VanEck Semiconductor ETF (SMH) Deep Value Research

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

• QHow high is its market ceiling? Is it expanding an existing pie, or creating an entirely new market?6/10

  A

  The ceiling is very high, and both “expanding an existing pie” and “creating an entirely new market” are happening at the same time. But one point has to be clear first: SMH is not a company. Its “ceiling” is the ceiling of the semiconductor industry profit pool it tracks. This is critical. Otherwise, the upside imagination for an ETF gets mistaken for the growth curve of a single company.

  Start with the size of the pie itself. Semiconductors are a long-runway market with both existing demand and incremental demand: the report cites WSTS as forecasting that the global semiconductor market will grow from about 772.2 billion USD in 2025 to 975.5 billion USD in 2026; under SIA’s framework, global semiconductor sales reached a record 791.7 billion USD in 2025, up 25.6% year over year (the report’s figures are consistent with official SIA data), and SIA expects industry sales in 2026 to approach 1 trillion USD. On the equipment side, SEMI expects global semiconductor equipment sales to reach a record 156.0 billion USD in 2027 (about 133.0 billion USD in 2025 and about 145.0 billion USD in 2026). This is a trillion-scale pie still expanding at a double-digit annual pace. The ceiling is nowhere near exhausted.

  The structure matters even more: this cycle is not just about “selling more of the same chips” into an existing pie. In SIA’s data, logic product sales surged 39.9% in 2025 to 301.9 billion USD, becoming the largest category, while memory grew 34.8% to 223.1 billion USD. Behind that is AI compute turning a new class of products with unit prices several orders of magnitude higher (data center GPUs, custom ASICs, HBM high-bandwidth memory, advanced packaging, high-speed networking chips) into a market from scratch. The clearest example is NVIDIA, the largest weight in the basket: fiscal 2026 revenue was 215.9 billion USD, up 65% year over year. A market created almost out of thin air by AI and then scaled quickly to 200 billion USD has never appeared with this slope in the traditional semiconductor cycle.

  So for SMH as an ETF, the honest conclusion is: by holding leaders such as NVIDIA, Broadcom, TSMC, and Micron, it owns both ceilings at once: “expanding the existing pie” (PCs, smartphones, automotive electronics, industrial) and “creating new markets” (AI data center compute and its supporting stack). At the industry level, the ceiling is high and is still being actively lifted by AI. That is the basis for the report’s “industry attractiveness 4/5, long-term upward” judgment.

  But a Baillie-style note of sobriety has to be added immediately: a high ceiling does not mean buying today lets you capture it. The report repeatedly stresses that SMH is currently a “good asset, bad price.” The industry pie can grow, and whether you can earn a satisfactory return at today’s roughly 585 USD price (NAV was 598.11 USD as of 2026-06-08, AUM had risen to about 65.0–68.0 billion USD, already far above the 40.984 billion USD AUM and 543.96 USD price at the report’s cutoff) are two separate questions. The ceiling answers whether the runway can still grow. The answer is yes. It does not answer whether the odds are attractive. That belongs to the valuation question.

  Jun 10, 2026
• QCan its revenue at least double over the next five years? Is growth mainly driven by volume, price, or new businesses?5/10

  A

  The question of “revenue doubling in five years” has to be translated for an index ETF. SMH has no “revenue” of its own; its return comes from growth in the underlying companies’ earnings per share plus valuation changes. In Baillie language, the real question is: can the underlying earnings of the companies in the basket double over the next five years? The answer is: very likely for the top weights, but probably not for the whole basket, and the driver is mainly volume plus new businesses, not price increases.

  First set the basis correctly. The report states clearly that SMH’s “business model” is simply to charge a 0.35% management fee to replicate an index; “what you as a holder truly bear and enjoy is the operating results of the underlying semiconductor companies.” So the real question is whether the underlying profit pool can double in five years.

  The conclusion differs sharply across two layers:

  The top AI engines are likely to double or more, driven by volume and new businesses. NVIDIA had fiscal 2026 revenue of 215.9 billion USD, up 65%. At that slope, if AI capex does not collapse, it could double again within two or three years; Broadcom had FY2025 revenue of 63.9 billion USD, with AI revenue up 65% year over year to about 20.0 billion USD, and custom ASICs are entirely incremental; AMD had FY2025 revenue of 34.6 billion USD, up 34%, with data center as the main engine. Growth at this layer comes almost entirely from volume (AI accelerator shipments, data center buildout) and new businesses (ASICs, HBM, networking), not simple price hikes. That is exactly the higher-quality growth Baillie values most.

