TSMC: A Deep-Dive Value Investing Study

Hunting ten-year five-baggers among great growth stocks — pressing the upside question: "Can it get much bigger?"

• QHow large is its market ceiling? Is it expanding an existing pie, or creating an entirely new market?7/10

  A

  The ceiling is very high, but the essence is expanding and nearly monopolizing an existing pie with extremely high value density, rather than creating a brand-new market from scratch. That is exactly why it can command a high valuation, while still not fitting Baillie Gifford's favorite type of company: a new category entering explosive growth.

  Start with the size of the pie itself. TSMC is not serving the broad "semiconductor industry" in a generic sense. Its business is more specifically advanced logic foundry + advanced packaging. Based on the company's own disclosures, the report records that the Foundry 2.0 industry it operated in grew 16% in 2025, while TSMC's U.S. dollar revenue grew 35.9%, clearly outpacing the industry. On TrendForce's definition, TSMC's share of the pure-play foundry market was about 67.6% in 2025Q1 and rose to 70.4% in Q4, while Samsung was only about 7%. In other words, this is the chokepoint pie for global compute and mobile SoCs, and TSMC has already taken about 70% of it.

  Why is this "expanding an existing pie" rather than "creating a new market"? Because foundry has existed for decades. TSMC did not invent a new demand category. It has continuously taken the layer with the highest value density, namely AI accelerators, GPUs, and flagship smartphone SoCs, away from customers' in-house capacity and competitors, while thickening the entire compute pie. The report's data confirms this upward migration of value: HPC accounted for 58% of revenue in 2025 and rose to 61% in 2026Q1; advanced processes at 7nm and below accounted for 74% of wafer revenue. The demand ceiling is determined by downstream AI/HPC capital spending. This is a real, still-expanding long runway, but it is a structural volume expansion in the existing compute market, not virgin territory opened up by TSMC alone.

  The other side of the ceiling is cyclicality and concentration. The report honestly notes that the industry is "extremely difficult for new entrants to disrupt, yet can still be disturbed by changes in technology paradigms, state-subsidized competition, export controls, customers building their own capacity, or a pullback in AI investment." In 2025, 75% of revenue came from North America, the top 10 customers together accounted for about 78%, and the largest customer accounted for about 19% and the second largest for about 17%. This means the ceiling is high, but it depends heavily on the capex rhythm of a small number of super customers, rather than a dispersed, self-propelled new consumer market.

  The honest conclusion for Baillie Gifford's question, "how large is the market ceiling?": the runway is long and penetration is still continuing, across advanced nodes, advanced packaging, and overseas capacity, so a high score around 4 is justified. But its growth comes from making an existing, extremely valuable pie larger and nearly monopolizing it, not from creating a brand-new market that did not previously exist. This is a world-class business, but on the Baillie Gifford dimension that prizes the imagination of explosive new-category growth most highly, it is not the most extreme type.

  Jun 10, 2026
• QCan its revenue at least double over the next five years? Will growth be driven mainly by volume, price, or new businesses?6/10

  A

  A doubling of revenue over the next five years, or a CAGR of about 15%, is achievable but sits near the optimistic upper bound rather than the base case. Growth will mainly be driven by volume plus high-end mix-related pricing, while the new business, advanced packaging, is an amplifier rather than an independent second engine.

  Start with the benchmark. TSMC's growth over the past five years has been extremely strong: the report shows revenue rising from NT$1,587.4bn in 2021 to NT$3,809.1bn in 2025, a CAGR of about 24%. In 2024→2025 alone, revenue grew 31.6% and net income grew 46.4%. In other words, it has already far exceeded a doubling over the past five years. But that exceptional growth rode the sweet spot of the AI/HPC boom, so extrapolating it into the next five years requires a discount.

  To answer whether it can double over the next five years, the key is whether the drivers can continue. They break into three parts:

  Volume, through capacity and penetration. This is the most tangible piece. The report records 2025 capacity of more than 17 million 12-inch-equivalent wafers, while 2026 capex guidance is as high as US$52bn–56bn, versus about US$40.9bn actually spent in 2025, with simultaneous expansions in the United States, Japan, Germany, and Taiwan. Sustained high capex will translate into continued volume growth in advanced-node capacity. That is the physical foundation for volume growth.

