TSMC: A Deep-Dive Value Investing Study

What is the cost structure? Is this business simple and transparent? Fact: In 2025 TSMC's gross margin was 59.9%, operating margin 50.8%, and net margin 45.1%; in Q1 2026 these rose further to 66.2%/58.1%/50.5%. At the same time, TSMC is extremely capital intensive: capital expenditure in 2023, 2024, and 2025 was NT$949.8bn / 956.0bn / 1,272.4bn, and the company expects 2026 capex to reach US$52bn-56bn. So this business is not actually "light": it is a manufacturing platform with high fixed costs, high technological barriers, and heavy capital investment, but one that can also enjoy enormous operating leverage when utilization is high and leading-edge nodes are scarce. The business model itself is clear, but financial judgment cannot rest on profit alone; it must look at cash flow and return on reinvestment.

Does it depend on a few customers, suppliers, policies, or key people? Fact: Yes, and the dependence is not low. The company relies on orders from leading customers and on a handful of high-end equipment suppliers, and it acknowledges that high-quality silicon wafers "come mostly from a limited number of suppliers." Meanwhile, its global expansion increasingly depends on government subsidies and industrial policy, such as the U.S. CHIPS Act and German subsidies. On governance, since 2024 C.C. Wei has served as both Chairman and CEO. This is not necessarily a bad thing, but it means investors must rely more heavily on board oversight and cultural checks and balances.

If the stock market closed for five years, would I be willing to hold? Opinion: If the purchase price were closer to intrinsic value, yes; at the current price, I would be willing to hold this company, but not necessarily willing to bear the opportunity cost that comes from stacking "high quality + high valuation + geopolitical tail risk."

Business comprehensibility score: 4.5/5.

Industry and Competitive Landscape Fact: TSMC does not sit in the generic "semiconductor industry" concept but in the more specific space of advanced-logic foundry and advanced packaging. By TSMC's own disclosure, the Foundry 2.0 industry grew 16% in 2025, while TSMC's USD revenue grew 35.9%, clearly outpacing the industry. Third-party data also show TSMC's share of the pure-play wafer-foundry market is roughly in the 67.6%–70.4% range, versus about 7.1% for Samsung Foundry, an enormous lead. Its strongest rival remains Samsung Foundry; a longer-term institutional challenge comes from Intel Foundry; in mature processes, UMC, GlobalFoundries, and SMIC each hold local pockets of the market. Long-term industry demand is broadly upward but not smooth: AI, HPC, EVs, and edge computing drive structural growth, while phone/PC/industrial orders are more cyclical.

Inference: This is "a good company in a good industry," but not "a good industry without cycles." The industry is extremely hard for new entrants to disrupt, yet it can be unsettled by technological paradigm shifts, national subsidy competition, export controls, customers building their own capacity, or a pullback in AI investment. For TSMC, the real competition is not "whether others can do foundry work" but "whether others can simultaneously approach it on leading-edge nodes, advanced packaging, yield, scale, customer trust, and delivery certainty." Seen from today, that bar remains astonishingly high.

Industry attractiveness score: 4/5.

Moat and Management

What exactly is the moat? TSMC's moat is not a single source but several advantages reinforcing one another:

Brand and trust. Fact: The pure-play foundry model means TSMC does not compete with customers for chip-design profit; instead it becomes the neutral platform "every IC innovator is willing to work with." The company summarizes its core competitiveness as Technology Leadership, Manufacturing Excellence, Customer Trust. This is not an empty slogan: when customers hand their most critical CPUs, GPUs, AI accelerators, and phone SoCs to TSMC, that itself is a vote of industry trust.

Scale advantage. Fact: TSMC's 2025 capacity exceeded 17 million 12-inch equivalent wafers, with a particularly strong lead in leading-edge nodes and advanced packaging; 2025 revenue was NT$3.81tn, and the 2026 capex plan is US$52bn-56bn. This capital intensity, talent density, supply-chain bargaining power, and learning curve make it very hard for latecomers to replicate within 3–5 years, even with money.