  Mature and cyclical weights are the drag, and they cannot double. TSMC has already reached 3.81 trillion TWD in 2025; doubling in five years would require about 15% annualized growth, a demanding task from a high base. Mature analog/communications leaders such as TI and Qualcomm are “excellent but not fast” assets with mid-single-digit to low-double-digit growth. Micron is a highly cyclical memory stock, and Intel had FY2025 revenue of 53.1 billion USD while still posting a GAAP net loss of about 2.0 billion USD. Their “doubling” would be more cyclical recovery than structural growth.

  Weight the two layers together, and SMH’s overall underlying earnings doubling in five years is possible but not easy, with a heavy bet on one variable: AI capex. Note that weights have already drifted: according to stockanalysis data as of 2026-06-06, Micron had risen to the third-largest weight (about 7.28%), while Broadcom had moved down to sixth (about 6.53%). That means the weight of highly cyclical memory has increased, diluting the purity of a “pure AI doubling” story. The report’s “key assumption” is exactly that “AI-related capex will not collapse significantly over the next 2–3 years.”

  The honest landing point: this is a basket of assets that can deliver real volume-driven growth, with five-year doubling potential that objectively exists and is driven by healthy volume plus new businesses rather than squeeze-the-tube price increases. But it is a passive basket. It cannot actively add to the fastest engines or remove the slowest drags, so its overall growth rate will be blunted by mature and cyclical stocks. The doubling story belongs more to NVIDIA/Broadcom and a few others than to SMH as an ETF.

  Jun 10, 2026
• QFive years from now, what will take over as the next growth engine? Does this “second curve” exist today?5/10

  A

  For SMH as a passive ETF, the “second curve” is a special question: the fund shell itself does not incubate new businesses, but its index rules ensure it can automatically board the industry’s next growth engine. Today’s first curve is AI data center compute. The next curves are already faintly visible, but most are extensions of the same leading companies rather than entirely new players.

  First, clarify the special nature of an ETF. A single company’s second curve depends on management actively incubating it. SMH is a basket that “holds U.S.-listed semiconductor leaders according to rules,” with reconstitution every March and September and quarterly rebalancing (the report cites the MarketVector methodology). This means: as long as a new curve is driven by leaders already in the basket, SMH will automatically follow it; but if the disruptor is a new company not yet included in the index, SMH will be late. That is the double-edged sword of a passive structure.

  Does the industry-level second curve exist today? Yes. There are mainly three, and all already have visible early forms in the underlying companies’ financials:

  First, advanced packaging and HBM high-bandwidth memory. The report repeatedly identifies “advanced packaging and high-bandwidth memory” as current expansion engines. As Moore’s Law process scaling gets more expensive and slower, stacking or placing multiple chips side by side in packages (CoWoS, HBM) becomes the main battlefield for continuing compute growth. This curve is carried by TSMC (packaging capacity) and Micron (HBM), both of which are in the basket. Micron’s weight has risen to about 7.28%, so in a sense the basket is already “automatically adding” to this curve.

  Second, custom ASICs and AI networking. As hyperscale cloud companies are no longer content to buy only general-purpose GPUs and instead develop custom accelerators, Broadcom has become the biggest beneficiary. In FY2025, its AI revenue rose 65% year over year to about 20.0 billion USD. This is both a “branch” beyond NVIDIA’s general-purpose GPUs and a second engine inside the basket.

  Third, AI inference and edge/device AI. Current compute demand is concentrated in training. In the next phase, as inference and on-device AI scale, AMD, Qualcomm, analog, and power semiconductors will see new demand. This curve is still earlier-stage. The report also notes that “discrete devices were still dragged down by weak automotive demand in 2025,” indicating that terminal recovery has not yet arrived. This is a curve that exists today but has not yet been realized.

  The Baillie-style honest landing point: the second curve objectively exists and can already be seen in the revenue structure of underlying companies today, but there are two concerns. First, most of these new curves are extensions of the same leaders (NVIDIA, TSMC, Broadcom, Micron), not independent new growth poles, so they are highly correlated with the first curve. Once the master switch of AI capex cools, several curves will slow together, rather than offsetting one another. Second, as a passive basket, SMH cannot actively overweight potential disruptors before they become index weights. In the report’s “signals that would trigger reassessment,” it lists “industry growth shifting from AI-driven to broad-based downward revision.” That is precisely the concern that the first curve and these not-yet-mature second curves may ebb at the same time.

  In short: the industry’s next baton exists and is visible, but it looks more like a deeper extension of the same AI wave than an independent insurance policy decoupled from the current cycle.

  Jun 10, 2026
• QWhat is its core competitive advantage? Will this moat widen or narrow over the next three to five years?6/10

  A

  The moat has to be viewed on two layers, and the report gets this judgment right: SMH’s ETF shell has almost no moat, while the underlying leaders it holds have very deep moats. Whether this “underlying moat” widens or narrows over the next three to five years depends almost entirely on one variable: AI capex.

