TSMC

IThe business

TSMC makes chips from designs its customers provide. It fabricates the wafers, packages the most advanced ones with its 3DFabric stacking technology, and tests them.⁵ In 2025 it made 12,682 products with 305 process technologies for 534 customers, and shipped about 15 million 12-inch-equivalent wafers.¹⁴⁵²⁴

In the twelve months to September 2026 depreciation and amortization were 15.1% of revenue, research and development 6.0% and capital spending 33.3%, which is 2.2 times depreciation; operating margin was 58.0% C (calculated).^1415172124 Revenue grew from NT$947.9B in 2016 to NT$3,809B in 2025, 16.7% a year, and was NT$4,945B in the TTM.⁵¹³¹⁴

TSMC defines its market as Foundry 2.0: all logic wafer manufacturing, packaging, testing and mask making. That industry grew 16% in 2025 against TSMC’s 35.9% in US dollars. In January TSMC forecast 14% industry growth for 2026, when it guided close to 30% for itself; in October it guided around 44%.²⁶¹⁶

What it sells

TSMC reports one segment. By the customer’s product, high performance computing, which includes AI accelerators, was NT$2,193B of 2025 revenue, 58% of the total, up 48%; smartphones were 29% and falling.⁵ In 3Q26 high performance computing was 68% of revenue and smartphones 21%.¹⁵ Advanced nodes carry the mix: 3-nanometer chips were 33% of wafer revenue in 3Q26, 5-nanometer 28% and the new 2-nanometer 8%.¹⁵

Who pays

The ten largest customers were 70% of revenue in 2023, 76% in 2024 and 78% in 2025.⁵ The 20-F does not name them. Its segment note gives letters: Customer A was 12% of revenue in 2024 (NT$352.3B) and 19% in 2025 (NT$727.0B), a rise of 106% in a year, and Customer B fell from 22% to 17% as its revenue rose 3% C (calculated).⁵ The largest customer changed in 2025.

By where the customer is headquartered, North America was 75% of 2025 revenue and 79% in 3Q26; China fell from 12% in 2023 to 9% in 2025 and 5% in 3Q26.⁵¹⁵ Substantially all sales are in US dollars, and more than half of capital spending is in other currencies.⁵

Where the plant is

80.0% of TSMC’s noncurrent assets were in Taiwan at the end of 2025, 13.9% in the United States, 3.0% in Japan and 1.7% in China C (calculated).⁵ TSMC and TSMC Arizona together booked 99.3% of consolidated net sales and 97.5% of net income C (calculated).⁵

Cycle or structure

Two things drive the outlook: demand for AI chips and the speed at which TSMC adds capacity. On demand, TSMC says AI accelerator revenue was in the high teens as a percentage of 2025 revenue and forecast it to grow at a mid-to-high-50s percent rate a year from 2024 to 2029. In October the CEO said the rate would probably be higher and gave no new number.²⁶¹⁶ The 2026 revenue guide has risen at each call.^24211714

The CEO also said consumer and price-sensitive end markets face rising component prices, and that TSMC is being prudent in its planning for 2027.¹⁶ Analysts on the call raised four limits on AI spending: data-center power, funding, HBM memory and advanced packaging. The CEO said TSMC sends its own teams to check data-center build-out and power in the United States, and that the feedback is good so far.¹⁶

The last down year was 2023: revenue fell 4.5% in NT dollars and 8.7% in US dollars, operating income 17.8% and gross margin 5.2 points C (calculated).⁵⁴⁰ The 20-F says a significant portion of operating costs is fixed.⁵ I read demand for leading-edge capacity as structural, because customers plan years ahead and TSMC sees their roadmaps. The spending that pays for that demand is the cyclical part. My bear case is an AI pause in which high performance computing, 68% of revenue now against 43% in 2023, falls 25% while the plant is already built.⁵¹⁵

IIThe moat

I think TSMC’s moat is a cost and process advantage: know-how in making the most advanced chips, paid for by scale that no rival matches, and held in place by what it costs a customer to redesign a chip for another factory. It is widening at the leading edge. It has one weak point: ten customers pay 78% of revenue.⁵

The best single measure is the return on the plant, because a foundry’s margin is the return on a large asset base. Over the last ten years TSMC earned between 25.5% and 46.4% a year on its average net plant, 33.1% on average, after a flat 17% tax C (calculated).⁵¹³ In the TTM it earned 59.3% on average plant at the same 17% tax C (calculated).⁵¹⁵²⁴ Each NT$1 of net plant added brought NT$0.55 of after-tax operating profit from 2019 to 2025 and NT$0.48 from 2016 to 2025 C (calculated).⁵¹³ 2019 was the weakest year for margin and 2025 the strongest, which flatters the first figure, and plant is carried at historical cost, which flatters both.

Our pricing will remain strategic, not opportunistic, to earn our value.

