What we said in September — and what happened
The business beat the plan; the price fell through our level; four facts were wrong
On 28 September we published TSMC at $451.95, a record high, with a verdict that held two thoughts at once: one of the finest businesses on earth, and a position to be sized for the risk across the Taiwan Strait. We said to accumulate, and to add with both hands near $380, about 17 times what analysts then expected TSMC to earn in 2027. On 5 October the ADR closed at a record $485.80. On 5 November it closed below $380 for the first time, and on 20 November at $319.15, 16% below our level. Here is the account.
| What we said on 28 Sep | What happened by 20 Nov | Grade |
|---|---|---|
| 'Add with both hands near ~$380' | Reached on 5 November; the price is now 16% below it. A buyer at $380 has lost 16% while the business grew. | ✗ Early on price |
| Q3 guide $44.6–45.8bn; TSMC 'tends to beat' | Q3 revenue $46.72bn, +41% in dollars, above the top of the range. | ✓ Right |
| 2026 growth 'slightly above 40%' | Raised to around 44% on 15 October; October sales +60%. | ✓ Beaten |
| Gross margin lower in Q3 as 2nm ramps | 66.4%, down 1.3 points, 'primarily due to margin dilution from 2-nanometer ramp' (finance chief); 2nm 8% of wafer revenue. | ✓ Right |
| Capital spending $60–64bn | Reiterated; 2026–28 to be 'significantly higher than the past three years', no number for 2027. | ◆ Rising |
| A rising dividend | NT$9.0 a share for Q3, +28.6%. | ✓ Right |
| Believe: Arizona earns acceptable returns | Overseas factories cut gross margin by ~2% this year, rising to 3–4% 'in the next few years'. | ◆ Amber |
| Believe: no blockade closes the Strait | No blockade; drills after President Lai's October speech and 'area denial' drills confirmed by Taiwan's defence minister on 10 November (as reported). | ◆ Holding; visible |
Our errors. Four facts in the September report were wrong, and two claims need amending. First, we wrote that US customers were about 82% of TSMC's 2025 revenue; the 20-F gives 74.4%, with China 8.6%, Taiwan 7.9%, Japan 3.9% and Europe, the Middle East and Africa 3.3%. Our feed left out Taiwan and "others", which inflated every other share (Part IX). Second, we presented the rise in earnings per ADR from $2.92 to $4.31 in three quarters as the run-rate of the business; the second quarter included about NT$63bn of gains on TSMC's stake in Vanguard International, which added NT$2.24 to earnings per share. Without it the second quarter was about $3.96 per ADR, still a fine quarter. Third, we compared a dividend of "$0.48 at the end of 2024" with "$1.11" now; the first figure was after Taiwan's 21% withholding tax and the second before it. Like for like, the quarterly dividend rose from NT$4 to NT$7 a share, +75%, not +131%. Fourth, we said the ADR dividend "paid in September" was $1.11; it went ex-dividend on 16 September and was paid on 8 October, at about $1.10.
The two amendments. We wrote that TSMC "may import up to 2.5 times" its US capacity duty-free during construction; we could not find a primary source for that figure and withdraw it until we can. And we said Berkshire Hathaway sold "nearly all" of its TSMC stake by early 2023; according to its filings, it sold all of it. Everything else we re-checked, including the second-quarter figures, the node mix, the guidance, the capital spending, the $265bn of commitments in Arizona, the tariff terms, the dividend amounts, the net cash and the names and dates of management, was correct.