  Start with the ETF shell. It basically has no moat. The report says directly that “SMH itself does not have a very wide operating moat.” Its advantages are only scale, liquidity, brand, and a transparent 0.35% fee; “any large ETF institution could theoretically create a similar product.” In other words, the moat you buy with SMH is not at the fund company level. It sits in the companies inside the basket. That premise cannot be blurred when answering this question.

  Look through to the underlying holdings, and the moats are real and extremely deep:

  • NVIDIA: CUDA software/hardware ecosystem plus system-level integration. Fiscal 2026 revenue was 215.9 billion USD, with GAAP gross margin of 71.1%. A 71% gross margin is hard evidence of pricing power and ecosystem lock-in. The report’s phrase that it “sells an AI computing platform deeply depended on by customers” is on target.
  • TSMC: a compound barrier of process technology, yield, capacity, and customer ecosystem. 2025 gross margin was 59.9%, and net income was 1.72 trillion TWD, while it continues to pour massive capex into 2/3 nanometer nodes and advanced packaging. Leading-edge process is a winner-takes-most field where replication costs are measured in tens of billions of USD and years.
  • ASML: de facto exclusivity in EUV lithography machines. The report says this equipment moat “cannot be replicated with just a few years and several billion USD.” It is one of the hardest single-point monopolies in the entire industry to challenge.
  • Broadcom: customer embedding in custom ASICs and networking chips. FY2025 free cash flow was 26.9 billion USD, up 39% year over year (the report’s 26.914 billion USD is consistent), and this strong cash machine validates its pricing power.

  But the report also correctly identifies the ETF’s structural flaw: it packages strong moats and weak moats together. Intel had FY2025 revenue of 53.1 billion USD but still recorded a GAAP net loss of about 2.0 billion USD and remains in a moat-repair phase. Micron’s memory industry has naturally shallower and more cyclical moats. So the basket’s overall moat is “leader-dominated plus mixed with cyclical/repair assets,” passively diluted. The report’s “moat strength score of 3/5” is a reasonable score for this mixture.

  Will it widen or narrow over the next three to five years? It depends on the scenario. This is the marginal change Baillie asks for, not a static metric:

  • If AI infrastructure spending stays high: NVIDIA’s ecosystem, TSMC’s leading-edge process/packaging, Broadcom’s ASICs, and ASML’s EUV moats will continue to widen, because high capex will further raise entry barriers for latecomers and reinforce leaders’ scale and ecosystem advantages.
  • If AI capex normalizes: the report is sober: “leaders will not disappear, but marginal pricing power and valuation support will contract.” In other words, the moat’s depth remains, but its width (the pricing-power premium) narrows.

  SMH’s distinctive moat risk is the passive structure itself: the report points out that “the biggest problem is precisely that it cannot actively remove companies whose moats are narrowing; it can only wait for index rules to handle them.” Put differently, you benefit when a leading company’s moat widens, but when a constituent’s moat collapses, you can only passively wait until the next reconstitution removes it.

  Landing point: the underlying moats are deep and will probably continue to widen over the next three to five years as AI investment continues. But the “holding method” for those moats is a passive basket. You enjoy the leaders widening their moats while being forced to absorb weaker companies narrowing theirs. That is why the report scores the moat at 3/5 rather than higher, and it is the price SMH pays versus directly holding one or two of the strongest leaders.

  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?4/10

  A

  This question has to be answered on two levels for SMH, and the answers are almost opposite: as a fund shell, SMH has no “DNA for self-reinvention” at all. It is a set of rules, not an organization that reflects. But through periodic index reconstitution, it mechanically gains a kind of “blood-replacement passive reinvention,” at the cost of sluggishness. As for “how it handles mistakes and bad news,” the meaningful subjects are the underlying companies, and they differ greatly.

  First, make the implicit premise of this chain question explicit: Baillie asks this to understand whether, when the core business is disrupted, the company has the DNA to tear itself down and start again, and whether it faces bad news by covering it up or correcting it directly. Applied to an ETF, this has to be split into two entirely different subjects.