Cost advantage and learning curve. Inference: TSMC may not have the lowest nominal wages or static depreciation costs, especially after overseas expansion; but on yield-ramp speed, unit cost of good die, process co-optimization, equipment utilization, and capacity-scheduling efficiency, it holds a deeper "systemic cost advantage." That is also why it can sustain high gross margins in the high-end market while customers still keep placing orders. Its gross margin rose from 54.4% in 2023 to 59.9% in 2025, and to 66.2% in Q1 2026; this is not mere accounting cosmetics but more the combined result of a high-end product mix, advanced-packaging scarcity, technology leadership, and scale efficiency. It contains both structural advantage and a cyclical-upswing component.

Switching costs and ecosystem position. Inference: The real lock-in TSMC has over customers is not just "the hassle of re-taping out at another fab," but the deep switching cost formed jointly by process libraries, the EDA/IP ecosystem, packaging coordination, design-technology co-optimization, mass-production yield databases, and delivery and confidentiality systems. Once a leading AI chip tapes out successfully at TSMC, migrating it to another leading-edge foundry carries yield risk, time cost, and intellectual-property risk.

Patent, technology, and regulatory barriers. Leading-edge processes, advanced packaging, and high-end equipment are natural barriers in themselves; export controls have, to some degree, sharpened the tiered scarcity of "compliance, equipment, and capacity." But this barrier does not benefit only TSMC; it may also constrain its China business and its pace of acquiring equipment globally.

Network effects and data advantage. Strictly speaking, TSMC does not have a standard network effect the way a social platform does; but it does have a weak industry-side network effect: the more top-tier customers achieve volume production on the same platform, the more IP, design tools, materials, packaging, and equipment resources optimize around that platform, which in turn attracts more customers. This is an "ecosystem-type moat" rarely seen in manufacturing.

Is the moat widening, stable, or narrowing? Opinion: It is currently stable to widening. In 2025 the share of revenue from leading-edge nodes rose further to 74%, 2nm entered volume production in 2025, and A16/N2P are advancing on schedule; meanwhile the importance of advanced packaging is rising, making TSMC's "process + packaging" advantage more complete than before. What could truly narrow the moat is not ordinary competition but long-term erosion from overseas manufacturing costs, key customers diversifying their capacity, or Samsung/Intel meaningfully catching up in advanced logic or advanced packaging.

Is management trustworthy, and is capital allocation rational? Fact: TSMC's governance record is sound overall. The 2025 Form 20-F shows management assessed the effectiveness of internal controls, and Deloitte issued an unqualified internal-control audit opinion. The company has long relied on cash dividends rather than routine share buybacks, and the 20-F explicitly states "purchases of the issuer's equity securities by the issuer and affiliated purchasers: not applicable," indicating it has not used buybacks to "dress up" earnings per share. The primary uses of cash over the past several years have been very clear: reinvestment in expansion + raising the dividend + maintaining a solid balance sheet, with no large, high-premium acquisitions that destroy value. Its recent sale of its VIS stake to focus on the core business is also consistent with "concentrating resources on the core."

Fact: But the "equity alignment" between management and shareholders is not strong. As of late February 2026, Chairman and CEO C.C. Wei personally held about 0.03%; the National Development Fund held about 6.38%. This means TSMC is not a "founder with very high ownership" type of company; it is more a world-class enterprise driven by systems, culture, and operating processes.

Opinion: On the three dimensions of "honest, rational, long-term oriented," TSMC's management is trustworthy overall; on "ownership alignment," it falls short of the founder-controlled companies Buffett most favors; on "capital allocation," it is clearly better than most tech manufacturers, because it almost never strays from its main course. The one thing to keep watching is whether global expansion will dilute the exceptional manufacturing returns TSMC has historically earned in Taiwan.

Moat strength score: 5/5. Management and capital allocation score: 4/5.

Financial Quality and Owner Earnings

The table below organizes TSMC's core financial metrics for the past five years on a consolidated basis, in NT$bn, with EPS as diluted earnings per share. Some margins and cash-conversion ratios are values calculated from annual-report/20-F data.