  Price / high-end mix. TSMC's "price increases" show up more as a shift in product structure: 2nm entered volume production in 2025Q4, N2P/A16 are progressing on schedule, and wafer ASP rises as nodes move forward. The report shows advanced nodes, defined as 7nm and below, already at 74% of revenue, while gross margin rose from 54.4% in 2023 to 59.9% in 2025 and further to 66.2% in 2026Q1. That is exactly the value capture brought by mix upgrade plus tight advanced packaging supply.

  New business, advanced packaging CoWoS. This is the most attractive incremental piece. CoWoS is the key capacity bottleneck for AI chips, and TSMC is expanding capacity sharply. But the honest point is that advanced packaging is currently an amplifier and companion to the main curve, advanced logic foundry. Its scale is not yet enough to support a doubling by itself. It strengthens wallet share from the same group of AI customers rather than opening a new revenue pole.

  The arithmetic of doubling and the honest judgment. Doubling in five years requires a compound growth rate of about 15%. The report's neutral DCF assumption sets "first-five-year growth" at 12%, with the optimistic case only reaching 15%. In other words, doubling corresponds to the report's optimistic scenario, not the neutral base case. The report's core concern is exactly this: the current NT$2,235 price, which had risen to about NT$2,295 by early June, already implies optimistic expectations of long-term ultra-high AI demand growth and overseas expansion that does not dilute returns. For Baillie Gifford's hard threshold of "can it double in five years?", the honest answer is: yes, but only if AI capex strength does not fade, 2nm yields and customer adoption both continue to beat expectations. This is conditional optimism, not a high-certainty base case. In the growth structure, volume + mix are the main drivers and advanced packaging is the assist, rather than a brand-new business driving the doubling.

  Jun 10, 2026
• QAfter five years, what will take over as the next growth engine? Does this "second curve" exist today?5/10

  A

  TSMC's "second curve" does exist today and is scaling: advanced packaging, namely CoWoS/SoIC. But it is essentially an extension and companion to the main curve, advanced logic foundry, rather than a new growth pole unrelated to foundry that can independently take over. This is the subtle gap between TSMC and growth companies that truly have a second engine.

  First, does the successor already exist today? The answer is yes, and it is visible in the financials. The report records that TSMC's moat has expanded in recent years from "advanced process" to a dual wheel of "process + packaging": 2nm entered volume production in 2025Q4, while A16/N2P are progressing on schedule. At the same time, the importance of advanced packaging has risen materially. CoWoS, Chip-on-Wafer-on-Substrate, has become the key capacity bottleneck for AI chips, and about 10–20% of the US$52bn–56bn capex in 2026 is allocated to advanced packaging such as CoWoS/SoIC. The report explicitly lists tight advanced packaging supply as one of the important sources of high gross margin in 2025. So does the second curve exist today? It does, and it is already contributing profit.

  But to answer Baillie Gifford's real question, "what will take over as the next growth engine?", we must distinguish between two kinds of second curves:

  The first is vertical extension of the main curve. Advanced packaging still serves the same AI/HPC customers and is tied to the same advanced logic wafers. It expands TSMC's wallet share per AI chip. This is a very high-quality incremental opportunity, but it is one of two gears in the same demand engine as foundry. If AI capex falls back, advanced logic and advanced packaging will come under pressure together, rather than hedging each other. The report's data confirms this shared source: HPC accounted for 58% of revenue in 2025 and rose to 61% in 2026Q1, while the prosperity of advanced nodes and advanced packaging is highly synchronized.

  The second is a truly independent new growth pole, a separate leg that can take over when the main business slows. TSMC's "third curve" in this sense is not yet clear: it does not make memory, it does not make its own chips, and overseas expansion is moving the same business into different geographies rather than a new business. The report indirectly acknowledges this: all tracking indicators, including 3nm/2nm/N2P/A16 volume production, CoWoS supply and demand, HPC share, and returns on overseas fabs, revolve around the same advanced foundry business.

  The honest conclusion for this Baillie Gifford question: TSMC's "second curve", advanced packaging, genuinely exists today and is scaling. Continued node migration, from 2nm→A16, also ensures that the main curve itself will not peak quickly, giving it better growth visibility than most manufacturing companies. But this second curve is an extension of the main curve rather than an independent substitute. It reinforces the same AI demand engine. If the overall valve of AI/HPC capital spending structurally falls back, TSMC lacks a new growth leg decoupled from it to take over. This is one reason the report rates it as "Watch" rather than an unconditional long-term hold: the business quality is exceptional, but the independence of the second curve is only "present, but not extreme" under the Baillie Gifford framework.