Evidence it is widening

Mix and price. Seven-nanometer and smaller chips were 58% of wafer revenue in 2023 and 74% in 2025, and 79% in the third quarter of 2026.⁵¹⁵ Revenue rose 76% from 2023 to 2025 while wafers shipped rose 25%, so revenue per wafer shipped rose 41% to about NT$254 thousand, on total revenue that includes mask, test and packaging C (calculated).⁵²⁴ Gross margin went from 46.0% in 2019 to 59.9% in 2025 and 65.8% in the TTM.^514172124

Share. TrendForce put TSMC at 70.2% of foundry revenue in 2Q25, 70.4% in 4Q25, 72.3% in 1Q26 and 72.5% in 2Q26: a gain of 2.3 points over four readings, none of which defines its base.^45464828 In 2Q26 TSMC’s revenue grew 12.1% on the quarter against 11.5% for the top ten foundries together. TrendForce says TSMC’s 5/4-nanometer and 3-nanometer capacity stayed fully booked and 2-nanometer contributed revenue for the first time.²⁸

Cadence. 2-nanometer entered volume production in 4Q25 with good yield, and TSMC expects the 2-nanometer family’s annual capacity to more than double in 2027.¹⁶ A14 enters risk production in 2027 and volume production in 2028. On the October call the CEO said customer activity on A14 so far is higher than it was on 2-nanometer, and that he expects A14 to be the bigger node.¹⁶

Packaging. Advanced packaging was about 8% of revenue in 2024 and slightly over 10% in 2025, and the CFO expects it in the low teens in 2026, growing faster than the company. Packaging, testing, mask making and others take 10% to 20% of the 2026 capital budget.²⁶ The CEO said TSMC is in production at 5 times the standard reticle size, moving to 9 and 14 times, and that no one else does 5 times; that is a management claim.¹⁶ TSMC is building nearly 20 fabs worldwide, 13 of them leading-edge and advanced-packaging fabs in Taiwan.¹⁶

Evidence it is narrowing

Concentration. The ten largest customers rose from 70% to 78% of revenue in two years, and Customer A’s revenue doubled in 2025.⁵ A customer that large can negotiate hard, and it can also build a second source.

Capacity deposits. Some customers pay temporary receipts to retain capacity. They were NT$291.1B at the end of 2024 and NT$189.9B at the end of 2025, down 35% in a year in which the CEO says demand ran above supply; the 20-F says refund or offset is decided by mutual consent.⁵¹⁶ I cannot tell from the filings whether the drop is prepayments being used up or customers needing to pay less to secure capacity.

Customers going direct. The 20-F says system companies increasingly design their own chips and work directly with foundries.⁵ On the October call an analyst asked about a large customer that manufactures too and might do more of it. The CEO answered that TSMC takes customers’ internal capacity into account and treats all of them fairly.¹⁶

Rivals. Samsung says its foundry has 2-nanometer design wins in high performance computing and plans a second generation of 2-nanometer mobile products in the second half of 2026.³⁸ Intel has 18A-P in risk production.²⁹ TrendForce says Samsung raised prices on 5/4-nanometer and more advanced nodes in 2Q26.²⁸ The 20-F warns that rivals may receive subsidies that TSMC cannot reach on equal terms.⁵

Dilution and cost. TSMC expects the 2-nanometer ramp to cost 3 to 4 points of gross margin in the second half of 2026, and overseas fabs about 2 points in 2026 and 3 to 4 points in the next few years.¹⁹¹⁶ The CFO says tool price inflation is real.¹⁶

Ten customers pay 78% of the revenue, and the moat holds only while none of them finds a second source it trusts at the leading edge.

IIIThe competitive advantage

TSMC’s advantage is scale at the leading edge. In 2Q26 it reported revenue of US$40.2B, 12 times the revenue of Samsung Foundry (US$3.26B, TrendForce’s figure), the next company in TrendForce’s ranking, and 7 times Intel Foundry’s segment revenue (US$5.77B, which includes sales to Intel’s own product groups) C (calculated).¹⁷²⁸²⁹

The margin gap shows the pricing power. TSMC’s gross margin was 67.7% in 2Q26, 35 to 42 points above UMC, GlobalFoundries and SMIC, and its operating margin was 60.3% against 21.8% at UMC, 17.8% at SMIC and 9.7% at GlobalFoundries C (calculated).^17353637 Part of the gap is node mix: the 20-F says competitors in more mature technologies are numerous and offer specialized processes, and UMC earned 37% of its revenue at 22/28-nanometer.⁵³⁶ The relevant comparison is at the leading edge. Intel reports a foundry segment loss (US$2.1B on US$5.8B of revenue, 36% of it) C (calculated); Samsung does not report foundry revenue, so its figure is TrendForce’s. Neither reports a foundry gross margin.²⁹²⁸

TSMC spent US$15.7B on plant in 2Q26, 8.6 times SMIC’s US$1.84B, the largest of the other foundries I can measure, and 38 times GlobalFoundries’ US$0.41B. It has budgeted US$60 to 64B for 2026, 31 times the US$2B UMC plans C (calculated).^1837353619