★ A record in October, a rout in November
Seven weeks in which the news from TSMC was good and the price fell by a third
| Date | What happened | TSM that day |
|---|---|---|
| 5 Oct | Record close, two weeks after our X-Ray. | +2.8%, $485.80 |
| 15–16 Oct | Q3 results above the guide; growth for 2026 raised to ~44%. The shares fell over two days. | −1.8%, −2.3% |
| 30 Oct | Intel −29% on its guidance, after Alphabet's fall the day before (as reported). | −6.0% |
| 5 Nov | Chip shares' second-worst day on record (SOX −12.25%); fears of delays in AI data centres (as reported). | −10.8%, $353.10 |
| 11 Nov | October revenue +59.7%. | +4.3%, $413.35 |
| 16 Nov | Nvidia's chief executive resigned on 12 November; the Taiex fell 6% (Focus Taiwan). | −6.3% |
| 20 Nov | Nvidia −14% after Reuters reported its accounting was the focus of an SEC probe (via CNBC); Nasdaq's worst week since March 2020. No TSMC news. | −9.4%, $319.15 |
Not one bad day was about TSMC. Read down the list of large moves and a pattern appears. On the two days that carried TSMC's own news, its results and its October sales, the news was good. The large falls came from Intel, from fears about the pace of AI data-centre building, from turmoil at the largest buyer of advanced chips, and from a sell-off that took the Nasdaq to its worst week in more than six years. Chip shares as a group, measured by the SOXX fund, fell 34% over the same period; TSMC's ADR fell about five points less.
That does not make the fall irrelevant. The market's question is not whether TSMC had a good quarter, but whether its customers will keep spending at this pace, and TSMC's revenue is now 68% high-performance computing, mostly AI. When the market doubts the buyers, it marks down the supplier. But the evidence so far is that the supplier is still being paid more each month than the month before, and that its order book is the best of any company in the chain.
Where the price sits. At $319.15 the ADR is 5% above its close at the end of 2025 and 0.8% above its 2026 closing low of $316.50, set on 30 March. In other words, the whole of the year's rally has been undone, while the company's growth forecast for the year went from "close to 30%" to "around 44%".
★★ The business ran ahead of the plan
A beat, a raised forecast, record monthly sales and a 29% higher dividend
On 15 October TSMC reported third-quarter revenue of $46.72bn, up 41% in dollars and above the top of its own range of $44.6–45.8bn. In Taiwan dollars revenue rose 51% and net income 69%. The gross margin was 66.4% and the operating margin 58.7%, above the guidance. Free cash flow was NT$477bn, about $15bn, after capital spending of $13.8bn. Then the monthly figures continued the story: September sales up 54.6% on a year earlier and October up 59.7%, the highest month in the company's history. For the fourth quarter TSMC guides revenue of $52.3–54.3bn, about 58% above a year earlier at the midpoint.
Through 2026 the forecast for the year rose at every call: "close to 30%" in January, "above 30%" in April, "slightly above 40%" in July and "around 44%" in October. A company that raises its forecast four times in a year is either very conservative or seeing demand it did not expect; in TSMC's case it is both. The chairman said on the October call that demand for advanced chips is still "higher than the supply" and that the growth of AI accelerators, which TSMC expects at a rate in the "mid- to high 50s" a year from 2024 to 2029, "probably will be higher".
| Q3 2026 | Value | vs a year earlier |
|---|---|---|
| Revenue | $46.72bn (NT$1,494bn) | +41.2% in US$; +50.9% in NT$ |
| Gross / operating margin | 66.4% / 58.7% | 59.5% / 50.6% |
| Earnings per ADR | $4.60 | $2.92 |
| High-performance computing | 68% of revenue | +81.5% in NT$ |
| Wafer revenue at 7nm or smaller | 79% (2nm 8%, 3nm 33%) | 77% in Q2 |
| Free cash flow | NT$477bn (~$15bn) | NT$139bn |
| Q4 guide | $52.3–54.3bn; GM 65–67% | ~+58% at the midpoint |
The margin, honestly read. The gross margin fell from 67.7% in the second quarter to 66.4% in the third, and TSMC guides 65–67% for the fourth. The finance chief gave two reasons: the new 2-nanometre lines, which are expensive until their volumes grow, and the factories abroad. The second is the more lasting. TSMC expects its overseas factories to cut the gross margin by about 2% this year and by "3% to 4% in the next few years", as Arizona, Japan and Germany take a larger share of output. Its own long-term target is a gross margin of "56% and higher". At two-thirds of sales today, there is a long way to fall before the target is tested. But the direction is set: TSMC is being paid by governments and customers to build outside Taiwan, and the price of that safety is a lower margin.