  First layer, the ETF shell: no subjective reinvention DNA, only rule-driven passive replacement. SMH does not “transform through R&D.” It only follows the rules of reconstitution every March and September and quarterly rebalancing, removing companies that no longer meet the threshold of “at least 50% of revenue from semiconductors (existing constituents can be relaxed to 25%)” and adding new leaders. The benefit is that if one technology path is disrupted, an old incumbent declines, and a new leader rises, SMH will eventually switch onto the new winner. That is a form of passive reinvention that “automatically follows industry evolution.” The drawback is exactly what the report repeatedly stresses: it “does not time the market itself, nor does it actively improve portfolio quality,” and it “cannot actively remove companies whose moats are narrowing; it can only wait for index rules to handle them.” So this reinvention is delayed and mechanical, not the active founder-led second founding Baillie looks for. If the core narrative (AI compute) is disrupted, SMH cannot turn ahead of time. It can only rebalance passively after the rules are triggered.

  On “how it handles mistakes and bad news,” the ETF shell is actually strong here. The report describes VanEck at this layer as “rules-based, transparent, clear on fees, and clear in tracking,” with official disclosure of daily holdings and premium/discount history. In other words, the fund has no room to “hide bad news”: all holdings, NAV, premiums, and discounts are public every day, and mistakes such as tracking error have nowhere to hide. This is the honesty advantage of a passive instrument. But on the other side, it also does not “correct mistakes,” because it makes no active judgments and has no concept of admitting error.

  Second layer, the underlying companies: this is where “DNA for self-reinvention” should truly be examined, and they are highly divergent:

  • Intel is both a negative and positive case study at once: its leading position in core process technology has been disrupted by TSMC, it still had a net loss of about 2.0 billion USD in FY2025, and it is in a painful reinvention phase (IDM 2.0, foundry transition). Whether it has reinvention DNA and can succeed remains unresolved. The report lists it as “in a repair phase, dragged by heavy assets.”
  • AMD is the model of successful reinvention: from a CPU maker near the edge, it rose again through the Zen architecture and data center strategy, with FY2025 revenue of 34.6 billion USD, up 34% year over year, and GAAP net income of about 4.3 billion USD. This is evidence of reinvention DNA.
  • NVIDIA has repeatedly moved from gaming graphics cards to accelerated computing and then to AI platforms. It is a classic case of continuous self-reinvention.

  Baillie-style landing point: asking whether SMH has “DNA for self-reinvention” as if it were a single company does not work. It is a rule set. Rules do not start companies; they only replace components, and slowly. It is highly transparent with bad news (it does not cover things up) but does not actively correct mistakes either (there is no notion of admitting error). The entities that truly do or do not have reinvention DNA are the underlying companies. The price of the ETF is exactly this: you neither fully capture the concentrated upside of AMD-style successful reinvention nor actively avoid the drag of Intel-style failed reinvention. You simply bear both according to their weights. That is the core reason the report scores the fund layer’s “management and capital allocation” at 3/5 and stresses that it is “not an exceptional active capital allocator.”

  Jun 10, 2026
• QDoes management, especially the founder, have a long-term view and deep alignment with the company? Is it willing to sacrifice current profits for the next five to ten years?3/10

  A

  The direct conclusion: SMH has no “founder,” so one of Baillie’s core questions mostly collapses when applied to this ETF. Fund manager VanEck is a diligent rule executor, not an “entrepreneur” whose interests are deeply aligned with yours and who is willing to sacrifice the present for ten years out. As for founder/management alignment at the underlying companies, it is highly mixed, and the ETF does not let you choose.

  First, clarify the spirit of this Baillie question: Baillie LTGG strongly prefers companies where the founder is still at the helm, personal wealth is deeply tied to the company, and leadership is willing to sacrifice current-period profits for a five-to-ten-year vision (the archetype is the Jensen Huang or Elon Musk style of long-term operator). The reason is that only this kind of person is likely to withstand Wall Street’s quarterly pressure and invest through long cycles. Apply that ruler to SMH, and the result is clear:

  The ETF shell structurally fails this question. VanEck is not a founder, and the fund manager does not hold a large equity stake in the “company” and share risk with you. The report’s assessment is accurate: at the fund level, this is “qualified rule execution” that is “rules-based, transparent, and clear on fees,” but it is “not a capital allocator that concentrates purchases when undervalued and reduces exposure when overvalued,” and it is “not exceptional active capital allocation.” More importantly, the design goal of a passive index fund is precisely not to make subjective judgments and not to sacrifice the present for a long-term vision. It merely replicates the index according to rules and charges a 0.35% fee. So the question “does management have a long-term view and willingness to sacrifice current profit for ten years out” is simply not applicable to the fund shell. It has neither long-term vision nor sacrifice. It only replicates. The report’s “management and capital allocation 3/5” score essentially says: competent execution, not exceptional foresight.