How should the financial quality be read? Fact: From 2021 to 2025, revenue CAGR was about 24% and net-profit CAGR about 30%, with free-cash-flow CAGR even higher but also more volatile; when revenue fell back in 2023, gross margin and free cash flow came under pressure in tandem, proving that the company, though strong, is still cyclical. The 2024–2025 recovery was very powerful, showing TSMC's clear profit elasticity in a recovering cycle and an AI upswing. Its operating cash flow has long exceeded net profit, indicating profit is not "paper profit"; but there have been quite a few years when free cash flow fell below net profit, mainly because of the industry's enormous capital expenditure.

Fact: The latest financial structure is very solid. In Q1 2026, the company held cash and marketable securities of NT$3,383.6bn against long-term interest-bearing debt of NT$901.0bn, a markedly net-cash position; total liabilities were 31.5% of total assets, and the current ratio was 2.5x. Interest expense in 2025 was about NT$12.4bn, and relative to 2025 operating profit, interest coverage is extremely high. In terms of "can it weather an economic downturn," TSMC is not a fragile balance sheet.

Fact: Operating metrics show no signs of deterioration. In Q1 2026, days sales outstanding were 26 days and days inventory outstanding 80 days, both within a controllable range; inventory rose slightly from end-2025 but showed no obvious loss of control. The share count is broadly stable: common shares were 25.93bn at end-2022 and about 25.93bn at end-March 2026; the company does no routine buybacks and shows no meaningful dilution.

Inference: TSMC's profit is largely "real cash profit," not profit piled up through aggressive accounting. The company has long been audited with effective internal controls; cash flow and profit move in the same direction, and there is no obvious sign of compressing R&D, deferring expense recognition, or abnormally eroding balance-sheet quality for short-term profit. The real thing to watch is not "fraud" but whether investors will mistake high-cycle margins for a permanent normal.

Owner Earnings Analysis Fact: In 2025, TSMC's operating cash flow was about NT$2,275.0bn, capex about NT$1,272.4bn, and traditional free cash flow about NT$1,002.6bn.

Assumption: But the key to Buffett-style "owner earnings" is not total capex, but maintenance capex. TSMC does not disclose maintenance capex, so it can only be estimated. Given the significant 2025–2026 expansion across the U.S., Japan, Germany, and Taiwan, and substantial 2nm/advanced-packaging build-out, a considerable portion should be treated as growth capex rather than the full necessary spend to maintain the existing competitive position. For a conservative investor, this report uses NT$900bn as a "conservative maintenance capex" estimate for 2025; this is higher than that year's depreciation and amortization, to avoid being overly optimistic.

Inference: Thus, estimating owner earnings ≈ operating cash flow − maintenance capex, TSMC's 2025 conservative owner earnings are about NT$1.38tn. This is clearly higher than traditional free cash flow, but also lower than the optimistic method of "net profit + all depreciation − all capex treated as growth investment." At a current market capitalization of about NT$58tn, that implies roughly 42x conservative owner earnings; even under looser assumptions, the multiple is not low. In other words, the market does not view TSMC as an ordinary manufacturer but prices it as a "super-scarce AI-infrastructure asset."

Intrinsic Value and Margin of Safety

Owner-Earnings Discount Method

The valuation below is an assumption, not a fact. It is meant to answer "roughly what expectations does the current price imply," not to give a precise target price.

Core Assumptions:

• Starting owner earnings: conservative NT$1.30tn, neutral NT$1.40tn, optimistic NT$1.50tn

• Growth in the first five years: 8% / 12% / 15%

• Growth in the next five years: 4% / 5% / 6%

• Discount rate: 11% / 10% / 9%

• Perpetual growth: 2.5% / 3.0% / 3.5%

• Share count: about 25.93bn shares (as of end-March 2026)

Opinion: For a conservative long-term investor, this result matters: the current price of NT$2,235 is already above this report's neutral valuation, and even above the midpoint of the optimistic scenario. If you believe AI demand will grow at very high rates for the long run, that overseas expansion will not dilute returns, and that TSMC can sustain its extraordinary 2026–2030 earnings for a very long time, then you can derive a higher value than this report does; but that already begins to rely heavily on "optimistic assumptions" rather than on a margin of safety.