Where TSMC wins

TSMC wins where a chip needs the newest node, advanced packaging and volume together. High performance computing was 68% of 3Q26 revenue, and the CEO’s claim is that all the AI players work with TSMC.¹⁵¹⁶ Capacity discipline helps: the CEO said TSMC does not allow customers to sublet or auction off allocated capacity.¹⁶

Where it loses

At mature nodes competitors are numerous, and the 20-F says rivals may also pursue customers who want to diversify their supply chains.⁵ Samsung says it has 2-nanometer design wins in high performance computing; it names no customers or volumes.³⁸ An analyst noted on the October call that rivals plan to adopt High-NA lithography earlier than TSMC, which plans to use it from 2030; the CEO said TSMC adopts a tool when it is mature and cost-reasonable.¹⁶

No rival publishes a leading-edge yield or a foundry margin. The gaps I can measure are 12 times Samsung Foundry’s revenue and 8.6 times SMIC’s plant spending in 2Q26.^17281837

IVManagement

C.C. Wei is chairman and chief executive. He has run TSMC since November 2013, as co-CEO until June 2018 and as sole CEO since; he became chairman in June 2024. He joined TSMC in 1998 and holds a Ph.D. in electrical engineering from Yale; he is in the 71 to 75 age band.⁵ He owns 7.45 million shares, 0.03%, worth about NT$17.8B or US$546M at the Taiwan price C (calculated).⁵⁵⁰³ Wendell Huang has been chief financial officer since September 2019 and joined in 1999.⁵

Pay

Wei’s pay for 2025 was NT$2,422.7M (US$77.2M), 0.14% of net income to the parent; he owns 0.03% of the shares C (calculated). Salary was 0.7% of it, bonus and profit sharing 37%, stock awards 15% and other pay, which includes a long-term incentive bonus, 47% C (calculated).⁵ All executive officers together were paid NT$9,060.0M, 0.53% of net income C (calculated).⁵

The 20-F does not state what the CEO’s bonus or long-term incentive measures. The articles of incorporation tie a pool of at least 1% of annual profits to all employees. For 2025 profits the 20-F lists two pools of NT$103.07B each, a business performance bonus and a cash profit sharing bonus; the accrual note covers the profit sharing line, 5.3% of operating income C (calculated). Whether the business performance bonus is a second cost the filing does not say; if it is, the two together are NT$206.1B, 10.6% of operating income C (calculated). Total employee benefits expense was NT$396.1B, 10.4% of revenue.⁵ TSMC adopted a clawback policy in August 2023 and recovered nothing in 2025.⁵

Directors and executive officers together own 0.23% of the shares. The National Development Fund, an R.O.C. government investment fund, owns 6.38% and holds a board seat. Incentives therefore run through pay and profit sharing, with no founder stake behind them.⁵

Wei’s salary was 0.7% of his 2025 pay; the rest was bonus, profit sharing, stock and a long-term incentive.⁵ The 20-F does not say what his bonus measures; the employee pools come from profit, so I read it as paying for profit, not return on plant. His stake of NT$17.8B is 7.3 times that year’s pay C (calculated).⁵⁵⁰³ Over a cycle they kept building through the 2023 revenue fall, paid out 66% of free cash flow and bought back almost nothing.⁵¹⁵¹³ Building has paid so far. I can’t see that it does at 35% of revenue.

The record against its own guidance

Revenue finished above the top of the guided range in 4Q25, 1Q26 and 3Q26, and at the top in 2Q26. Gross margin finished above the range in the first three of those quarters and inside it in 3Q26.^1417212439 The 2026 growth guide rose from close to 30% to around 44% in nine months. When the demand is weaker, management says so: it is prudent about consumer markets in 2027.¹⁶

The bench and the board

The board has ten directors, seven independent, among them Sir Peter Bonfield, Michael Splinter, Moshe Gavrielov, L. Rafael Reif, Ursula Burns, Lynn Elsenhans and Chuan Lin. All terms end in 2027.⁵ The co-chief operating officers are Y.P. Chyn, who joined in 1987, and Y.J. Mii, who joined in 1994. Four senior vice presidents were promoted on 10 February 2026, and Ray Chuang became chief executive of TSMC Arizona on 1 October 2025.⁵ The Compensation and People Development Committee has reviewed the succession pipeline since February 2023, and the 20-F names no successor to Wei.⁵

One flag: in October 2024 TSMC told U.S. and Taiwan authorities that a chip it made for one customer might have been diverted to a restricted entity. It has cooperated with information requests since, and no outcome is reported.⁵

Capital allocation

Management has paid out two-thirds of free cash flow and kept the rest, which is why NT$2,985B of net cash reserves sit on the balance sheet. The decision that mattered was the size of the build in 2021 to 2023, when plant spending ran at 44% to 53% of revenue. In 2023 revenue fell 4.5% and TSMC still spent NT$949.8B on plant, 44% of revenue. By 2025 revenue was 76% higher C (calculated).⁵¹⁵¹³

