Quanta (2382.TW) Zen Horizon Research Report

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 market, or creating an entirely new one?5/10

  A

  The absolute height of the ceiling is substantial, but Quanta is expanding an existing market rather than creating a new one, and it receives the thinnest slice of that market's profit pool.

  Start with the size of the market itself. Quanta's core battleground is AI data-center rack-scale ODM assembly. Demand in this segment is defined by NVIDIA's rack shipments: Morgan Stanley estimates annual shipments of GB200/GB300 NVL72 racks will rise from about 29,000 racks in 2025 to 70,000~80,000 racks in 2026, while value per rack keeps rising by generation. A GB300 rack is about USD 3.99 million, while a Rubin VR200 rack is estimated at about USD 7.8 million, nearly double. So viewed through absolute revenue capacity, the ceiling is indeed moving higher. Quanta's FY2025 revenue has already exceeded TWD 2.12 trillion, up 50.5% YoY, and 2026Q1 revenue reached TWD 809.2 billion, up 66.6% YoY. The trajectory is still upward.

  But is this creating a new market? No. AI data centers and hyperscale servers already existed. Rack-scale deployment is an evolution in which NVIDIA upgrades the selling unit from boards to full racks. The market is created and defined by NVIDIA through GPUs. Quanta is the outsourced manufacturer serving an existing large market, doing the work of assembling products invented by others at scale. The report states this plainly: Quanta assembles products defined by the world's most powerful supplier, NVIDIA, for the world's richest customers. It does not have the ability to define the market, nor does it own the source of demand.

  The more important issue is the quality of the ceiling: Quanta receives the thinnest layer of the profit pool. NVIDIA GPU gross margin is about 70%, while Quanta's quarterly gross margin for rack assembly has fallen to 4.78%, the lowest in nearly fifteen quarters. In other words, the revenue ceiling is high, but the profit ceiling is tightly compressed by upstream NVIDIA and downstream cloud customers. The larger the market gets, the thinner the layer of cream Quanta can reach. From Baillie Gifford's market-ceiling perspective, the truly valuable company is the one that can create and dominate a new market while capturing most of the value. That company is NVIDIA. Quanta is the carrier in this story: enormous in scale, weak in pricing power.

  In one sentence: the height of the ceiling is real and attractive, but Quanta is growing its share in a large existing market created by others, earning only an assembly spread rather than opening a new frontier. Its growth is follow-on volume expansion, not market-defining expansion.

  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?5/10

  A

  Very likely, and it may well more than double. Growth is mainly driven by a dual engine of price, as value per rack rises by generation, and volume, as rack shipments double. It has little reliance on a true new business.

  Start with momentum. Quanta has already nearly doubled over the past two years: FY2025 revenue was TWD 2.1236 trillion, up 50.5% YoY; 2026Q1 quarterly revenue was TWD 809.2 billion, up 66.6% YoY; and TTM revenue was about TWD 2.447 trillion. The report states that AI server order visibility has extended into 2027, with monthly revenue maintaining 60%+ YoY growth. From this starting point, another doubling over the next five years, roughly a 15% five-year CAGR, is not an aggressive assumption. It is closer to conservative. If the AI capex cycle does not reverse, the probability of achieving it is high.

  Breaking down the sources of growth, the main drivers are price and volume, not new businesses:

  • Price, as value per rack rises by generation: This is the most certain component. NVIDIA's product cadence from GB200 to GB300 to Rubin VR200 lifts value per rack in each generation. A GB300 rack is about USD 3.99 million, while a Rubin VR200 rack is estimated at about USD 7.8 million, nearly double. Under the Buy & Sell model, high-priced GPU/HBM components are fully included in Quanta's revenue. The same number of racks, but more expensive racks, can lift revenue by itself.
  • Volume, as rack shipments double: Morgan Stanley estimates annual NVL72 rack shipments will rise from about 29,000 racks in 2025 to 70,000~80,000 racks in 2026. As the world's second-largest rack-scale ODM, Quanta follows this volume ramp.
  • New businesses, essentially none: Quanta has not opened a true second growth curve. Automotive electronics still account for <10%, and notebooks have shrunk to less than 15%. Neither is a successor-level new engine. Growth comes from the same business, AI rack-scale assembly, through simultaneous increases in volume and price.