The prudent word. One phrase from the chairman deserves an owner's attention: consumer and price-sensitive products are "challenged due to the impact of rising component prices", and TSMC is "being prudent" in its planning for 2027. Smartphones are 21% of revenue. If AI spending slowed at the same time as consumer demand, TSMC would feel both. Neither has happened yet.
★ Where the 29% went
Taipei, the Taiwan dollar and the ADR premium; not the profits
An American owner of TSMC holds three things at once: a share in the company, priced in Taipei; an exposure to the Taiwan dollar, in which the company reports and pays; and a US-listed certificate that usually trades at a premium to the Taipei shares. All three moved against the owner in the last eight weeks, and taking them apart shows how little of the fall was about the business.
The Taipei shares fell from NT$2,475 to NT$2,000, 19.2%, with the Taiwan market (the Taiex lost 6% on 16 November and 4.4% on 19 November). That accounts for about $87 of the ADR's $133 fall. The Taiwan dollar weakened from 31.78 to 32.94 to the US dollar, 3.5%, which took about $13. The premium that American buyers pay over Taipei shrank from about 16% in September to about 10% on 19 November; on 20 November it looked smaller still, about 5%, but only because Taipei had closed before the US fall of 9.4% that day. Together the premium and the last day took about $33.
| Piece | 28 Sep | 20 Nov | Effect on the ADR |
|---|---|---|---|
| Taipei shares (2330) | NT$2,475 | NT$2,000 | −$86.7 |
| Taiwan dollar per US$ | 31.78 | 32.94 | −$12.9 |
| ADR premium over Taipei | ~16% | ~10% (19 Nov); ~5% (20 Nov)* | −$33.2 |
| Consensus 2027 earnings per ADR | ~$21.8 | ~$22.7 | +4% |
What this means for an owner. The premium is the least understood of the three. It exists because many American investors cannot or will not buy shares in Taipei, and it rises when they are eager and falls when they are not. Buying the ADR at a 16% premium in September meant paying 16% more for the same company than a buyer in Taipei. At 10% the premium is closer to its usual range, and its fall has been part of the cost of owning the ADR this autumn. It can fall further. It can also rebuild, and when it does it will add to the return of the ADR over and above the company's.
The fourth line of the table is the one that matters most. Analysts' expectations for TSMC's 2027 earnings rose while the price fell. The share price and the earnings have moved in opposite directions, which is how a price becomes cheaper without the business becoming worse.
The customers: two that matter, and the week one of them shook
Top ten 78% of revenue; the largest end-buyer's turmoil hit the supplier
TSMC's 20-F does not name its customers, but it measures them. In 2025 its ten largest customers provided 78% of revenue; the largest 19% (NT$727bn) and the second 17% (NT$645bn). TSMC does not name them, and we do not guess which is which. Either way, two companies provide more than a third of the revenue, and high-performance computing, the AI spending the market now doubts, provides two-thirds.
The Nvidia week. Nvidia, whose most advanced chips TSMC makes, had a difficult November. Its chief executive resigned on 12 November, as reported by Focus Taiwan. On 16 November the Taiex fell 6% and TSMC's ADR 6.3%. On 20 November Nvidia's shares fell 14% after Reuters reported, according to CNBC, that Nvidia's accounting was the focus of an SEC probe; that is a report, not a finding. TSMC's ADR fell 9.4% the same day. Focus Taiwan quoted investors' worry that Nvidia, under new leadership and pressure, might squeeze its suppliers. That is a possibility, not a fact: TSMC has no rival that can make Nvidia's most advanced chips at its scale, and a customer under pressure has less power to bargain, not more, with the only supplier able to serve it.