  Jun 10, 2026
• QWhat is its core competitive advantage? Will this moat widen or narrow over the next 3 to 5 years?7/10

  A

  The core advantage is an integrated triad of technology leadership + manufacturing excellence + customer trust, layered with scale and ecosystem lock-in. Over the next 3 to 5 years, the moat will most likely be stable to widening, but the pace of widening will be partly offset by high-cost overseas expansion and potential competitive catch-up.

  The moat does not come from a single source. It comes from multiple advantages that the report repeatedly emphasizes and that reinforce each other layer by layer:

  Share and scale, the hardest objective evidence. In the pure-play foundry market, TSMC's share was about 67.6% in 2025Q1 to 70.4% in 2025Q4, while Samsung was only about 7.1%; the annual share gap widened from 55 percentage points in 2024 to 62.7 percentage points in 2025. This "lead still widening" is itself a direct reading that the moat is expanding. The report records 2025 capacity of more than 17 million 12-inch-equivalent wafers and revenue of NT$3,809.1bn. Capital intensity, talent density, and the learning curve mean that later entrants are "hard to replicate within 3–5 years even if they have money."

  Neutrality and trust from the pure-play foundry model. This is TSMC's unique structural advantage: it does not design or sell its own branded chips, so it does not compete with customers for profits. The report notes that this makes TSMC a neutral platform that "all IC innovators are willing to work with." The fact that customers entrust their most critical CPUs, GPUs, and AI accelerators to it is itself an industry vote of trust. Samsung has its own chip businesses inside the group, creating an inherent trust discount. This is a soft moat that TSMC's rivals cannot symmetrically copy.

  Switching costs and ecosystem position. The report points out that the real lock-in is not merely "changing fabs after tape-out is inconvenient." It is the deep switching cost jointly created by process libraries, the EDA/IP ecosystem, packaging collaboration, yield databases for mass production, and confidentiality systems. Once a leading AI chip successfully enters mass production at TSMC, migration to another foundry involves yield risk as well as time and intellectual-property costs. This creates a rare ecosystem-type moat within manufacturing.

  Why judge "widening" rather than "stable"? There are three forward-looking pieces of evidence: first, 2nm entered volume production in 2025Q4 and A16/N2P are progressing on schedule, so the node gap is still opening; second, advanced packaging, CoWoS, has become a key AI bottleneck, making the "process + packaging" advantage more complete; third, advanced-node revenue share has continued rising to 74%, deepening high-end customer binding.

  But the "narrowing risks" must be stated honestly, and this is where the report is restrained. The report is clear that what could truly narrow the moat is not ordinary competition, but "long-term erosion from overseas manufacturing costs, key customers diversifying capacity, and Samsung/Intel significantly catching up in advanced logic or advanced packaging." The most realistic slow erosion is dilution of returns from overseas expansion. 2026 capex of US$52bn–56bn is being deployed heavily in the United States, Japan, and Germany, and overseas fabs have higher cost structures than Taiwan fabs. Over time, this may dilute the exceptional returns of manufacturing in Taiwan.

  The conclusion for Baillie Gifford's question, "will the moat widen or narrow over the next 3 to 5 years?": the direction is stable to widening, supported by three confirmations: a widening share gap, node-generation gap, and packaging complementarity. This is the basis for giving the moat a high score around 5. But the slope of widening will be flattened by high overseas costs and competitive catch-up. It is not a frictionless upward line. The moat itself is TSMC's strongest and most defensible dimension.

  Jun 10, 2026
• QIf its core business is disrupted, does it have the DNA to reinvent itself? How does it handle mistakes and bad news?6/10

  A

  TSMC's reinvention DNA shows up as a gradual regenerative ability to keep crossing technology paradigms along the same main channel, from furnace tubes to FinFET to GAA nanosheets, rather than as a radical transformation gene that can build a new business from scratch after the core business is disrupted. It handles bad news pragmatically, transparently, and with quantifiable disclosure, but this reinvention ability has not yet been tested by a true life-or-death crisis.

  Break this Baillie Gifford question into two layers: reinvention DNA + how it handles mistakes and bad news.

