balance sheet and a deliberately conservative cash-only bridge. Cash of NT$3,134.22bn is reduced by NT$1,064.95bn comprising current long-term liabilities plus noncurrent bonds, bank loans and leases. The resulting NT$2,069.27bn translates to US$63.85bn at the dated spot rate, or about $12.31 per ADS. We exclude marketable securities and equity holdings from this bridge and do not add the later cash increase while retaining stale debt. This is a dated analytical convention, not reported September net cash.

Quarterly free cash flow covered the NT$155.59bn cash dividend, leaving capacity for other investment and financing uses. That does not make every remaining dollar distributable. New fabs require committed equipment purchases, cash taxes and working capital; subsidiaries have their own obligations; and geographic expansion can require capital before revenue arrives. Shareholders benefit from the liquidity, but valuing the cash dollar-for-dollar while ignoring those commitments would be too generous.

The annual report shows one vote per ordinary share and five ordinary shares per ADS. ADS holders exercise rights through the depositary arrangements, which are not identical to directly holding the local shares. The company’s major operations are ownership-consolidated subsidiaries in the reviewed filings. We do not apply a fabricated VIE deduction, nor do we pretend that legal ownership removes geopolitical, regulatory, currency-conversion or minority-interest risks. Those risks belong in the valuation assumptions and margin of safety.

Three valuation routes

The disagreement between cash and earnings models is the point.

Our anchor is $410 per ADS, essentially in line with the $411.27 reference quote. Conviction is three out of five and uncertainty is high. The three routes are not three independent measurements of a knowable price. They expose different assumptions about reinvestment, margins and the valuation multiple. We give the cash model 40% weight and each of the other routes 30%, obtaining $412.85 before rounding the published anchor. The route range is $318–$489, distinct from the wider bear and bull scenario outcomes.

The cash model is intentionally demanding about reinvestment. A 35% normalized free-cash-flow margin already assumes strong economics through a large expansion. It is not derived by simply annualizing Q3 cash flow. Terminal value supplies 75.5% of the operating present value, so the 10% discount rate and 3% terminal growth rate matter enormously. A lower rate or a longer period of rapid growth can lift the answer, but those changes should be visible assumptions rather than hidden ways to match a preferred target.

At the reference price, the same revenue path, discount rate and terminal-growth assumptions require a normalized free-cash-flow margin of approximately 45.7% rather than our 35%. That reverse calculation is not a forecast and does not mean the company must literally report that exact annual margin. It shows the cash burden embedded in this particular model. The trailing issuer-defined FCF yield is only 2.14%, so sustained growth and future cash conversion do much of the valuation work.

A high-uncertainty business does not become a BUY merely because the earnings route looks attractive. Our entry discipline requires 25% below the $410 anchor, or $307.50, with the operating thesis intact. That is a valuation threshold, not a promised order or automatic recommendation to buy during a crisis. Conversely, stronger sustained cash conversion could raise the anchor before the stock reaches that price. The signposts make that alternative explicit rather than anchoring the decision permanently to today’s model.

Where the Street differs

A fresh target is useful context, not an independent valuation proof.

The provider reports a $582.71 consensus price target, with a $560 median and a $500–$700 range. Its rating snapshot contains 18 buys and seven holds. These aggregates combine analysts with different dates, models and horizons; they are not a fresh post-result poll. A dated October 15 Needham record raised the target to $560. That is the freshest record in the collected feed, while several other targets still date from July, August or September. We preserve that age distinction.

Our main 2027 revenue assumption is US$220bn, slightly below the provider’s NT$7.37tn annual revenue estimate translated at the same spot rate. More importantly, our cash route refuses to value every dollar of accounting profit as immediately distributable. That is why our $410 anchor is below the aggregate target despite an optimistic operating outlook. This is a difference in cash conversion and valuation discipline, not a claim that the Street has missed the existence of AI demand.

The annual consensus table is context, not a substitute for primary accounts. The provider’s historical statement feed disagreed with the issuer on FY2025 and Q4 revenue, so those historical financial rows were excluded from the core dataset. We also avoid treating a provider EPS field as dollars when its scale reflects a different currency or receipt convention. For our earnings model, consolidated dollar net income is divided by the explicit ADS-equivalent share count, making the denominator and currency reproducible.

Management scorecard

Execution is strong; capital discipline must survive the next phase.

Management deserves credit for delivering revenue above its own Q3 range and operating margin above the high end, while keeping gross margin within its guide. The Q4 outlook extends the growth trajectory. It also creates a clear hurdle that can be tested in the next release. A scorecard is more useful than accepting every confident statement at face value: we separate delivered numbers from plans and give capital allocation the same attention as revenue growth.

The completed VIS stake transaction is a useful example of why accounting and allocation should be discussed together. It generated a large recognized gain and changed the classification of the remaining holding. That is different from selling more leading-edge wafers. The Sony joint venture announced in August is another distinct capital decision: TSMC is a minority investor, and regulatory completion and execution remain relevant risks. Neither transaction is automatically evidence that the foundry’s recurring margin has improved.