  The underlying companies are where founder/management alignment truly exists, but it is highly divergent, and you cannot select:

  • Models of excellent capital allocation and long-term shareholder returns: the report notes that ADI returned about 96% of free cash flow to shareholders in FY2025, TI “returned 6.5 billion USD to shareholders over the past 12 months” and has long reduced share count, and Broadcom “continued to raise dividends and generate strong free cash flow.” These are positive signs of rational capital allocation.
  • Models of long-term investment: NVIDIA and TSMC have continued to invest in R&D and capacity through cycles for years (TSMC keeps investing in 2/3 nanometer and advanced packaging), which is typical “sacrificing the current period for the long term.” Strictly speaking, though, most of them are now professional management systems rather than founder-controlled businesses.
  • Repair-phase cases where alignment and foresight remain to be proven: Intel remains in a stage of “high capex, with earnings repair not yet complete.”

  Baillie-style honest landing point: this question exposes the fundamental tension between an ETF and the Baillie framework. Baillie is looking for “a leader with vision, aligned interests, and willingness to sacrifice the short term for the long term.” SMH gives you “a rule set that does not make judgments plus a basket of companies with mixed management quality.” The report says it precisely: “you cannot own only the former (excellent management teams such as ADI, TI, Broadcom) without also owning the latter (Intel’s repair-phase management).”

  So the answer is: the fund manager is diligent but has no real vision, and its interests are not deeply aligned with yours; several underlying companies do have exemplary management, but the passive basket takes away your right to bet only on the best leaders. If what you truly value is Baillie-style founder long-termism, the purer approach is to directly hold the few leaders whose management you trust most (which is exactly what the report suggests: “the most relevant peer alternative may instead be directly holding a very small number of the strongest leaders, such as TSMC or NVIDIA”), rather than buying an ETF that packages good and bad management together.

  Jun 10, 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth model sustainable and not dependent on harming society or regulation?6/10

  A

  This question has to be split into two layers: “indispensability” and “whether the growth model is sustainable and not dependent on harming society or regulation.” First layer: SMH the ETF itself is highly replaceable (if it disappeared, investors could switch to another fund), but the underlying leaders it holds are vital arteries the global digital economy cannot do without. Second layer: the underlying companies’ growth model is broadly compliant and creates positive externalities, but there is one unavoidable concern: geopolitics and export controls, a structural risk the report repeatedly flags.

  Start with indispensability. There are two layers, with opposite answers:

  The ETF shell is almost completely replaceable. If SMH disappeared tomorrow, investors would not “miss” it, because similar semiconductor ETFs such as SOXX are readily available substitutes. The report is blunt: “any large ETF institution could theoretically create a similar product.” The fund’s value is the convenience of liquidity and low fees, not indispensability.

  But look through to the underlying holdings, and indispensability is extremely high. If TSMC disappeared tomorrow, the world’s most advanced AI chips and high-end smartphone chips would have almost nowhere to be manufactured, and the entire digital economy would seize up. Its customers would “miss” it at a catastrophic level. The report discloses that TSMC’s top 10 customers account for about 78% of revenue, and its largest customer about 19%; that deep binding itself is evidence of indispensability. If NVIDIA disappeared, global AI training compute would face a break in supply. Its fiscal 2026 revenue of 215.9 billion USD and gross margin of 71.1% are hard indicators that “customers cannot do without it and are willing to pay a high price.” ASML’s EUV lithography machines are also one of a kind. So the honest answer is: when you buy SMH, you buy a basket of vital companies whose disappearance would shake the world, but the ETF shell carrying them is itself worthless in terms of irreplaceability.

  Now address sustainability and the society/regulation layer. This is the implicit premise that must be added, and it is SMH’s biggest non-valuation risk:

  On the positive side, semiconductor growth is broadly healthy and creates positive externalities: it powers AI, cloud computing, automotive electronics, and industrial automation, raising productivity across society. It does not monetize by harming consumers or exploiting regulatory loopholes (unlike some models based on data abuse or addictive design). The report’s judgment that “the industry itself remains in a long-term growth phase” holds.

  But there is one unavoidable structural concern: geopolitics and export controls. In the risk section and tracking indicators, the report repeatedly names “trade restrictions” and “changes in U.S. export controls, tariffs, and subsidy policies,” and cites TSMC’s 20-F statement that “trade restrictions ... could affect profitability.” Leading-edge semiconductor processes and advanced packaging are highly concentrated in a few regions and companies, exposing the whole basket to U.S.-China technology rivalry, export control lists, tariffs, and localization subsidies. This is not endogenous unsustainability of the “company harms society” kind. It is exogenous unsustainability where “the global supply chain and market access that growth depends on could be cut apart by politics.” If core holdings’ business models are impaired by controls, the report lists this as one of the “signals that would trigger reassessment.”