  Two caveats need to be stated plainly:

  First, a doubling of revenue is not the same as a doubling of profit. Under the Buy & Sell model, the surge in revenue is largely the result of pushing more expensive GPUs through the income statement. Quarterly gross margin has already been diluted to 4.78%, and net margin is only about 2.63%. The report puts it sharply: the explosive growth in revenue has barely translated into proportional improvement in margins. For this doubling question, the numerator, revenue, is close to a win. The denominator, profit quality, is the real uncertainty.

  Second, the price driver contains a self-limiting variable. The report notes that Quanta plans to shift some orders to a Consignment model from 2026 H2, under which customer-supplied GPU components will no longer enter the income statement. This is a tradeoff of revenue for gross margin, and it will discount revenue growth.

  In one sentence: the probability that revenue doubles over five years is high. The main engine is price times volume, driven by higher value per rack and shipment ramp, not new businesses. But this is a doubling where revenue is likely and profit remains uncertain. The real question has never been whether revenue can double, but whether doubled revenue can bring doubled profit.

  Jun 10, 2026
• QFive years from now, what will take over as the next growth engine? Does this second curve exist today?4/10

  A

  Frankly, Quanta does not have a clear second curve today that is independent of AI servers. Its next growth engine is still an extension of the same curve: next-generation NVIDIA platforms plus customer-designed ASIC racks. It is not a new growth pole. This is a structural weakness of being a leading contract manufacturer.

  First, define what a second curve should mean. Baillie Gifford asks this question to understand whether, when today's main engine, AI rack-scale assembly, matures or cools, another business has already been seeded and can drive the next decade. By that standard, Quanta does not look encouraging.

  The future growth points Quanta discusses are almost all still inside the main AI server curve:

  • Ramp of next-generation NVIDIA platforms: GB300 reaches full production, and the Rubin VR200 rack price is about USD 7.8 million, nearly double the GB300's roughly USD 3.99 million. This lifts revenue to another level, but it is a bigger engine in the existing machine, not a new engine.
  • Customer-designed ASIC racks, including Google TPU, AWS Trainium, and Meta MTIA: The report treats this as a net positive. These chips still require ODM rack assembly, and Quanta has higher penetration among ASIC customers than Hon Hai, which can diversify reliance on NVIDIA and improve gross-margin structure because ASIC racks do not include sky-high GPU components. But fundamentally, this is still assembling compute racks for data centers. It diversifies the main curve; it is not a second curve.

  What could be a real second-curve candidate? The report offers only two, both weak:

  • Automotive electronics: Still <10% of consolidated revenue. The base is too small to take over for a company with annual revenue of TWD 2.4 trillion.
  • Notebooks: Already an old business in decline, with share down to less than 15%. It is the business being replaced, not the replacement.

  This is exactly the gap between Quanta and growth companies with a true second curve. It is not a platform company that can keep incubating entirely new businesses internally. It is a receiver that shifts existing manufacturing capability into the next hottest segment. The last shift, from notebooks to AI servers through QCT, founded in 2012 and patiently built for a decade, was highly successful. But where the next shift goes is not answered today. Quanta is heavily betting that the main AI capex curve lasts long enough. The report's core risk is precisely that more than 80% of revenue is exposed to AI servers. If cloud capex growth slows or reverses, earnings leverage will fall sharply.

  In one sentence: Quanta has no visible, independent second curve today. The next growth engine is still the generational extension of the main AI rack curve, Rubin plus ASIC racks. That ties its fate deeply to a single AI capex cycle and leaves it without the kind of regenerative capacity Baillie Gifford prizes, where a new business can take over even after the core business peaks.

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

  A

  Quanta's core competitive advantage is NVIDIA rack certification, L11 rack-level system integration capability, and more than thirty years of supply-chain scale. The moat is moderate, rated 2.5~3 / 5 in the report. Over the next three to five years, the likely path is that it can protect orders but not profit. The moat is likely to stay flat or even narrow slightly. It can defend its position as the world's second-largest rack-scale ODM, but not a respectable gross margin.