  Reinvention DNA: present, but it is the "same-channel paradigm leap" type. Over decades, TSMC has repeatedly proven one thing: when transistor physics approaches its limits, it can be among the first to switch to the next technology paradigm and maintain leadership. The latest example is that 2nm, or N2, entered volume production in 2025Q4, the company's first use of GAA nanosheet transistors, a leap from FinFET to gate-all-around. The moat recorded in the report has also expanded in recent years from "advanced process" to the dual wheel of "process + packaging," another form of active capability regeneration. This shows a very strong engineering culture of continuously surpassing itself within the existing track.

  But two kinds of "reinvention" must be separated honestly. What Baillie Gifford is really asking is whether, if the core business is disrupted, the company can change its life path and survive. TSMC's regenerative ability is highly concentrated in the main channel of "advanced logic foundry + packaging." It does not make memory, does not make its own chips, and overseas expansion is moving the same business into different geographies. If a disruptive paradigm replacement occurs, such as a new computing paradigm that bypasses silicon-based advanced lithography, or customers building large-scale internal capacity and changing the foundry model itself, TSMC does not have a ready-made "second life." All of the report's tracking indicators revolve around the same business, indirectly confirming that this single-channel focus is both an advantage and a constraint. Compared with the Baillie Gifford calibration names, TSMC's regenerative quality is closer to "continuously optimizing the existing business" than to the crisis reinvention of lunr/SDGR, which can rewrite themselves when the core is disrupted.

  How it handles mistakes and bad news: this item is excellent and verifiable. This is a highlight of TSMC's governance quality:

  • It discloses bad news quantitatively and without avoidance. The report records that in the 2025 20-F, the company directly disclosed that earthquakes in Taiwan in 2024 and early 2025 caused losses of about NT$3bn and NT$5.3bn respectively, and acknowledged that prolonged interruptions in electricity, water resources, or natural disasters would weaken capacity. It lays out risks item by item and with amounts, rather than dressing them up.
  • Governance and internal controls stand up to audit. The report shows that in the 2025 Form 20-F, management assessed the effectiveness of internal controls, and Deloitte issued an unqualified internal-control audit opinion. Cash flow and profit move in the same direction, with no signs of compressing R&D for short-term profit or abnormal balance-sheet erosion.
  • It admits mistakes and stays focused in capital allocation. The recent sale of the VIS stake to focus on the core business fits the discipline of concentrating resources on the core.

  The conclusion for this Baillie Gifford question: TSMC's attitude toward mistakes and bad news is world-class: transparent, quantitative, and pragmatic. That deserves a high score. But the "reinvention DNA" deserves a discount. Its regenerative power is continuous paradigm migration within the same channel, strong in engineering culture but weaker in radical transformation after core disruption, and this capability has not yet been tested by a true life-or-death crisis. On Baillie Gifford's "resilience and regeneration" dimension, it is robust and reliable, but not extreme.

  Jun 10, 2026
• QDoes management, especially the founder, have a long-term perspective and deep alignment with the company? Is it willing to sacrifice current profit for 5 to 10 years out?5/10

  A

  Management is world-class in long-term perspective, honesty, and capital discipline, clearly better than most technology manufacturing companies. But on the Baillie Gifford/Buffett dimension that prizes very high founder or management ownership and deep alignment with the company, TSMC has a clear weakness: it is driven by institutions and culture, not founder control.

  Start with the hardest sub-question, "alignment of interests," using primary data:

  Equity alignment is weak. As of March 31, 2026, Chairman and CEO C.C. Wei personally held 8,152,610 shares, about 0.03% of total share capital, with total share capital of about 25.93bn shares. The report also records that the National Development Fund held about 6.38%. In other words, TSMC is not a company with very high founder ownership. This must be acknowledged honestly. The helmsman's wallet is far less aligned with minority shareholders than in Baillie Gifford calibration names where founders hold 30%+ stakes, such as the lunr founder's 33.5% stake and the SDGR founder's heavy ownership. On the Baillie Gifford hard metric of "deep alignment of interests," TSMC is objectively weak.

  But on "long-term perspective" and "willingness to sacrifice current profit for 5 to 10 years out," TSMC is textbook-level. This is what makes it distinctive: even without founder control, its systems and culture are deeply long-term oriented. The most direct evidence is capex discipline. 2026 capex guidance is US$52bn–56bn, and the report records capex of NT$949.8bn / 956.0bn / 1,272.4bn in 2023–2025. This willingness to invest heavily today for capacity leadership 3–5 years later, even at the cost of sharply lowering current free cash flow, is the essence of "sacrificing current profit for 5 to 10 years out." The report also confirms that free cash flow has often been below net income, mainly because industry capex is enormous. Management chose to put money into the future rather than beautify the current period.