  Baillie-style landing point: the underlying leaders’ indispensability is textbook-level (their disappearance would shake the world), and their growth model itself is compliant and socially beneficial; the ETF shell has no indispensability at all. The real sustainability question is not “does it harm society,” but “can geopolitics allow this global supply chain to keep functioning smoothly.” That is a systemic risk SMH cannot diversify away and can only passively absorb.

  Jun 10, 2026
• QWhat are the unit economics of this business (gross margin, incremental returns)? Do they improve or deteriorate with scale? Where does the money earned go?6/10

  A

  Unit economics also have to be split into two layers: the ETF shell’s unit economics are extremely simple and strong (0.35% fee, lower marginal cost as scale grows); look through to the underlying holdings, and the picture is a mix of “high-margin asset-light design companies + high-capex asset-heavy manufacturers + cyclical memory makers.” Overall gross margins and incremental returns are good, but they are visibly dragged down by heavy-capital segments, and most of the money earned is not paid to you. It is reinvested into capacity, R&D, and buybacks.

  First layer, the fund shell’s unit economics: simple and scalable. SMH charges a 0.35% management fee, and the marginal cost of operating an index fund declines as scale increases. The report disclosed that AUM had grown from 40.984 billion at the cutoff (according to Yahoo Finance, about 67.9 billion USD as of 2026-06-08). The larger the scale, the lower the operating cost per unit. This is the ETF shell’s only meaningful “scale improves economics” logic. But this layer’s profit belongs to VanEck, not to holders. For you, it is simply a cost item.

  Second layer, the underlying companies’ unit economics: this is what you truly own, and it is highly divergent:

  Asset-light, high-margin design/platform companies whose economics improve with scale:

  • NVIDIA: fiscal 2026 gross margin was 71.1%, with operating cash flow of 102.7 billion USD and free cash flow of about 96.7 billion USD. A 71% gross margin plus very high cash conversion is a model case of unit economics improving as scale expands, with very high marginal profit on each additional GPU.
  • Broadcom: FY2025 free cash flow was about 26.9 billion USD, up 39% year over year. It is a strong cash machine with excellent incremental returns.

  High-margin but extremely capital-heavy manufacturing leaders whose incremental returns are consumed by capex:

  • TSMC: 2025 gross margin was 59.9%, and operating cash flow was 2.27 trillion TWD. Gross margin is impressive, but the report correctly notes “strong cash, but extremely large capex.” To maintain leadership in advanced processes, every round of growth requires massive capacity investment upfront, materially diluting free cash flow. As scale grows, the capex threshold also rises.

  Cyclical memory makers whose unit economics swing violently with the cycle:

  • Micron: the report says its “net capital expenditures still reached 13.8 billion USD,” and that the memory industry “can quickly move from high profit to loss in a downcycle.” Unit economics do not improve monotonically with scale; they swing up and down with the cycle.

  Repair-phase companies whose unit economics have not yet turned positive:

  • Intel: FY2025 still had a GAAP net loss of about 2.0 billion USD. Under heavy-asset drag, incremental returns are negative.

  “Where does the money earned go?” This is the decisive point in Baillie’s question and also the root of why SMH’s valuation is expensive. The cash generated by semiconductor leaders is mostly not handed to you through dividends (SMH’s 30-day SEC yield is only 0.29%; the report calls it “extremely low cash yield” and “not an income asset”). It mainly flows to three places: ① capex for capacity expansion and advanced process/packaging (TSMC, Micron); ② R&D (to maintain AI and process leadership); ③ buybacks and dividends (large shareholder returns by mature leaders such as TI, ADI, Broadcom). The report’s characterization is precise: SMH “can create a large amount of real cash flow over the long term, but a large portion of that cash flow continues to be used for capacity expansion, R&D, buybacks ... rather than being acquired by you at a low valuation.”

  One detail in the report needs correction: the report says AMD FY2025 free cash flow was 5.519 billion USD, but AMD officially disclosed FY2025 free cash flow of about 2.1 billion USD (GAAP net income about 4.3 billion USD). The report clearly overstates this, and AMD’s cash conversion quality is weaker than the report describes. That also supports the judgment that underlying cash-flow quality is highly mixed.

  Landing point: overall unit economics are a weighted average of “top students (NVIDIA, Broadcom) + heavy-capital giant (TSMC) + cyclical stock (Micron) + repair case (Intel).” Gross margins and incremental returns are objectively good but diluted. Scale improves economics for design companies; for manufacturers, it is “better, but only after heavy spending”; for cyclicals, it depends on the cycle. Most of the money earned is reinvested into the industry itself rather than distributed to shareholders. That is the essence of SMH as a “high-quality growth asset pool” rather than a “cash cow,” and it is also the micro reason it lacks a margin of safety at the current price.