  First, the real side of the moat. It can indeed protect its ability to win orders:

  • Scarce certification: Only 4~5 ODMs globally have obtained NVIDIA Blackwell rack certification. Certification involves deep engineering collaboration across cooling, power supply, and system stability, and new entrants cannot acquire it quickly.
  • L11 rack integration is a real barrier: A single rack consumes more than one hundred thousand watts, with GB200 NVL72 at about 120,000 watts and GB300 even higher. The engineering complexity of liquid cooling, power delivery, mechanical structure, and system testing is far above standard racks, separating Quanta from vendors that only make motherboards or nodes.
  • Scale and supply chain: More than thirty years of notebook ODM experience have honed component procurement, capacity scheduling, and global delivery across Taiwan/China/U.S./Mexico/Thailand. When AI server lead times are extremely tight, this is genuine competitiveness.
  • Customer ties: Years of cooperation with Google/Meta and others, plus order visibility to 2027, mean switching costs are not low.

  But the weak side of the moat matters more. It does not protect profits, and that weakness is getting wider and deeper:

  • Pricing power is extremely weak: Quarterly gross margin is 4.78%. Value-chain profits are squeezed from both ends by upstream NVIDIA, whose GPU gross margin is about 70%, and downstream cloud customers.
  • Customers are also designers and can split orders across multiple suppliers at any time: This is hard evidence that the moat is narrowing. The report notes that at the GB300 stage, Quanta's share has been diluted by new entrant Inventec from about 40% in GB200 to about 30%, proving that share can be diluted. Third-party data also show Hon Hai at about 52%, Wistron at about 21%, and Quanta at about 19% of cumulative GB200+GB300 shipments in 2025. The exact definition differs slightly from the report's statement of about 2,100 rack shipments and global number two, but the conclusion is consistent: Quanta is firmly number two, and share can be reshuffled.
  • The product is highly standardized and replaceable: Differences among ODMs are more about capacity, yield, and delivery timing than irreplaceable technology.
  • Peers show Quanta is not the strongest: Wiwynn's recent gross margin is about 8.8% in Q3 2025, the highest among Taiwan peers, far above Quanta's below 5%. In the same segment, Quanta is not the strongest in profitability.

  Will the moat widen or narrow over the next three to five years? My judgment is that the width stays flat, while the profit side narrows slightly:

  • Positive factors, protecting orders: the rack-scale trend, rising value per rack by generation, with Rubin VR200 about USD 7.8 million per rack, and continued need to assemble customer ASIC racks keep the barrier of certification plus L11 capability effective.
  • Negative factors, eroding profit: generational gross-margin decline is structural. Morgan Stanley estimates ODM gross margin will fall from about 2.7% for GB300 to about 1.9% for Rubin VR200. New entrants such as Inventec continue to split orders, and customers naturally have stronger bargaining power.

  In one sentence: this is a classic high-scale-barrier, low-pricing-power moat. Certification and scale can defend Quanta's number-two position and secure massive orders, but they cannot defend margins, and share can be sliced away by new entrants. Over the next three to five years, the river is unlikely to widen, and its profit depth is becoming shallower. This is the fundamental reason the report cannot assign a Buy rating and why Quanta does not deserve the valuation of a true wide-moat growth stock.

  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?5/10

  A

  It has some DNA for reinvention. Quanta has historically proven one successful core-business migration, from notebooks to AI servers. Its attitude toward bad news is also reasonably candid, as it proactively disclosed the gross-margin collapse at its investor conference. But this DNA is more follow-on migration than proactive disruption, and it relies heavily on the founder personally. Whether the ability to reinvent is systematic and institutionalized remains uncertain.

  First, whether the reinvention DNA truly exists. The evidence is positive:

  Quanta's most convincing act of reinvention was transforming from the world's largest notebook ODM into the world's second-largest AI server rack-scale ODM. The key was not the 2023 explosion, but 2012, when it founded QCT (Quanta Cloud Technology) to enter cloud white-box servers, a full ten years before this AI wave. When the ceiling of the mature notebook business, with low single-digit gross margins for years and slowing growth, had begun to appear, Quanta bet ten years early on the next arena and cashed in when NVIDIA rack demand erupted in 2023. This shows a forward-looking gene of planting seeds for the next stage while the main business is still making money, which is a quality Baillie Gifford values.