  Capital allocation is highly rational and friendly to long-term shareholders. The report records that the company has long relied mainly on cash dividends and does not conduct routine share repurchases. The 20-F explicitly states "issuer and affiliated purchaser purchases of equity securities: not applicable", showing that it did not use buybacks to beautify earnings per share. In recent years, cash use has been clear: reinvestment in expansion + higher dividends + maintaining a sound balance sheet. There have been no large, high-premium acquisitions that destroy value, and the recent sale of the VIS stake sharpened focus on the core. The report scores capital allocation clearly above most technology manufacturers because TSMC "almost never deviates from the main channel."

  Honesty and internal controls are credible. The report records that in the 2025 Form 20-F, management assessed internal-control effectiveness, and Deloitte issued an unqualified internal-control audit opinion. The overall governance record is strong.

  The conclusion for this Baillie Gifford question: look at two sides. "Long-term perspective + willingness to sacrifice the current period for the long term + capital discipline + honesty" are close to full marks, and that is the basis for a high management score. But "very high founder/management ownership and deep interest alignment" is a clear weakness, with the CEO holding only 0.03% and no founder control. TSMC is a world-class company driven by systems, culture, and an operating system, not by founder-wallet alignment. The only risk that needs continuous monitoring is the one flagged by the report: whether global expansion will dilute the historically exceptional returns of manufacturing in Taiwan. That is the true long-term test of management's capital-allocation judgment.

  Jun 10, 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth sustainable and not dependent on harming society or regulators?7/10

  A

  If TSMC disappeared tomorrow, global AI and high-end electronics would miss it at the level of an irreplaceable paralysis. It is the single-point lifeline of modern compute. Its growth model is highly healthy and does not rely on harming users or regulatory arbitrage, but indispensability itself has pushed it into the center of geopolitics and export controls. That is the true variable for its sustainability.

  This Baillie Gifford question has two layers: indispensability + social/regulatory sustainability.

  First layer: indispensability, extremely high and close to an industry single-point lifeline. Objective evidence: TSMC's share of the pure-play foundry market was about 67.6% in 2025Q1 to 70.4% in 2025Q4, while Samsung was only about 7%. The report records that the world's most critical CPUs, GPUs, AI accelerators, and flagship smartphone SoCs are almost all taped out at TSMC. HPC accounted for 58% of revenue in 2025 and rose to 61% in 2026Q1, while advanced processes at 7nm and below accounted for 74% of wafer revenue. Even more important is advanced packaging: CoWoS is already the key capacity bottleneck for AI chips, and is also almost monopolized by TSMC. In the short term, no company can approach it simultaneously in advanced nodes + advanced packaging + yield + scale + delivery certainty. So how much would customers miss it if it disappeared tomorrow? The answer is that the entire AI wave would lose speed. This is indispensability at the highest intensity.

  Second layer: whether the growth model is sustainable and does not depend on harming society or regulators. TSMC is quite clean on this layer, but has a unique geopolitical constraint. There are two sides:

  The healthy side, not relying on harm. TSMC earns money "properly": it is a neutral foundry platform, does not compete with customers for profits, does not harvest user data, and does not profit from regulatory arbitrage or gray areas. Growth comes from real technology leadership and capacity investment. The report records gross margin rising from 54.4% in 2023 to 59.9% in 2025 and 66.2% in 2026Q1, driven by high-end mix and yield rather than exploitation. Its indispensability is built on "others cannot do it," not "others are not allowed to do it." That is the healthiest source of moat.

  The unique constraint side, the real regulatory/social sustainability risk. Precisely because it is so indispensable, TSMC has become a focus of geopolitics. The report explicitly discloses that its principal management and most core production facilities are located in Taiwan, and that geopolitical events, military conflict, and escalation of trade and export controls may affect business, supply chains, and securities market prices. The company also discloses that its Nanjing fab in China depends on renewal of U.S. export licenses. In other words, TSMC's indispensability is a double-edged sword: it makes the world unable to live without TSMC, and it also pulls TSMC into great-power competition. Export controls, subsidy competition, and the politicization of capacity are real threats to "sustainability." The report lists this as "the tail risk that must not be underestimated."