  Jun 10, 2026
• QWhat conditions must all hold for it to rise fivefold in ten years? Are those conditions realistic? What expectations are embedded in today’s stock price?3/10

  A

  For SMH to rise fivefold over ten years (about 17.5% annualized), three things must all hold: ① AI-driven high growth must continue for many more years, with underlying earnings compounding at a high rate; ② leaders’ margins and pricing power must not fall meaningfully; ③ the market must continue to assign a reasonably high valuation multiple, with no multiple compression. Frankly, the probability of all three holding at once is not high, and today’s roughly 585 USD price has already prepaid a considerable portion of the optimism. That is the core of the report’s “current price lacks a margin of safety” conclusion.

  First, quantify what “fivefold in ten years” requires. Baillie wants a condition list, not a slogan:

  Fivefold in ten years = about 17.5% annualized. SMH’s total return ≈ underlying earnings growth + valuation multiple change + minimal dividend (the 30-day SEC yield is only 0.29%, essentially negligible). So to achieve 17.5% annualized, the following all have to be true:

  1. Underlying earnings compound at a high rate for a long time. If the valuation multiple is unchanged ten years from now, underlying EPS has to compound at roughly 17% annualized. That means asking a basket already at trillion-USD scale and containing mature and cyclical stocks such as TSMC/TI/Micron/Intel to maintain something close to today’s AI peak growth for ten years without fading. The report’s “key assumption” is exactly that “AI-related capex will not collapse significantly over the next 2–3 years,” but it only dares to assume 2–3 years. A fivefold return needs nearly ten years.
  2. Margins do not fall. NVIDIA’s 71.1% gross margin and TSMC’s 59.9% gross margin must broadly hold. The report warns that if AI capex normalizes, “marginal pricing power and valuation support will contract.” Historically, semiconductor gross margins have never stayed at cyclical peaks for a long time.
  3. Multiples are not compressed. This is the most fragile link, discussed below.

  What expectations are embedded in today’s price? This is the core of the question, and it must be stated with current data:

  The official portfolio P/E cited at the report cutoff was 38.56x. But after the price rose from 543.96 to about 585, stockanalysis showed the portfolio P/E had risen to about 51x as of 2026-06-06. Valuation has not fallen; it has risen, meaning today’s embedded expectations are more optimistic and more fragile than at the report cutoff.

  But there is an honest counterpoint that must be stated, otherwise this would mislead: the high trailing P/E is partly a “point-in-time distortion” because AI earnings have not yet fully entered the TTM denominator. Take top-weight NVIDIA: its TTM P/E was only about 32x as of 2026-06-09, about 40% below its ten-year median of 52.85x, because earnings growth has caught up with the share price. The whole semiconductor sector’s forward P/E is about 22x, far below the basket’s roughly 51x trailing PE. In other words: on trailing 51x, SMH looks frighteningly expensive; on forward 22x, it is merely “reasonably high pricing for a high-growth industry.” The truth lies between the two. It is priced for “high growth continuing to materialize.” As long as earnings keep growing rapidly, today’s high trailing multiple can be “digested” by future earnings.

  So is fivefold realistic? Baillie-style honest verdict:

  • Possible, but it requires nearly perfect consecutive outcomes. The report’s optimistic scenario shows that “only under a near-optimistic case would annualized returns over the next ten years reach roughly 10%–12%.” Note that the report’s own optimistic upper bound is only 10%–12% annualized, corresponding to about 2.6–3.1x over ten years, still short of fivefold. Fivefold (17.5%) requires assumptions even more optimistic than the report’s optimistic scenario to all hold.
  • The more likely neutral outcome is 5%–7% annualized (about 1.6–2x), with the conservative case even at 0%–2%. The report explicitly gives 5%–7% for the neutral scenario.
  • The current price has already prepaid a large portion of the optimistic case. The report judges the current price to be at a “significant premium to conservative intrinsic value and a clear premium to reasonable value.” Its ideal buy range is 250–380 USD, while the current price of about 585 USD is far above that.

  Landing point: a fivefold ten-year return is not impossible, but it requires AI prosperity, high margins, and high valuation multiples to remain unbroken for ten years. That is a low-probability combination of “rolling three good outcomes in a row.” At today’s roughly 585 USD price, under the report’s framework, you have already paid a high price for a “near-perfect AI future,” leaving almost no margin of safety. The report’s most precise sentence is: “this is a classic case where the business is not wrong; the odds are wrong.” To bet on fivefold, you are not betting on whether the companies will grow. You are betting on whether the market will still pay a high price when growth slows, and historically that is the least reliable link.