  But this DNA has clear limits in type:

  • It is follow-on migration, not disruptive creation. Quanta transferred existing strengths in precision manufacturing, component management, and scaled delivery into the next hot arena, from notebooks to AI servers. It is good at quickly absorbing a new wave, but not at inventing the wave itself. If AI rack assembly is one day disrupted by a more fundamental technology, such as customers fully building their own production lines or a major change in manufacturing paradigm, can Quanta migrate again? History offers a positive reference, but no guarantee. The previous migration worked because QCT had been planted ten years in advance. Today, no comparable next chess piece is visible, as discussed in the second-curve question.

  Next, how it handles mistakes and bad news. The attitude is fairly candid:

  • Proactive disclosure of bad news: The sharp drop in 2026Q1 gross margin to 4.78%, the lowest in nearly fifteen quarters, was a clear negative. Quanta proactively and clearly explained it at the 2026-05-14 investor conference, acknowledging that the causes were the doubling of high-priced GB300 racks and memory price increases. The stock fell about 8.5% the next day, but the company did not gloss over the issue. It also raised the full-year outlook for general servers. This stance of stating the bad news plainly while offering a response is better than many companies that obscure poor performance.
  • It offered a remedy: Facing the gross-margin weakness, Quanta explicitly proposed introducing a Consignment model for some orders from 2026 H2 to ease the pressure. This is a pragmatic response of recognizing the issue and adjusting tactics, even though the cost is trading revenue for gross margin.

  However, one governance concern weakens the durability of this DNA: the reinvention capability is highly tied to founder Barry Lam personally. The report notes that Barry Lam is already 77, with no clear successor, and that both sons have no intention of taking over. Quanta's corporate-governance evaluation is only in the second tier, top 6~20%, not top-tier. The historical successful migration carried a strong imprint of founder vision. Whether this gene of betting ten years early on the next stage can be institutionalized after this central figure steps back is the real question.

  In one sentence: Quanta has completed one textbook reinvention, from notebooks to AI servers through QCT's ten-year early positioning, and it is willing to disclose bad news candidly and offer remedies. The DNA is real. But it is follow-on migration rather than disruptive regeneration, and the vision is highly tied to 77-year-old Barry Lam, with institutional succession uncertain. This is a company that has proven it can reinvent itself once, but the next time is not assured.

  Jun 10, 2026
• QDoes management, especially the founder, have a long-term horizon and deep alignment with the company? Is it willing to sacrifice current profit for the next five to ten years?6/10

  A

  Founder Barry Lam is deeply aligned with the company, with about 25.6% ownership and zero pledging, and his long-term horizon is supported by evidence, especially the QCT investment ten years in advance. But willingness to sacrifice current profit for the next five to ten years is only neutral to slightly weak. Quanta looks more like it is borrowing and expanding for growth than actively compressing current profit to invest in the future, and succession uncertainty at age 77 is a governance risk hanging over the company.

  Start with alignment of interests. This part is solid:

  • Founder Barry Lam, born in 1949, controls about 25.6% of the company, with 10.76% personally plus 14.82% through Chien-Yu Investment, and zero pledging. Zero pledging is an important positive because the major shareholder has not used shares as collateral for financing and faces no forced liquidation risk. His interests are genuinely tied to the company's long-term value.
  • Co-founder, vice chairman, and president C.C. Leung owns about 2.14%, and is also a founding veteran still in office.
  • Quanta has paid dividends for 32 consecutive years. FY2025 cash dividend was TWD 15.6, with a payout ratio of about 80%. Management has treated shareholders well over the long term and distributed real cash to them, a long record of shareholder orientation.