Capital spending was NT$7,042B from 2016 to 2025, 39% of revenue, while revenue grew 4.0 times C (calculated).⁵¹³ Dividends paid were NT$2,736B, 66% of cumulative free cash flow of NT$4,161B C (calculated).⁵¹³ The cash dividend a share was NT$6 in 2016, NT$18 in 2025 and NT$24 in 2026, and management describes a policy of a sustainable and steadily increasing dividend.¹⁹¹⁶ Buybacks are close to nil: one NT$3.1B program in 2024.⁵ Net cash reserves of NT$2,985B are 4.8% of market value C (calculated).¹⁵⁵⁰¹⁴

The 2026 budget of US$60 to 64B is about 1.5 times 2025, and management expects 2026 to 2028 to be significantly higher than the US$101B spent in the last three years.¹⁹¹⁶²⁶ The CFO says that if TSMC does its job right revenue will outpace capital spending, and that tool price inflation is real.¹⁶

We have no intention to leave food on the table for others.

In 2026 TSMC will spend about 35% of revenue on plant and still grow revenue about 44%, and the price asks how long both can be true.

VFinancials

Free cash flow is what is left after the plant, and the plant is a third of revenue.

Gross margin went from 46.0% in 2019 to 59.9% in 2025 and 65.8% in the TTM. Operating margin went from 34.8% to 50.8% and 58.0%.^514172124 Operating cash flow was 63.3% of TTM revenue and free cash flow 29.9% C (calculated).^14172124

Depreciation is the line to watch. Depreciation and amortization were NT$220.2B in 3Q26, up 35% on the year and 14.7% of revenue, and capital spending is running at 2.2 times depreciation, so the charge keeps rising C (calculated).¹⁵ The margin before depreciation, operating income plus depreciation and amortization over revenue, was 73.1% in the TTM. In every year from 2016 to 2025 it was between 61.6% and 68.9%, and 68.0% on average from 2020 to 2025 C (calculated).^51314172124 That one fact carries most of my valuation: today’s margin is above anything TSMC reported in the last decade.

Earnings quality

TTM earnings per share of NT$98.26 include about NT$2.24 from a gain on the sale of VIS shares in 2Q26, TSMC’s own figure; without it they are about NT$96.02 C (calculated).^1914172124 Net interest income was NT$31.7B of the NT$34.6B of non-operating items in 3Q26.¹⁵ The 20-F books the surtax on undistributed earnings in the year the earnings arise, and the quarterly reports book it a year later. FY2025 net income to the parent was NT$1,697.6B on the 20-F basis and NT$1,717.9B on the quarterly basis; the annual columns below follow the 20-F and the TTM follows the quarterlies.⁵²⁷ The effective tax rate was 17.0% in 2025 on the 20-F basis and TSMC guides 17% to 18% for 2026.⁵²⁶

Government money

Free cash flow is operating cash flow less plant spending. Government grants, U.S. investment tax credits and interest received are reported in investing cash flow and are outside it. They were NT$284.9B in the TTM, 17% of TTM capital spending, and TSMC says the NT$198.7B in 3Q26 was mainly U.S. Treasury investment tax credits.^15182225 The 20-F separately shows plant grants of NT$75.2B in 2024 and NT$76.3B in 2025, 6.0% of 2025 capital spending, before any tax credits.⁵ Interest received was NT$52.5B in the first half of 2026; twice that, NT$105B, would leave about NT$180B of credits, grants and other items, 11% of TTM capital spending. The split is not disclosed, so that is my estimate E (estimate).²⁰ I leave all of it out of the cases, and the bridge shows what it is worth.

The balance sheet

At 30 September 2026 TSMC held NT$4,017B of cash and marketable securities against NT$1,032B of interest-bearing debt: net cash reserves of NT$2,985B, with NT$225B of long-term investments on top.¹⁵¹⁴ Of the cash, NT$363B was declared and payable as dividends, NT$7 a share for each of 1Q26 and 2Q26. The 1Q26 dividend went ex on 16 September and was paid on 8 October, so the Q4 2026 price no longer carries it; I deduct NT$181.5B. The 2Q26 dividend goes ex on 10 December, so the price still carries it.^41424314 TSMC still issues bonds, NT$86.9B in 2025.⁵

VIValuation

First, how this goes badly wrong

Working back from losses, three ways stand out. An AI pause with the plant built: high performance computing is 68% of revenue, so a 25% fall there takes my bear value to NT$887.¹⁵⁵⁴⁰ Plant that costs more per dollar of revenue: at US$1.05, my base value is NT$1,842 E (estimate).¹⁵ And Taiwan, which my cases leave out: a 44% chance of losing the business takes the base to NT$1,180.³³

What an owner could really take out

Over the last twelve months TSMC earned NT$2,547.9B. After NT$1,646.9B of plant, free cash flow was NT$1,480.7B, 58% of that, or about NT$57 a share against earnings of NT$98.26 C (calculated).^14172124 Stock pay, NT$1.2B in 2025, hardly moves it.⁵ Some plant buys growth. Spending only the NT$746.3B of depreciation would leave about NT$92 a share, 93% of earnings E (estimate).^14172124 Spending is 2.2 times depreciation, so I read it nearer NT$57, and my cases take cash after all the plant.