  Jun 10, 2026
• QWhy has the market not realized all this yet? Is it because investors do not understand it, look down on it, or cannot look far enough? What will become the “narrative inflection point”?3/10

  A

  For SMH, this question has to be turned around. Baillie’s original intent is: “why has the market not yet recognized how good this stock is (does not understand it / looks down on it / cannot look far enough), creating undiscovered upside?” SMH is the opposite: the market has already fully recognized, even over-recognized, its merits (NAV return of about 140% over the past year). So there is no “underpriced perception gap” here. The real perception gap is on the other side: the market may be underestimating the drawdown that high valuations can suffer when growth slows. The “narrative inflection point” is therefore not a positive catalyst being recognized, but the negative trigger of AI capex cooling.

  First answer “why has the market not realized it” directly: it already has. That is exactly the problem.

  The report’s data is very clear: SMH had about 140.37% NAV return over the past 1 year, about 33.87% annualized over the past 5 years, and about 35.70% annualized over the past 10 years, and the portfolio valuation rose further from the report-cutoff P/E of 38.56x to about 51x on stockanalysis as of 2026-06-06. This combination of “excellent history + elevated valuation” is a classic sign of the market fully pricing it, even enthusiastically pricing it. It is not an obscure stock that investors “do not understand, look down on, or cannot look far enough.” It is a star asset under a hot spotlight. The report’s line is sharp: “excellent past performance often comes with valuation expansion” and “historical returns are very strong, easily inducing momentum chasing.”

  So for SMH, Baillie’s three reasons for what “the market cannot see” basically do not apply:

  • Do investors not understand it? No. Semiconductors are among the most intensively researched sectors today, with very full institutional coverage.
  • Do they look down on it? No. It is an object of capital pursuit, not a neglected orphan.
  • Can they not look far enough? Partly the opposite. The market may be looking “too far and too optimistically,” already pricing in the best outcomes for the next 5–10 years. The report’s counterargument is exactly: “you see the halo of semiconductor leaders; the market sees an almost perfect future.”

  Where is the real perception gap? In downside asymmetry. That is what is worth discussing as the thing the market may not have fully recognized for SMH:

  The market has fully priced the upside (long-term AI growth), but it may be systematically underestimating the damage from the combination of “earnings revision + multiple compression” when high-valuation assets deflate. The report gives a quantified warning: from the current price, if the AI investment cycle cools, leading companies’ growth rates fall, and valuations compress from high levels to more common ranges, “a 50%–65% mark-to-market drawdown in SMH during a mid-cycle decline would not be exaggerated.” That is the real perception gap easily ignored by bull-market sentiment: not “upside has not been discovered,” but “the magnitude of downside is underestimated.”

  Here, a counterpoint also needs to be added honestly: not all signals point to “excessive optimism.” As noted above, top-weight NVIDIA’s TTM P/E has fallen to about 32x, about 40% below its ten-year median, and the sector’s forward P/E is about 22x. This shows earnings are genuinely catching up with prices, and it is not a pure valuation bubble. So the more precise statement is: there is no “market does not understand it” perception gap for you to arbitrage, but by buying at a high trailing multiple, you bear odds risk: if growth falls short, both multiples and earnings can be hit at the same time.

  “What will become the narrative inflection point?” This is the implicit premise that must be added, and almost all inflection points are on the negative side:

  Because the upside narrative is already fully priced, the inflection points that can materially change SMH’s path are listed comprehensively in the report’s “signals that would trigger reassessment.” The core items are:

  1. AI capex cooling: a slowdown in hyperscale cloud capex growth or advanced packaging supply-demand returning to balance would be the biggest inflection point. The report lists “changes in AI investment pace” as the second-largest risk.
  2. Deterioration in leaders’ cash-flow conversion: NVIDIA, TSMC, and Broadcom showing “cash flow materially weaker than profit” for several consecutive quarters.
  3. Industry growth shifting from AI-driven to broad-based downward revision (WSTS/SIA data turning lower).
  4. Geopolitics/export controls impairing the business models of core holdings.
  5. Multiple compression: when “portfolio valuation remains high, but profit growth clearly falls to mid-to-low single digits,” the Davis double hit begins.

  Baillie-style landing point: SMH is not an undervalued opportunity where “the market has not yet recognized its value.” Its merits have been fully, even excessively, priced; the 140% return over the past year is the evidence. Its real perception gap is that the market may be underestimating downside asymmetry in high-valuation assets. The narrative inflection point will not be the unveiling of some positive news, but a cooling signal from the master switch of AI capex. The report’s final recommendation is therefore restrained and honest: “a good asset and a good entry point are not the same thing.” For SMH, understanding the story is easy. The hard part is admitting that the story already has a high price tag.

  Jun 10, 2026