1. How much of the Q4 margin range reflects new-node ramp costs, overseas fabs, mix and currency, and which components are expected to reverse?
2. What sustained cash-flow margin should shareholders expect once the present capacity expansion reaches normal utilization?
3. How much of customer-related cash timing supports current OCF, and what obligations accompany those balances?
4. Which investments are genuinely discretionary if demand slows, and which equipment or construction commitments remain unavoidable?
5. What milestones will determine whether the Sony minority investment earns an acceptable return without distracting from core capacity needs?

The annual auditor is Deloitte & Touche, and the reviewed 20-F contains unqualified opinions on the IASB financial statements and internal control over financial reporting under PCAOB standards. This packet does not transform that audit opinion into a claim about a specific current inspection result. Likewise, voting rights, ownership structure and a long operating history reduce some uncertainties without eliminating export-policy, Taiwan-location or capital-cycle risks. A credible management team still needs a purchase price that compensates the investor for those risks.

Risks and competing cases

The strongest bear case accepts that the business is excellent.

The bull case is straightforward: leading-edge demand remains strong, the new capacity fills, advanced-node mix supports pricing, and cash conversion catches up with accounting profit. In that outcome, the earnings route may be the better guide than a conservative five-year cash model. The bear case does not require a collapse in technological competitiveness. It requires only that reinvestment remain heavy, cash margins fail to reach the embedded expectation, or investors demand a larger risk premium on very long-dated growth.

Geopolitical and export-policy risks are not well represented by a small, tidy probability plugged into one spreadsheet cell. Their severity can overwhelm an ordinary earnings model. We reflect that uncertainty through a high rating, a wide scenario range and a demanding entry cushion, while acknowledging that no discount makes every tail risk acceptable to every investor. This packet is a company valuation analysis, not a portfolio suitability assessment.

Catalysts and next checks

Dates are labeled according to what is actually known.

The next catalyst need not be a dramatic announcement. A detailed filing can materially improve the quality of the cash and tax analysis even if it does not change the headline earnings. Monthly revenue can help assess the quarterly guide, but it cannot establish quarterly profitability or cash conversion on its own. We will distinguish actual results from calendar expectations and will not use a scheduled call date as proof that financial statements have been released.

The decision rule is deliberately two-sided. A price near $307.50 would provide the specified 25% cushion to the current anchor, provided the fundamentals remain intact. Sustained cash margins above the model could justify a higher anchor without such a pullback. Conversely, a deterioration in utilization, working capital or export access could reduce fair value even after a large share-price decline. The next review should update the assumptions with evidence rather than mechanically repeat today’s rating.

Appendix and method

Every denomination, unavailable item and model assumption stays visible.

Method: primary earnings releases and presentations supply the consistent nine-quarter selected series. Detailed interim filings supply historical opex, below-the-line, balance-sheet and cumulative cash-flow components. Standalone Q2 and Q3 cash flows are computed as current cumulative cash flows less the preceding cumulative period; Q4 is annual less nine months. Annual IASB statements are retained separately for structural and reporting-basis checks. All source tables and transformations remain in the underlying dataset.

Rounded presentation margins can create small differences when multiplied by revenue. The current operating-profit bridge is explicitly approximate; the combined tax-and-NCI residual cannot be split without additional statements. Platform revenue is calculated from rounded percentage shares and is labeled approximate. Platform operating profit is unavailable because the issuer has one reportable segment. We do not turn an unavailable segment figure into an assumed zero or a proportional allocation.

Currency method: quarterly dollar revenue uses the issuer’s own presentation. The valuation uses NT$32.409 per U.S. dollar from the same market snapshot as the ADS quote, while historical quarter-average rates remain in the source series. A U.S. ADS represents five ordinary shares. The share denominator is 25.932 billion ordinary shares divided by five. The June cash-only bridge is deliberately dated and excludes investments; it is not updated with a later cash balance while retaining older debt.

Model method: the DCF uses analyst revenue, cash-margin, discount and terminal-growth assumptions. The earnings and operating-profit routes use explicit forward margins and multiples. They are not issuer forecasts. The three-route weighting produces a rounded fair-value anchor; scenario probabilities produce a separate expected value. The sensitivity grid varies initial growth and cash margin while holding other inputs fixed. It does not represent every possible combination of economic and political outcomes.

Data limitations: the quarterly analyst-estimates endpoint was unavailable. Annual consensus figures are shown with their currency scale and analyst counts, but ambiguous provider EPS fields are not used. Historical provider statement discrepancies are preserved in the research audit and excluded from authoritative financial series. The current price is intraday, not a close. No claim is made that the observed share-price change was caused solely by this release. There is no previous packet signpost score to grade.

• Current Q3 release and presentation
• Detailed Q2 2026 statements
• FY2025 Taiwan-IFRS annual statements
• FY2025 IASB 20-F and structure
• FMP market and analyst data

Glossary: OCF is cash generated by operating activities. Issuer FCF is OCF less cash capital expenditure on property, plant and equipment. ADS means American depositary share. NCI is the portion of consolidated income or equity attributable to outside owners of subsidiaries. HPC is the issuer’s high-performance-computing application category. EV is equity value less the specified net-cash bridge. A terminal value estimates cash flows beyond the explicit model period and is highly sensitive to assumptions.