Relative Valuation Method

Fact: The market currently assigns TSMC a valuation markedly higher than most traditional manufacturers or mature foundry peers. By recent public market data, TSMC's Taiwan shares carry a TTM P/E near 29.7x and a P/B near 9.6x; UMC's U.S. ADR has a TTM P/E near 28.2x, P/B near 3.4x, and EV/EBITDA near 11.4x; GlobalFoundries has a TTM P/E near 51.1x, P/B near 3.4x, and EV/EBITDA near 15.9x; Samsung Electronics at the group level has a TTM P/E near 22.1x, P/B near 3.8x, and EV/EBITDA near 10.9x. Among these, Samsung Electronics is not a pure-play foundry and cannot be compared directly; but it at least shows that TSMC's current valuation embeds a substantial "quality-scarcity premium."

Inference: TSMC deserves a higher valuation than UMC, GFS, or even the Samsung group, because its leading-edge share, customer stickiness, scale, and profit quality are clearly better; but "deserves to be more expensive" does not equal "any price is reasonable." If part of the AI-cycle dividend is mistaken for a permanent normal, then relative valuation will be distorted. At most the relative method can prove "TSMC should be a premium asset"; it cannot prove "it is cheap right now."

Asset or Liquidation Value Method

Fact: As of Q1 2026, total shareholders' equity was about NT$5,932.4bn, corresponding to book net assets per share of roughly NT$229; cash and marketable securities were about NT$3,383.6bn, markedly higher than long-term interest-bearing debt of NT$901.0bn.

Opinion: For TSMC, the asset method is not the main valuation anchor, because its core value lies not in the steel and concrete on its books but in leading-edge processes, manufacturing know-how, customer trust, the ecosystem, and execution. Conversely, under "liquidation thinking," the current share price is far from supported by net cash or book value; it is mainly supported by future cash flows and the durability of the moat. The asset method is therefore better used for TSMC as a downside-boundary reminder than as the main anchor for fair value.

Composite Valuation Conclusion:

• Conservative intrinsic-value range: NT$800–1,050

• Fair intrinsic-value range: NT$1,150–1,650

• Optimistic intrinsic-value range: NT$1,900–2,600

• Current price relative to fair value: roughly in the premium zone

• Required margin of safety: at least 20%–30%

• Ideal buy-price range: NT$1,100–1,400

• Acceptable holding-price range: NT$1,400–1,900

• Clearly overvalued price range: above NT$2,100

These ranges are not precise points but a "value band" that folds in cash-flow assumptions, capital intensity, and geopolitical risk.

Margin-of-safety judgment: Insufficient. The current price is not absurd, but for a balanced, conservatively inclined long-term investor it already lacks an adequate buffer. If future growth falls short, margins revert, or the valuation multiple contracts, returns can easily shift from "an excellent company's excellent compounding" to "an excellent company's ordinary returns." This is the classic case of a good company, but not necessarily a good price.

Risks, the Bear Case, and Opportunity Cost

The most important risk is not short-term price volatility but permanent loss of capital.

Competition and technology-substitution risk. If Samsung meaningfully improves in 2nm/advanced packaging, or if Intel Foundry achieves a key customer breakthrough in domestic U.S. manufacturing and advanced packaging, TSMC's bargaining power and the pace of share gains could both slow. TSMC itself states clearly in its 20-F that if it cannot maintain technology leadership or respond in time to market changes brought by AI, its competitive position could be impaired.

Customer-concentration and AI-cycle risk. A structure of 19% largest customer, 17% second largest, and 78% top ten means that any change in one or two customers' design cadence, internal inventory, capex decisions, or foundry-diversification strategy will significantly affect TSMC. In particular, the strong profit of 2025–2026 is to a large degree tied to the AI/HPC boom and the scarcity of leading-edge nodes and advanced packaging. If AI investment slows, both margins and valuation could fall back.