What I can’t judge

I can’t say how much plant is needed to stay in place, how long AI demand lasts, or what the Taiwan risk is worth. So I show a range, NT$887 to NT$3,056, not one value.

What the price implies

On the Taiwan price TSMC trades at 24.1 times trailing earnings, 21.4 times my estimate of 2026 earnings and 17.4 times my estimate of the next twelve months; the ADR trades at 28.6, 25.4 and 20.6 times E (estimate).^5014172124 The median at the year-ends of 2016 to 2025 was 20.9 times, in a range of 11.7 to 26.9, so today’s 24.1 times is 15% above that median C (calculated).⁴⁴⁵¹³

The Taiwan price is today’s margin, held, and more. The TTM margin before depreciation of 73.1%, held for fifteen years on my revenue and plant, gives NT$2,218 a share; the price needs 76.4%, against a best year of 68.9% in the last decade E (estimate). Revenue growth of 12.0% a year in years 2 to 10, against my 8.7%, also gets there E (estimate). Solved for one growth rate of free cash flow, the price needs 13.6% a year for fifteen years from the TTM’s NT$1,480.7B, to about NT$10.0 trillion in year fifteen C (calculated), between my base case (NT$6.3 trillion) and my bull case (NT$10.4 trillion) E (estimate).^14172124

On my revenue and plant, the Taiwan price needs a margin before depreciation of 76% held for fifteen years, or revenue growth about 3 points a year faster than my base case; the ADR price needs more revenue than my bull case.

The ADR premium

One ADR is five ordinary shares. At the NT$32.57 to the dollar I hold, the ADR’s close of US$431.81 on 29 October equals NT$2,812.81 an ordinary share, 18.7% above the Taiwan close of NT$2,370 on 30 October C (calculated).⁴⁹⁵⁰³⁴ The New York session of 30 October had not closed when I pulled the data, so the ADR line is the older of the two; against the Taiwan close of the same date, 29 October (NT$2,415.00), the premium was 16.5% C (calculated).^4950333 Since my earlier pricing the Taiwan close is 0.6% lower than the NT$2,385.00 of 22 October and the ADR 3.2% higher than the US$418.51 of 23 October C (calculated).¹² Taiwan did not trade on 23 October, a day the exchange’s schedule does not list as a holiday.³⁴ Over the past year it had a median of 17.2% and a range of 7.0% to 29.7% C (calculated). The Taiwan stock is 72% above its November 2025 low close of NT$1,375 and 8% below its Q4 2026 high of NT$2,585 C (calculated).³³⁵⁰

The 20-F gives no reason for the premium. It does cap new depositary receipts at the number the regulator approved plus those already cancelled, and it requires a holder who withdraws shares to register with the exchange and appoint a local agent, a custodian and a tax guarantor.⁵ My inference is that these rules limit arbitrage; I cannot confirm that they explain the premium. The Taiwan price is 18% above my base value and the ADR price 40% above it; I do not model the premium closing.

The discount rate

I use 10.5%, the midpoint of the capital asset pricing model’s range: 9.9% with a 10-year beta of 1.17 and 11.0% with a 5-year beta of 1.43, on a 5.00% Treasury yield and a 4.20% equity risk premium C (calculated).³⁰³¹³² Substantially all of TSMC’s sales are in US dollars, so I use a US-dollar rate, keep the exchange rate at NT$32.57 and put no country premium in it.⁵³ At NT$2,370 my base cash flows return 9.4% a year, below the 9.9% to 11.0% that the model gives, so the price includes more margin or growth than my base case; at the ADR price they return 8.4%, 1.5 points under the lower rate. At 11.0% my base value is NT$1,875, 21% under the Taiwan price E (estimate).

Three cases

I build each case from revenue down to free cash flow: a margin before depreciation, depreciation at 19% of average net plant rising to 23% (it was 18% to 27% in 2016 to 2025), an 18% tax rate (the top of TSMC’s 2026 guide), net new plant equal to the next year’s revenue change over the revenue a dollar of plant supports, and working capital at 2% of each year’s revenue change C (calculated).⁵¹³²⁶ Applied to the TTM at a 17% tax, the method gives NT$1,479B of free cash flow, against TSMC’s reported NT$1,480.7B C (calculated).^14172124