| Concentration | Figure | Read |
|---|---|---|
| Ten largest customers, 2025 | 78% of revenue | ◆ High |
| Largest / second customer | 19% / 17% (unnamed) | ◆ Two-thirds of a third |
| High-performance computing, Q3 | 68% of revenue, +81.5% | ◆ AI-led |
| Customers headquartered in the US, Q3 | 78.9% of revenue | US policy matters as much as Taiwan's |
| TSMC's own stakes in AI buyers (approved 10 Nov) | Up to $500m in Anthropic at its IPO; up to $500m of OpenAI preferred | ◆ Small, but new |
A new habit to watch. On 10 November TSMC's board approved investments of up to $500m in Anthropic's shares at its IPO and up to $500m in OpenAI's preferred shares. These are AI laboratories whose spending drives demand for TSMC's customers' chips. A billion dollars is small beside TSMC's $90bn of net cash. But in our Broadcom report we measured how suppliers have begun to finance the buyers of their products. TSMC has until now stayed outside that loop. These two investments are a small step towards it, and we will watch for a larger one.
The moat, rechecked
Wider than in September; the cost of building it abroad is rising
| The claim (September) | Evidence now | Verdict |
|---|---|---|
| A process lead rivals cannot close | 2nm in volume 'with good yield'; capacity to more than double in 2027; 7nm and below up from 77% to 79% of wafer revenue. | ✓ Wider |
| Advanced packaging as a second moat | The chairman: 'we are the only one' producing 5x-reticle advanced packaging today (Q3 call). | ✓ Intact |
| Pricing power | Gross margin 66.4% with 2nm and overseas dilution; 'tool price inflation is real', and so is TSMC's ability to pass it on. | ✓ Strong |
| Intel and Samsung as foundries | Intel −29% on 30 October on its own guidance; nothing new from Samsung's foundry. | ✓ No threat yet |
We keep the moat at 10. Nothing in the last eight weeks narrowed it, and the 2-nanometre ramp, the doubling of its capacity planned for 2027 and the first risk production of the next node (A14) in 2027 all widen it. The moat's cost is a different matter. TSMC has "nearly 20 fabs worldwide in progress", 13 in Taiwan and five or six abroad, and its capital spending for 2026–28 will be "significantly higher" than in the past three years. A moat that must be rebuilt at every node, and now on three continents, is a moat with a large maintenance bill. TSMC can afford it; the owner should know it exists.
- That AI spending by TSMC's customers keeps growing through 2027, so that earnings reach roughly the $22–23 per ADR analysts expect, about 35% more than this year.
- That overseas factories cut the gross margin by 3–4 points at most, as TSMC says, and not by more.
- That no blockade or conflict closes the Strait, and that you own only as much as you could bear to lose if one did.
The competition and the map
No rival gaining; a larger military presence; tariffs unchanged
| Pressure | What changed since September | Read |
|---|---|---|
| Intel Foundry | Shares −29% on 30 October after a weak Q4 guide; analysts raised Intel's EMIB-T packaging on TSMC's call. | ✓ No gain |
| Samsung Foundry | Nothing new that we could source. | ✓ No gain |
| The Strait | Five days of drills after President Lai's 10 October speech; 'area denial' drills confirmed by Taiwan's defence minister on 10 November (as reported). No blockade and no named exercise. | ◆ Louder |
| US tariffs | Section 232 duty of 25% on certain advanced chips (January 2026) with exemptions; the 15% US–Taiwan rate stands; nothing new in October or November. | ◆ Unchanged |
| US export controls | A Commerce rule of 15 November on remote access to advanced AI chips adds due-diligence duties for foundries and packagers (WilmerHale); TSMC is not named. | ◆ More paperwork |
The map, again. In September we wrote that the one risk no dial measures is geography, and we gave TSMC a score lower than its dials because of it. Nothing in the last eight weeks has changed that judgment, but the risk has been more visible: drills in October, "area denial" drills confirmed in November, and a Taiwan stock market that lost 6% on 16 November and 4.4% on 19 November. None of this is a blockade. All of it is a reminder that TSMC is building 13 of its new factories in Taiwan and five or six abroad, so that in absolute terms more of the world's advanced chip-making is concentrated on the island, not less.
Our rule from September stands without change: own as much TSMC as you could bear to lose if the island were cut off, and no more. That rule does not depend on the price. At $319 it lets an owner buy more of the business for the same amount of money, but it does not let an owner put more money at risk.