  The conclusion for this Baillie Gifford question: a full-mark-level score for indispensability is entirely reasonable. This is one of the rare assets on earth whose disappearance tomorrow would paralyze an industry. Its growth model itself is healthy and does not rely on harming society or regulatory arbitrage. That is also clean. But the sustainability variable is not "will users abandon it" (they will not); it is "will geopolitics and export controls materially sever it." That is the reverse cost of TSMC's indispensability and one of the core reasons the report rates it as "Watch" rather than a mindless long-term hold.

  Jun 10, 2026
• QWhat are the unit economics of this business, including gross margin and incremental returns? Does scale make it better or worse? Where does the money it earns go?7/10

  A

  The unit economics are an extremely rare combination in heavy-asset manufacturing: high gross margin + high cash conversion + strong operating leverage. During upcycles, scale makes the business better because leverage amplifies returns, but there are two concerns: overseas expansion will dilute marginal returns, and current margins contain an obvious cyclical boom component. The money it earns mainly goes to reinvestment in capacity expansion and dividends, with strong discipline.

  Break this Baillie Gifford question down item by item.

  Gross margin and profitability: an outlier in manufacturing. The report records TSMC's 2025 gross margin at 59.9%, operating margin at 50.8%, and net margin at 45.1%, verified by net income of NT$1,717.88bn / revenue of NT$3,809.1bn. In 2026Q1, these rose further to 66.2% / 58.1% / 50.5%. For a high-fixed-cost, high-capital-investment manufacturer to reach nearly 60% gross margin and 50% operating margin is almost unique in global manufacturing. The quality of unit economics is extremely high.

  Does scale make it better or worse? In an upcycle, it makes it "better." TSMC is a classic high-operating-leverage business. Once high fixed costs are spread over high utilization and tight advanced-node supply, incremental margins are extremely high. The report confirms this elasticity: gross margin rose from 54.4% in 2023 to 59.9% in 2025 and 66.2% in 2026Q1, "not merely accounting beautification, but more like the combined effect of high-end product mix, tight advanced packaging supply, technology leadership, and scale efficiency." In other words, during the AI upcycle, bigger means more profitable, and incremental returns rise.

  But two concerns must be stated honestly, and this is where the report is restrained:

  • Cyclicality is real. The report clearly notes that when revenue fell in 2023, gross margin and free cash flow came under pressure at the same time, "proving that although the company is strong, it remains cyclical." A meaningful part of today's exceptionally high margin comes from AI/HPC strength and tight advanced packaging supply, so it should not be treated as a permanent normal state. The report repeatedly warns that the real risk is whether investors mistake boom-period margins for a permanent normal.
  • Incremental returns may be diluted by overseas expansion. This is the most realistic slow erosion of unit economics. 2026 capex guidance is US$52bn–56bn, with large amounts allocated to the United States, Japan, and Germany. The report infers that TSMC "may not have the lowest nominal wages or static depreciation costs, especially after overseas expansion," and higher overseas cost structures may gradually dilute the historically very high returns on capital.

  Where does the money it earns go? Discipline is extremely strong. The report records that the main uses of cash are very clear: reinvestment in capacity expansion + higher dividends + maintaining a sound balance sheet. It has long relied mainly on cash dividends and does not conduct routine share repurchases. The 20-F explicitly states there were no repurchase operations, and there have been no large, high-premium acquisitions that destroy value. Although capex consumes large amounts of cash flow, with free cash flow often below net income, this is active investment for future capacity leadership rather than passive bleeding.

  The owner-earnings perspective. The report specifically notes that traditional free cash flow, about NT$1,002.6bn in 2025, understates TSMC because a substantial portion of massive capex is "growth investment" rather than investment required merely to maintain its competitive position. The report conservatively estimates maintenance capex at about NT$900bn, corresponding to "conservative owner earnings" of about NT$1.38tn in 2025, clearly higher than accounting free cash flow. This shows that its true cash-generating ability is stronger than it appears on the surface.

  The conclusion for this Baillie Gifford question: the unit economics are top-tier: nearly 60% gross margin, strong operating leverage, real cash profit, and extremely strong capital discipline. This is the solid basis for a high score on "unit economics and incremental returns," and where it is far superior to cash-burning growth stocks. But two caveats must remain: current margins contain a cyclical boom component, not a permanent normal state, and overseas expansion is a real dilution risk for marginal returns. "Bigger is better" holds in the upcycle, but is not an unconditional linear extrapolation.