Base case. Year one, to September 2027, is US$223B: 4Q26 at the guidance midpoint of US$53.3B plus three quarters that average US$56.6B, 42% on the TTM E (estimate).¹⁴ Revenue then grows 19%, 14%, 11% and 9% in years 2 to 5, and 7% falling to 4% after that. Calendar-2029 revenue is about US$311B, 13% above the company’s January path of about US$275B.²⁶²⁴ The margin before depreciation falls from the 71.5% the 4Q26 guide implies to 68.0% in year six, the 2020 to 2025 average C (calculated).¹⁴⁵ Capital spending ends at 25% of revenue, below every year from 2016 to 2025 (the low was 30.6% in 2018), because each dollar of plant carries US$1.06 of revenue C (calculated).⁵ Return on net plant, after an 18% tax and on average plant, goes from 58.6% in the TTM to 65% in year one and 42% in year ten, against 33.1% on average over ten years E (estimate). Terminal reinvestment is 8.6% of after-tax operating profit for 3% growth, a return on new plant of 35%, above the ten-year average. That is the generous end, and I cannot defend it for plant that is 25% of revenue beyond the record, so the downside table prices returns on new plant of 20%, 15% and 10.5%.

I did not shrink the fade. Management names 6 to 8 points of dilution; my gross margin falls 10.5 points, and the 2.5 to 4.5 points beyond the named dilution are my assumption that the margin before depreciation returns to its 2020 to 2025 average. Management’s own floor of 56% is a fall of 9.8 points, so I am 0.6 point under it. The first table under “What closes the gap” prices a year-ten gross margin of 56%, 60% and 62%.²⁶¹⁶¹⁹

Bear case. An AI pause with the plant already built. High performance computing, 68% of 3Q26 revenue against 43% in 2023, falls 25% in year two, which takes 17.0 points off company revenue; the rest falls 8.7%, as the whole company did in 2023 in US dollars, which takes off 2.8 more. Revenue falls 19.8% to US$164B, 23% below the 4Q26 annualized run rate of US$213B.¹⁵⁵⁴⁰ The margin before depreciation falls to 64% in year two and to its 2019 level of 62% from year three, because a significant portion of costs is fixed; operating margin in year two is 41.7%, against 42.6% in 2023 on a revenue fall of 8.7%. Gross margin is about 46% in year ten and the return on net plant settles near 27%, the bottom of its ten-year range E (estimate).⁵²⁶

Bull case. Demand runs ahead of supply for years. Revenue is US$641B in year ten, 15% a year from the TTM; the margin before depreciation stays near today’s, at 71.5%; each dollar of new plant supports US$1.00 to US$1.20 of new revenue, where I use US$0.85 to US$0.92 in the base case E (estimate).¹⁴

Terminal growth differs by case (2.5%, 3.0%, 3.5%); at 3.0% in all three the bear is NT$902 and the bull NT$2,959, so about NT$112 of the spread is that choice. The 25%, 50% and 25% weights are a convention, not probabilities; they give NT$1,989, which I do not use for the margin of safety E (estimate).⁴⁹⁵⁰

I add NT$2,953B of net cash reserves and investments, NT$114 a share, at carrying value, after the dividend deduction.^14152041 The terminal value is 40% of the value of the cash flows and 37% of the base value including cash E (estimate).

Does the revenue add up

The base case needs revenue to grow 26.2% a year from US$122.4B in 2025 to US$311B in calendar 2029. At the company’s forecast rate AI accelerators reach US$127B, 41% of base revenue, and everything else must grow 16.3% a year; on the company’s own path of US$275B everything else grows 10.2% a year E (estimate).²⁶ Year-ten revenue of US$474B at 2025’s revenue per wafer needs 58 million wafers, 3.4 times today’s capacity of more than 17 million; at revenue per wafer rising 5% a year it needs 36 million, 2.1 times E (estimate).⁵²⁴¹⁶

What closes the gap

Of the single changes in the bridge, only the bull revenue path takes the base value to the Taiwan price. The TTM margin held gives NT$2,218, 2 points more growth in years 2 to 10 gives NT$2,201 and plant as productive as in my bull case gives NT$2,211; any two of the three reach the price E (estimate).⁵⁰ Government money is the largest item I leave out. At 11% of capital spending for ten years it adds NT$79 a share; kept through the terminal value it adds NT$164 and leaves the base value 8% under the price. The split is not disclosed, so both are upper bounds E (estimate).^15182225

The downside changes are larger: growth 2 points lower gives NT$1,841, a return on new plant of 15% from year ten gives NT$1,724, Customer A’s 19% of revenue moving to a rival in year two gives NT$1,558, and AI accelerators growing at half the company’s rate gives NT$1,500 E (estimate).⁵²⁶

What the cases leave out

The cases hold no cross-strait conflict, no blockade and no tariff beyond today’s. The 20-F says financial markets have viewed past developments between the R.O.C. and the P.R.C. as occasions to depress share prices, and it does not quantify a conflict.⁵ Neither do I. Most executives and the principal production sit in the R.O.C., and 80.0% of noncurrent assets are in Taiwan.⁵ The table is arithmetic only: the cash stays and the operating value is multiplied by one minus a chance.