Management, insiders & capital allocation
Directors buying, the dividend rising, and $147bn of approved spending
Capital allocation. We lower the score from 9 to 8, for one reason and with one caution. The reason is the new habit described in Part V: up to $1bn in the shares of two AI laboratories, small in money and large in principle. The caution is the scale of commitments. In 2026 TSMC's board approved about $146.6bn of capital appropriations across four meetings, including $40.9bn on 10 November, a further US$20bn capital injection into its Arizona subsidiary in May, and a Japanese joint venture with Sony. Capital spending this year is $60–64bn, and the company has declined to give a number for 2027 except that it will be higher. A company earning a return on equity of 45% should invest every dollar it can at those returns. The owner's risk is the cycle: semiconductor history is full of companies that built capacity at the top of a boom.
The rest is exemplary. Net cash rose to about NT$2,985bn, roughly $90.6bn, $17.47 per ADR. The dividend rose 29%. Capital spending, though rising, was $13.8bn in the third quarter against $15bn of free cash flow after it. And the company told owners in plain words what it does not know: the 2027 capital budget, the full cost of the overseas factories, the effect of component prices on consumer demand.
Insiders. Since 28 September the 67 insider transactions reported on Forms 4 have all been purchases; there have been no sales. Most are small monthly purchases that look like regular plans. Four are discretionary: the two directors named above, and two vice presidents. Insider buying at TSMC is modest in size, but the direction during a 29% fall is unambiguous.
The numbers
Per ADR, in dollars, and six faults in our feed
In the right units. TSMC reports in Taiwan dollars per share; Americans own ADRs worth five shares, priced in US dollars. Every figure in this report is per ADR in US dollars unless marked. Over the twelve months to September TSMC earned $15.54 per ADR (adding the company's own quarterly US$ figures), on revenue of about $157bn. Free cash flow was about NT$1,481bn, roughly $45bn or $8.67 per ADR, after capital spending that is now about 35% of revenue. Return on equity in the third quarter was 45.5% annualised.
The quality of earnings. One adjustment matters. The second quarter included about NT$63bn of gains on TSMC's stake in Vanguard International, which we failed to flag in September. Without it, trailing earnings are about $13.51 per ADR rather than $13.86 at the time of our report. The third quarter had no gain of that size; its non-operating income of NT$34.6bn was close to normal. Taxes, depreciation and capital spending all look ordinary for a business growing at 40%.
The balance sheet. Cash and marketable securities of NT$4,017bn (about $126bn, as the finance chief put it) against bonds and bank loans of NT$1,032bn: net cash of about $90.6bn, 5.5% of the market value. Interest is covered about 277 times by operating profit. There is nothing to worry about here except the size of the commitments in Part VIII.
| What the screen shows | What is wrong, and the number to use |
|---|---|
| Revenue by region, 2025: four regions, US ~82% | Leaves out Taiwan (NT$299bn) and others (NT$70bn), and doubles some 2023–24 rows. US share is <b>74.4%</b> (20-F). This caused our September error. |
| Dividend history per ADR | Switches from after-tax to before-tax amounts in September 2025, creating a false +33% jump. Use NT$ per share: <b>NT$4 → 5 → 6 → 7 → 9</b> a quarter, 2024–26. |
| Dividends: the NT$7 (ex 10 Dec) and NT$9 (ex 11 Mar) missing; trailing dividend null | Both declared by the board (11 August and 10 November). The screen can show no dividend and misses the 29% rise. |
| Buffett Desk: '1.8% of free cash flow returned' | Dividends were about <b>57%</b> of free cash flow over 2021–25; the 1.8% looks like Taiwan dollars divided by US dollars. The desk also still uses 2025 data. |
| Q3 free cash flow NT$510bn | The filing gives operating cash flow of NT$919.8bn less capital spending of NT$443.2bn: <b>NT$476.6bn</b>. |
| Annual earnings per ADR, 2024–25 | Mixes Taiwan's accounting standard with international IFRS; the 2024 figure matches neither. Use the company's quarterly US$ per ADR. |
The dividend
Small, rising fast, and now a larger claim on the cash — tested
On 10 November the board raised the quarterly dividend for the third quarter from NT$7 to NT$9 a share, NT$45 per ADR, about $1.37 at today's exchange rate before Taiwan's 21% withholding tax. It goes ex-dividend for ADR holders on 11 March 2027 and is paid on 8 April. The second-quarter dividend of NT$7 goes ex on 10 December and is paid on 7 January. At four times NT$9 the forward yield is about 1.7%. That is not an income stock, but it is a dividend growing faster than most income stocks' earnings.