Geopolitical and export-control risk. This is the tail risk that can least afford to be underestimated. TSMC clearly discloses that its principal management and most of its core production facilities are located in Taiwan, and that geopolitical tensions, military conflict, and escalating trade and export controls could all affect its business, supply chain, and securities-market price; the company also discloses that its Nanjing, China fab depends on the renewal of U.S. export licenses. For a value investor, this is not a risk that can be brushed off with "historical volatility."

Natural-disaster, infrastructure, and utility risk. TSMC's 2025 20-F states directly that Taiwan earthquakes in 2024 and early 2025 caused losses of about NT$3bn and NT$5.3bn respectively, and warns that a prolonged interruption of power, water, or a natural disaster could significantly impair capacity and delivery capability.

Capex and overseas-expansion return risk. The 2026 capex guidance is as high as US$52bn-56bn. If overseas expansion carries high costs over the long run and falls short of Taiwan's home efficiency, or if subsidy realization, labor, construction, and the volume-production ramp fall short of expectations, TSMC's historically very high returns on capital could be diluted.

What is the strongest bear case? The strongest short thesis is actually quite simple: This is a great company, but the market is already pricing in the story that "it will retain a near-monopoly on high-end AI chip manufacturing for the next decade, advanced packaging will stay scarce, margins will hold high for the long run, and geopolitics will not truly hit the valuation." If any one of these conditions is discounted, then even if the company itself remains excellent, investors who bought at a high valuation could still face years of low returns. In other words, someone shorting TSMC is not necessarily betting the company will get worse, but betting that "its excellence is not enough to support the current price."

What facts would overturn this report's investment judgment? If the following facts occur, this report's "current margin of safety is insufficient" judgment would need to be revised upward, even toward more positive:

• 2nm/N2P/A16 volume-production progress, yield, and customer adoption keep beating expectations;

• Overseas fab return rates prove that overall ROIC is not meaningfully diluted;

• 2026–2028 owner-earnings growth stays above 15% and does not depend on more aggressive capital investment;

• Valuation falls back closer to fair value while the fundamentals remain strong.

Conversely, if the following facts appear, I would acknowledge the research logic needs to be revised downward:

• Leading-edge-node/advanced-packaging share is continuously taken by Samsung or Intel;

• Gross margin falls below roughly 53%–55% for the long run with no clear path to repair;

• Capex stays high for the long run but free cash flow and owner earnings no longer grow in tandem;

• Key customers begin to systematically diversify their high-end capacity;

• Taiwan-related political/military/export-control risk materializes.

Comparison with Other Opportunities Compared with the S&P 500, TSMC is a single asset of higher quality and higher concentration; the S&P 500 currently trades at a P/E of about 21.5x, while TSMC is about 29.5x. Compared with the U.S. 10-year Treasury yield of about 4.6%, TSMC's potential excess return at the current price is not obviously and overwhelmingly superior, unless you have very high conviction in long-term AI demand and TSMC's moat. The conclusion is: it is not obviously better than buying the index, much less obviously better than a risk-free asset; its advantage lies in long-term business quality, not in being cheaply valued right now.

If I could hold only 5 assets, would it qualify for the portfolio? Opinion: On business quality, it qualifies; on current price, for fresh capital allocation it may not, for now, deserve priority over cheaper high-quality opportunities or a more diversified index position.

Investment Checklist and Final Judgment

Investment Checklist

Final Investment Conclusion

【Final Rating】 Watch

【One-Sentence Investment Thesis】 TSMC is one of the very few world-class companies that combine technology leadership, scale advantage, customer trust, and cash-generation power, but at the current Taiwan share price, the company is still great, while the margin of safety is no longer great.

【Core Bull Case】

• The pure-play foundry model is inherently neutral, does not compete with customers, and forms a unique trust moat over the long run.

• It holds hard-to-replicate technology, scale, yield, and ecosystem advantages in leading-edge nodes and advanced packaging, with a pure-play foundry market share of roughly 70%.