Trade rules also move. The 20-F lists a Section 232 proclamation of a 25% tariff on certain advanced computing chips unless imported for designated U.S. use. It says the Supreme Court struck down the IEEPA reciprocal tariffs in February 2026, a Section 122 replacement followed, Section 301 investigations that include Taiwan opened in March, and the effect on the January agreement capping reciprocal tariffs at 15% is unclear. It gives no dollar effect.⁵

Other uses of the same dollar

The Treasury yield I use is 5.00%.³⁰³¹³² At NT$2,370 my base case returns 7.8% a year over ten years, 2.8 points more, with a bear value 63% under the price C (calculated).⁵¹⁴ The closest comparison on this site is NVIDIA, which sells into the same AI demand: 9.3% a year on its base case, same ten-year basis.

Two different questions

How good is the business, and how good is the price? TSMC took 72.5% of foundry revenue in the second quarter and earned 59% on net plant. The price is 18% above my base value. This looks like a great business at a price that already pays for much of the good outcome. The price needs two of three things: today’s margin held, growth 2 points faster, or more revenue per plant dollar. At one chance in three each, two or more come true about 26% of the time E (estimate).²⁶

The margin of safety

• PriceNT$2,370.00Close, Q4 2026 [50]
• Base valueNT$2,006My estimate E (estimate)
• Margin of safety belowNT$1,180set from my bear case: the price at which the bear value is 75% of the price paid C (calculated)

I set the margin of safety from the bear case instead of as a discount to the base value, because one cycle and one geography drive this business. At NT$1,180 the bear case is worth NT$887 discounted at 10.5%, 25% under the price, and the bear cash flows still return 8.4% a year over fifteen years; the base case returns 15.5% a year E (estimate). That return does not cover Taiwan: the base value falls to NT$1,180 if the chance of losing the operating business is 44%. NT$1,180 is 14% below the NT$1,375 close of November 2025, the lowest of the past year, so the stock has not traded at that price in a year.³³ At that price TSMC would trade at 12.0 times trailing earnings and 8.6 times my next-twelve-month estimate, and the Taiwan line equals US$181 an ADR at parity E (estimate).¹⁴³ The price sits 101% above the margin-of-safety price.

Where this could be in 5 and 10 years

For the base to be right at five years, revenue has to reach US$366B by September 2031, up from US$157B now. The margin before depreciation stays near 69% and capital spending near 26% of revenue. For ten years it needs US$474B, a margin of 68%, and each new dollar of plant still supporting about 90 cents of new revenue E (estimate). At year 10, 70% of the base value is the terminal value, so that figure leans hardest on my 3% growth after year 15 and my 35% return on new plant. The base returns 7.8% a year, below the 10.5% I discount at, because the price starts above my value.

The cases split on revenue and margin. The discount rate is 10.5% in all three. The bear has revenue falling 19.8% in year two and a margin before depreciation of 62%, so year-10 revenue is US$234B. The bull has US$641B at a margin near 72%. That is 2.7 times the revenue and 4.3 times the value at year 10 C (calculated).

The most likely way the base is wrong is plant. TSMC says 2-nanometer capacity costs substantially more to build than 3-nanometer. If a dollar of revenue needed US$1.05 of plant by year ten, the base value today would be NT$1,842 instead of NT$2,006 E (estimate).

A reader's guide: these are estimates of value, not forecasts of the share price, and the market price can sit above or below value for years.

VIIThe case for the price

What most people miss

Most of TSMC’s growth now comes from each wafer earning more. In 3Q26 dollar revenue rose 41.2% on the year and wafers shipped rose 9.6%.¹⁵ Revenue per wafer went from about US$8,100 to about US$10,440, up 29% C (calculated). That counts packaging and test. Much of the rise looks like mix: 2- and 3-nanometer chips went from 23% of wafer revenue to 41%.¹⁵ Capacity has usually grown in the mid to high single digits a year.¹⁶

TSMC does not normally add capacity to a node that has reached its target. For 3-nanometer it planned three new fabs, added a fourth in Tainan on the October call, and is converting 5-nanometer tools.¹⁹¹⁶ My reading: converted tools stretch a dollar of new plant.

The bull case

On 15 October C.C. Wei said the gap between demand and supply is still quite big, and that AI accelerator growth will probably beat the guided mid-to-high-50s rate. He also said there is “no ceiling” on profitability, though no jump overnight.¹⁶

I have Gavin Baker of Atreides secondhand, from a 24/7 Wall St. report on his May podcast. It quotes him saying the wafer shortage, which TSMC controls, will prevent a bubble. If TSMC did what Jensen Huang wanted, he said, Nvidia’s GPU sales could reach US$2 trillion in 2026 or 2027. He also said companies are starting to go to Intel and Samsung.⁵⁶ His fund’s 13F for 30 June lists no TSMC position.⁵⁴

I haven’t read the broker notes either. Taiwan press reported Citi at NT$4,000 on 8 October, above my bull value, expecting 2027 revenue growth above 40%.⁶⁰⁶¹ In July it saw wafer prices rising through 2027.⁵⁷ After the results, MarketBeat’s tracker shows Needham at US$560 and Barclays at US$670 for the ADR. My bull value is about US$469 an ADR.⁶²

Brad Gerstner’s Altimeter reported 1,429,355 ADRs at 30 June, up 4.7% in the quarter and 6.9% of its 13F total.⁵²⁵³ ARK’s ARKK fund had 167,325 ADRs on 30 October, 1.00% of the fund.⁵⁵ I found no public case on TSMC from Gerstner or Cathie Wood, only these positions.