| Test | TSMC | Result |
|---|---|---|
| 1. Covered by earnings | 29% of earnings in the first nine months of 2026 | ✓ Pass |
| 2. Covered by free cash flow | 40% of free cash flow in nine months; <b>63%</b> of trailing free cash flow at the new NT$9 rate | ◆ Pass, tighter |
| 3. Growing | NT$13 a share for 2023 → NT$22 for 2025 → NT$9 a quarter now | ✓ Pass |
| 4. Never cut | No quarterly cut in the years we checked (2023–26) | ✓ Pass |
| 5. Balance sheet behind it | Net cash covers about 3.2 years of dividends at the new rate | ✓ Pass |
| 6. A stated policy | 'Sustainable and steadily increasing' per share, annual and quarterly; 'continued and increasing' in 2027 | ✓ Pass |
The one test to watch. At NT$9 a quarter the dividend would take about 63% of the last twelve months' free cash flow, before the "significantly higher" capital spending of 2026–28. That is still comfortable for a company with $90bn of net cash and earnings growing 40%. But it means the dividend's growth from here depends on free cash flow keeping up with capital spending. In US dollars the ADR holder also carries the currency: when the Taiwan dollar weakens, as it did this autumn, the dividend in dollars rises less than the dividend in Taiwan dollars.
★ Valuation
14 times next year's earnings for the finest factory on earth
Our method, unchanged. In September we valued TSMC at 20–24 times what analysts expected it to earn in 2027, which gave about $420–520. We keep the same anchor and widen the range, because the market has shown how quickly the multiple of a chip company can move. In the first case, AI spending pauses and 2027 earnings merely equal 2026's, valued at 14 times, about half the multiple at the end of 2025. In the second, the consensus for 2027 arrives and the market pays 18 times, below the 28.8 times of the end of 2025, for the island, the capital intensity and the overseas dilution. In the third, the consensus arrives and the market pays 23 times.
| Case | Earnings per ADR | Multiple | Value |
|---|---|---|---|
| AI pause: 2027 = 2026 | $16.79 | 14× | ~$235 |
| The price | $22.71 (2027) | 14.1× | $319.15 |
| Our level ◆ | $22.71 (2027) | 16.7× | $380 |
| Consensus, 18× | $22.71 (2027) | 18× | ~$409 |
| Consensus, 23× | $22.71 (2027) | 23× | ~$520 |
What is priced in? At $319.15 a buyer earning 10% a year needs free cash flow per ADR to grow about 15% a year for ten years, with growth of 3% after that; or earnings per ADR to grow about 17% a year to 2031 if the market then pays 15 times. TSMC's own plan is for revenue to grow at close to 25% a year from 2024 to 2029, and 2026 is running at 44%. The price therefore assumes that the company falls well short of its own plan. That is possible if AI spending turns, but it is not the likely case on the evidence we have.
Our level. We said in September that $380 was where we would add with both hands, at about 17 times 2027 earnings. The earnings expected for 2027 have risen since, so $380 is now 16.7 times, a slightly better price for the same level. We keep it. The ADR is 16% below it today, and our advice is the same as in the postscript of 17 November: add, and keep the position to a size whose loss you could bear. We do not lower the level to chase the fall, and we do not raise it because the business is better: the level is where the price is good enough, and the price is better than good enough now.
What would change it. Two things would make us stop adding: a sign that TSMC's customers are cutting their orders, which would show first in the monthly sales, and a change in the Strait from exercises to interference with shipping. The first is measurable every month. The second, we hope, never will be.