  Jun 10, 2026
• QWhat conditions must hold simultaneously for it to rise 5x in 10 years? Are those conditions realistic? What expectations does today's share price imply?3/10

  A

  A 5x increase over 10 years, or about 17%/year total return, requires four optimistic conditions to hold simultaneously. The threshold is high and is not the base case. Today's share price of about NT$2,295 already implies optimistic expectations of long-term ultra-high AI growth + overseas expansion not diluting returns + high margins staying elevated for a long time + no material geopolitical shock. The price is already above the report's neutral valuation and even past the midpoint of the optimistic scenario, leaving insufficient margin of safety. This is TSMC's weakest link: a good company, but not a good price.

  First quantify the hurdle. A 5x return over 10 years corresponds to a compound total return of about 17%/year including dividends, clearly above the report's expected annualized return range for TSMC: conservative 0–4%/year, neutral 5–8%/year, and optimistic only 9–12%/year. In other words, the 17%/year "5x" target is not reached even by the report's optimistic scenario. That alone shows that buying at the current price, a 5x return over 10 years is not a realistic base-case expectation. It is a tail result that requires multiple layers of optimism.

  What conditions must hold simultaneously for a 5x return over 10 years? None can be missing:

  1. AI/HPC demand must sustain structurally ultra-high growth for at least several years and not fall back into a short-cycle order squeeze. Current profits are highly dependent on the AI boom. In 2026Q1, HPC already accounted for 61% of revenue, and the report warns that "if AI investment slows, margins and valuation may both fall back."
  2. 2nm/N2P/A16 and advanced packaging must enter volume production on schedule, with yields and customer adoption continuing to beat expectations. 2nm entered volume production in 2025Q4, but to support a 5x outcome, node leadership must continue translating into pricing power and share.
  3. Overseas expansion must not materially dilute overall ROIC. 2026 capex is as high as US$52bn–56bn, with large amounts going to the high-cost United States, Japan, and Germany. The report lists "overseas returns not being diluted" as a key assumption.
  4. Taiwan geopolitical and export-control risks must not produce a materially damaging event. The report calls this "the tail risk that must not be underestimated," and one that traditional valuation models cannot fully digest.

  Only if all four conditions hold simultaneously is a 5x outcome possible. If any one is discounted, "excellent compounding from an excellent company" falls back into "ordinary returns from an excellent company." The report's judgment is exactly that these conditions are realistic but not assured: conditional optimism, not a high-certainty base case.

  What expectations does today's share price imply? It has already priced in most of the optimistic narrative above. This is the core of the issue:

  • Using NT$2,235, the 2026-05-21 close, the report calculates TTM PE at about 29.5x and PB at about 9.6x. On owner earnings, it corresponds to about 42x conservative owner earnings. The market is "not treating TSMC as an ordinary manufacturer, but pricing it as a super-scarce AI infrastructure asset."
  • Valuation has been getting more expensive, not cheaper. By early June, the share price had risen to about NT$2,295, after setting an all-time high of NT$2,440 on June 3. Its U.S. ADR currently has TTM PE of about 34–36x and forward PE of about 28x, both materially above the level when the report was written. In other words, the report's conclusion of "insufficient margin of safety" only becomes more valid at the latest price.
  • The report's three intrinsic-value ranges are: conservative about NT$790, neutral about NT$1,230, and optimistic about NT$1,930. The current NT$2,295 is already above the midpoint of the optimistic scenario. The report states directly that the current price already implies expectations of "long-term ultra-high AI demand growth, overseas expansion not diluting returns, and abnormal profitability lasting for a long time." That has already begun to rely clearly on optimistic assumptions rather than margin of safety.

  The honest conclusion for this Baillie Gifford question: a 5x return over 10 years requires four optimistic conditions to hold simultaneously, so the hurdle is high and it is not the base case. Today's price has already pulled forward most of that optimistic narrative: the price is above the report's neutral and even optimistic midpoint, TTM PE is near historical highs and still rising. The conclusion matches the report: the company passes; the price does not yet pass. The report's ideal buying range is NT$1,100–1,400, leaving a 20–30% margin of safety. The current price is clearly in the waiting/premium zone, not the action zone. This is TSMC's lowest-scoring dimension under the Baillie Gifford framework. The issue is not that the company is weak; the issue is that the price has already pulled forward too much.