Against reality

This is about the Taiwan price of NT$2,370. The ADR at US$431.81 needs US$800B of revenue in year ten, more than my bull case E (estimate).⁴⁹

My model offers three levers. Keep today’s 73.1% margin before depreciation for fifteen years. Grow revenue 2 points a year faster than my base. Or earn US$1.00 to US$1.20 of new revenue on each new dollar of plant, against my US$0.85 to US$0.92. Any two reach NT$2,424 to NT$2,477 E (estimate). Only my full bull revenue path does it alone. So if TSMC pulls two of those levers and beats its targets, the price is possible. Today’s numbers don’t show that yet: the price is 18% above my base value and no lever is proven.

The company’s January path is about US$275B of revenue in 2029. My base is US$311B, and 2 points of extra growth makes it about US$323B, 18% above that path E (estimate).²⁶ TSMC has not updated the path, though 2026 growth has gone from about 30% to about 44%.²⁶¹⁴

Revenue is close: the 4Q26 guide annualizes to US$213B, 4% under the US$223B my base needs for the year to September 2027 C (calculated).¹⁴ Margin is harder. The last twelve months sit at 73.1%, but the 4Q26 guide implies 71.5%, and management expects overseas fabs to cost 3 to 4 points of gross margin.¹⁴¹⁶ Plant is the least proven. Spending is 33% of revenue, and TSMC says 2-nanometer capacity costs substantially more to build than 3-nanometer.¹⁴²⁶ The CFO’s answer is that revenue will outpace spending if TSMC does its job.¹⁶

Where I come down

Revenue per wafer moved me most. It shows TSMC can outgrow its wafer count. Today’s margin lasting fifteen years did not move me, because management itself names the dilution ahead. Gross margin at 65% or more through the 2-nanometer ramp, and a 2027 budget growing more slowly than revenue, would raise my values.

VIIIRisks, and what would change my view

1. AI demand and the build-out cycle. HPC was 68% of 3Q26 revenue, and the CEO said demand is above supply by a wide gap.¹⁵¹⁶ If AI customers pause, TSMC’s fixed costs and a plant budget of US$60 to 64B meet falling revenue at once.⁵¹⁹
2. Capital intensity and depreciation. Capital spending is 33% of revenue and 2.2 times depreciation. Depreciation rose 35% in 3Q26.¹⁵ A 2027 plant budget that grows faster than revenue would lower my base value.
3. Margin dilution. The 2-nanometer ramp costs 3 to 4 points of gross margin in the second half of 2026 and overseas fabs 2 points, rising to 3 to 4 points.¹⁹¹⁶ The 4Q26 gross margin guide of 65% to 67% is down 40 basis points at the midpoint.¹⁴
4. Customer concentration. The ten largest customers are 78% of revenue. Customer A is 19% and grew 106% in 2025.⁵ A change at one customer moves the total.
5. Taiwan. 80.0% of noncurrent assets, most executives and the principal production are in the R.O.C.⁵ The cases do not price a conflict.
6. Trade and export controls. The Nanjing fab’s validated end-user status expired in December 2025 and was replaced by an annual license. China was 9% of 2025 revenue and 5% of 3Q26 revenue. The Section 301 investigations opened in March 2026 include Taiwan.⁵¹⁵
7. Exchange rate. Substantially all sales are in US dollars and over half of capital spending is in other currencies. Each 1% fall in the dollar costs 0.3 point of operating margin.⁵
8. Natural hazards and utilities. Earthquakes in April 2024 and January 2025 cost about NT$3B and NT$5.3B net of insurance. The 20-F lists power outages, higher electricity prices and drought-related water limits as risks.⁵
9. Succession. The CEO is in the 71 to 75 age band, the committee reviews the pipeline, and no successor is named.⁵
10. The ADR premium. A holder at US$431.81 pays 18.7% more than the Taiwan line for the same share, and the premium has ranged from 7.0% to 29.7% in a year.³³⁴⁹⁵⁰

What would raise my base value: gross margin holding at 65% or more through the 2-nanometer ramp, a 2027 plant budget that grows more slowly than revenue, and a disclosed split showing credits and grants above 15% of capital spending or credits that continue after 2026 construction starts. What would lower it: gross margin below 60% for two quarters, plant spending above 40% of revenue while growth falls under 25%, the ten largest customers above 82% of revenue, customer deposits falling further while demand is above supply, or a large customer moving a volume 2-nanometer product to a rival.^141551926