Risks, probes & controversies
Verified 22 November 2026 against TSMC's 6-Ks, the 20-F and attributed reports
Probes. In April 2025 press reports said TSMC could face a US penalty of $1bn or more over chips that reached Huawei through a third party; we found no resolution and no mention in the 20-F. On 12 November TSMC said some employees' personal data had appeared on an external website; it said the data matched a cloud provider's systems and that it had found no evidence its own systems were compromised.
Customers. The resignation of Nvidia's chief executive and the reported SEC inquiry into Nvidia's accounting concern a customer, not TSMC; we record them because they moved TSMC's price and because a large customer in difficulty is a risk to its suppliers' plans.
| Test | Today | What would change our mind |
|---|---|---|
| Monthly sales, ~10 Dec (November) | October +59.7% | A sharp slowdown in two consecutive months would be the first sign customers are cutting orders. |
| Q4 results and 2027 guidance, mid-January | Q4 guide $52.3–54.3bn | A 2027 capex number far above revenue growth would test the dividend and the returns. |
| Gross margin | 66.4%; guide 65–67% | Overseas dilution beyond the 3–4 points guided would lower the earnings power. |
| The dividend, ex 10 Dec and 11 Mar | NT$7, then NT$9 | Another rise in 2027 as promised; a pause would say capex comes first. |
| Stakes in AI laboratories | Up to $1bn approved | A larger commitment would put TSMC inside the financing loop we measured at Broadcom. |
| The Strait | Exercises, no blockade | Any interference with shipping: stop adding, whatever the price. |
In September I told you TSMC was one of the finest businesses on earth and that I would add with both hands at three hundred and eighty dollars, keeping the position to a size whose loss you could bear. Eight weeks later you can buy it for three hundred and nineteen. A reader who took my advice at three hundred and eighty is sixteen per cent behind, and I owe that reader an account.
The account is short, because the business did everything I hoped. It sold more in the third quarter than it had told us it would. It raised its forecast for the year for the fourth time, to about forty-four per cent growth. October was the best month in its history. The board raised the dividend by twenty-nine per cent, and the directors bought shares while the price fell. If you had shown me those facts in September and asked me where the price would be, I would not have guessed lower.
The price fell for reasons that live outside the company. Chip shares fell by a third. The Taiwan dollar weakened. The premium American buyers pay for the certificate over the shares in Taipei shrank. And the largest buyer of advanced chips had a week of turmoil at the top. None of that changed what TSMC earns. Some of it, the premium and the currency, may reverse; some of it, the market's doubts about AI spending, may prove right. I cannot tell you which, and I do not need to.
I also got four facts wrong, and I have corrected them in this report. The most important was the share of revenue from American customers, which is about three-quarters, not four-fifths. It does not change the verdict, but you should be able to trust the numbers you read here, and a wrong number is a wrong number.
So the advice is the same, at a better price. Keep adding below three hundred and eighty dollars. At fourteen times next year's expected earnings, you are paying for TSMC roughly what the market pays for an ordinary company, and getting the finest factory on earth. But keep the second rule as firmly as the first. The island is no safer at three hundred and nineteen dollars than it was at four hundred and fifty-two. A lower price lets you buy more of the business for the same money. It does not let you put more of your money on one island.
— The Buffett Lens. A great business at a falling price is a gift; the size of the gift you accept is still yours to set.
Sized for the StraitOne of the finest businesses on earth, ahead of its own plan and 29% cheaper than in September. ★ At $319.15 the ADR trades at 14× 2027 consensus, within a range of ~$235 (AI pause) to ~$520; our central value is ~$409. Keep adding below $380, and keep the position to a size whose loss you could bear. November sales on ~10 December; 2027 guidance in mid-January.
Articles that take this apart
MoatsA shortage is not a moatScarcity and pricing power produce identical numbers for a year or two. Then capacity arrives, and only one of them is still there.
MarketWho actually gets paid in the AI build-outEnormous sums are moving. Following where they settle, rather than where they are announced, separates the businesses collecting cash from the ones committing it.
PortfolioThe risk you cannot diversify awaySome risks get smaller when you own more things. A few do not get smaller at all, and for those there is exactly one tool — and it is not diversification.