• Revenue, profit, and cash flow were all strong in 2025 and Q1 2026, showing AI/HPC demand is still being realized.

• The balance sheet is very solid, with ample net cash, enough to support large-scale reinvestment and cycle resilience.

• Capital allocation is focused on the core business, with few acquisitions and few buybacks, centered on reinvestment and dividends, friendly to long-term shareholders.

【Core Bear Case】

• The current valuation is already high and lacks an adequate margin of safety.

• Customer concentration is high, and high profitability is closely tied to the current AI boom.

• Taiwan-related geopolitics, export controls, and the tail risk of war or blockade cannot be fully absorbed by traditional valuation models.

• The high cost structure from overseas expansion could gradually dilute historical returns on capital.

• However excellent this business is, it still carries the cyclical volatility inherent to the semiconductor industry.

【Key Assumptions】

• 2nm/N2P/A16 and advanced packaging ramp on schedule and maintain a yield lead;

• The structural growth in AI/HPC demand can persist for at least several years;

• Overseas expansion does not significantly lower the overall return rate;

• No substantive damaging event occurs from geopolitics or export controls;

• The company continues to maintain high-quality governance and capital discipline.

【Fair Buy Price】 The more suitable long-term buy range: NT$1,100–1,400. The basis: within this report's conservative-to-neutral valuation range, leaving at least a 20%–30% margin of safety is more consistent with "balanced, conservatively inclined" capital discipline.

【Target Holding Period】 More than 10 years. TSMC is not suited to deciding by "will it rise next quarter," but better suited to deciding by "will the value chain of global high-end compute and logic foundry keep concentrating in it over the next decade."

【Expected Annualized Return】 The following is an assumption, not a promise:

• Conservative scenario: 0%–4%/year

• Neutral scenario: 5%–8%/year

• Optimistic scenario: 9%–12%/year

These ranges already account for: a high entry valuation, future growth that may stay strong, but a valuation multiple that may pull back.

【Maximum Loss Risk】 If a severe geopolitical shock, a loss in leading-edge-node competition, or a marked reversal of the AI boom occurs, a phased price drop of more than 50% is entirely possible; under an extreme Taiwan-risk scenario, permanent loss of capital could be larger. The biggest risk here is not a "quarterly earnings miss" but valuation and geopolitics compressing at the same time.

【Tracking Metrics】 Going forward, keep tracking:

• 3nm/2nm/N2P/A16 volume-production progress and yield

• Supply-demand and gross-margin contribution of advanced packaging, especially CoWoS

• HPC/AI revenue share and the share of the top two customers

• Whether gross margin and operating margin can stay high

• The capex/revenue ratio, and the return rate of overseas fabs

• Operating cash flow, free cash flow, and owner-earnings coverage of net profit

• Net-cash level and changes in long-term debt

• Export controls and the status of the Nanjing fab's licenses

• Taipower electricity supply, water resources, earthquakes, and operational interruptions

• The pace of technology catch-up by Samsung and Intel Foundry

【Signals That Trigger Re-Evaluation】

• A marked narrowing of the competitive advantage in leading-edge nodes or advanced packaging

• Gross margin falling below roughly 53%–55% for the long run

• Capex staying high for the long run while cash returns deteriorate markedly

• The largest customer shifting orders significantly or building its own substitute

• Taiwan-related risk materializing

• The current high valuation continuing to climb while fundamental growth begins to slow

【Final Recommendation】 If what you seek is "owning an outstanding company for the long run," TSMC is almost certainly worth continuing to study and keeping on the watchlist for the long run; if what you seek is "buying an outstanding company at a clear discount," then right now looks more like a waiting zone than an aggressive entry zone. The calm conclusion is: the business passes, the price does not yet.

Open questions / limitations

• TSMC does not separately disclose maintenance capex, so owner earnings can only be estimated as a range.

• Samsung's foundry business has no independently listed financials, so cross-company valuation can only partly borrow Samsung Electronics' group-level figures, with limited comparability.

• The current price uses public market data close to May 21, 2026, and will change with the market thereafter.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.