  Jun 10, 2026
• QWhy has the market not realized all this yet? Does it not understand, look down on it, or fail to look far enough? What will become the narrative inflection point?3/10

  A

  Unlike a typical "undervalued growth stock," the market understands and respects TSMC. Its excellence is already fully priced, and TTM PE is at a historical high. The real cognitive gap is not that the market fails to see the good, but whether the market over-believes that the good can last forever and underestimates geopolitical tail risk. Therefore, for TSMC, the narrative inflection point is not "value discovery." It is more likely "excessive optimism being disproved."

  This Baillie Gifford question usually assumes that "the market has not realized all this yet." For TSMC, that premise must be honestly reversed, because the reality is that the market has long realized it and may even have over-realized it.

  First, prove that the market does not look down on it and is not failing to look far enough. The objective evidence is valuation itself. The report records current TTM PE of about 29.5x and PB of about 9.6x, based on NT$2,235, and about 42x on owner earnings. The market is "not treating TSMC as an ordinary manufacturer, but pricing it as a super-scarce AI infrastructure asset." By early June, the share price had risen to about NT$2,295, after reaching an all-time high of NT$2,440 on June 3, and the U.S. ADR had TTM PE of about 34–36x, a 60%+ premium to its 10-year historical average. An asset trading in its most expensive historical range, far above the S&P 500 at about 21.5x, is by definition not being "looked down on" by the market. The market not only understands the moat; it has paid in advance for many years of future high growth.

  So where is the real "cognitive gap"? It has two opposite directions:

  The part where the market may be "overly optimistic," the more likely direction of imbalance. The report exposes the strongest bear case: the market is pricing in that, over the next 10 years, TSMC will still almost monopolize high-end AI chip manufacturing, advanced packaging will remain tight, margins will stay high for a long time, and geopolitics will not truly hit valuation. At least two variables in that narrative are easy for the market to systematically underestimate. First, current margins contain a cyclical boom component, and the report warns about whether investors are mistaking boom-period margins for a permanent normal. Second, Taiwan geopolitical tail risk cannot be digested by traditional valuation models, and the report calls it the risk that "must not be underestimated." The market is treating these two variables as negligible rather than potentially material.

  The part where the market may be "not looking far enough," which is smaller and not necessarily positive. If AI demand truly beats expectations for the long term and overseas returns are not diluted, then the current price may not be expensive after all. But the report is clear that reaching a value above its estimate "has already begun to rely clearly on optimistic assumptions rather than margin of safety." So even if there is upside cognitive gap from not looking far enough, it is built on optimistic assumptions, not on a safety cushion.

  What will become the narrative inflection point? For TSMC, the inflection point is two-way, and at the current price the elasticity of downward disconfirmation is larger:

  • Downward narrative inflection point, the one that deserves more caution: clear signs of structural slowing in AI/HPC capital spending; gross margin staying below about 53%–55% for a long time without a clear repair path; Samsung or Intel continuously taking share in advanced nodes / advanced packaging; key customers systematically diversifying high-end capacity or building their own alternatives; Taiwan-related political, military, or export-control risks materializing. If any one occurs, the market will rapidly re-rate from "priced for perfection," and high valuation will amplify the decline. The report warns that in an extreme case, "a temporary share-price decline of more than 50% is entirely possible."
  • Upward narrative inflection point, requiring optimistic delivery: 2nm/A16 volume production and yields continue to beat expectations, overseas fabs prove that returns have not been diluted, owner earnings growth in 2026–2028 stays above 15% without relying on more aggressive capital investment, and valuation falls closer to reasonable value. The report says that only when "valuation falls back while fundamentals remain strong" would its judgment of "insufficient margin of safety" need to be upgraded to positive.

  The honest conclusion for this Baillie Gifford question: TSMC is not a hidden gem the market cannot understand. It is an asset the market understands too well and may have paid too fully for. The cognitive gap is not "is it good," but "can the good last forever + is tail risk being underestimated." The narrative inflection point is more likely to come from excessive optimism being disproved than from value being discovered. This echoes the report's overall judgment: the company passes, the price does not yet pass. It should be a premium asset, but at the current point it is not a cheap entry.

  Jun 10, 2026