  Now look at long-term horizon. There is real evidence:

  • The strongest evidence is the 2012 founding of QCT to enter cloud servers. While the notebook core business was still a cash cow and the AI wave had not yet appeared, Quanta placed a ten-year early bet on the next arena and realized it in 2023. This is textbook long-term positioning and reflects that the founder did have the foresight to bet early for the next decade.

  But on willingness to sacrifice current profit for the next five to ten years, the answer needs a discount:

  • Quanta's current shape is more borrowing and expanding for high growth than actively sacrificing current profit to invest in the future. The evidence is that operating cash flow has been negative for two consecutive years, about -TWD 39.5 billion in FY2024 and about -TWD 10.4 billion in FY2025, and the balance sheet has moved from net cash to net debt of about TWD 118.2 billion. But this is not management voluntarily giving up profit for the long term. It is the passive result of advancing cash for sky-high GPU/HBM components under the Buy & Sell model while working capital is consumed by high growth. In other words, thin profit and cash drain are the fate of the business model, not fully a strategic choice.
  • A subtler conflicting signal: Quanta maintains a high 80% payout, which treats shareholders well but distributes cash rather than retaining it for reinvestment, while the board resolved on 2026-02-26 to pursue a GDR cash offering of about 245 million shares, about 6.4% of share capital, to raise about USD 2.7 billion. Generous dividends and dilutive fundraising point in opposite directions. They reflect tension between maintaining a shareholder-return image and raising funds for growth. This is not a pure long-termist posture of sacrificing the short term and retaining every dollar for reinvestment.

  The largest governance uncertainty is succession:

  • Barry Lam is already 77 and has said he wants to work for another 5~10 years, but there is no clear successor. Both sons have exited Quanta's operations and have no intention of taking over. The likely path is professional-manager succession plus continued family ownership, but governance continuity is a medium- to long-term uncertainty. The report also notes that Quanta's corporate-governance evaluation is in the second tier, top 6~20%, not top-tier. Baillie Gifford prefers companies where the founder is deeply involved and succession is clear. Quanta satisfies the first half and gets stuck on the second.

  In one sentence: Barry Lam's interests are deeply aligned with Quanta, with 25.6% ownership and zero pledging; long-term vision is evidenced by QCT's ten-year early positioning; and 32 consecutive years of dividends show fair treatment of shareholders. These all pass. But sacrifice of short-term profit for the long term is more passive acceptance of thin ODM economics than active strategic choice. Add the tension between an 80% payout and GDR fundraising, plus the hard flaw of unclear succession at age 77, and this dimension is only neutral to positive with clear blemishes.

  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 regulatory arbitrage?5/10

  A

  If Quanta disappeared tomorrow, customers, namely cloud providers, would suffer short-term pain but would not miss it in a fatal way. It occupies an important but replaceable position: the world's second-largest rack-scale ODM. Without it, capacity would tighten and delivery schedules would be hit, but Hon Hai, Wistron, Inventec, and Wiwynn could fill the gap after a painful ramp. Its growth model is clean and sustainable, not dependent on social harm or regulatory arbitrage.

  Start with how much customers would miss it. This is a test of indispensability, and Quanta scores in the middle:

  • Reasons it would be missed, with real short-term pain: Quanta is one of only 4~5 ODMs globally with NVIDIA Blackwell rack certification and is the world's second-largest rack-scale ODM. Morgan Stanley data show Quanta ranks firmly second in cumulative GB200/GB300 rack shipments. AI server lead times are extremely tight, and capacity is a hard constraint. Abruptly removing a major ODM would strain cloud providers' compute expansion plans in the short term and delay delivery. L11 rack integration, involving liquid cooling, power delivery, and system testing for racks above one hundred thousand watts, also has real engineering barriers. Not any factory can immediately step in.
  • Reasons it would not be missed fatally, given high replaceability: This is the fate of the ODM business. Customers, including Google/Meta/AWS/Microsoft, are themselves the system designers and purchasers, and they are accustomed to splitting orders across Quanta, Hon Hai, Wistron, and Inventec. The report provides hard evidence: at the GB300 stage, Quanta's share fell from about 40% to about 30% after new entrant Inventec took share, proving that share already moves among suppliers and a single ODM can be replaced. If Quanta disappeared, cloud providers would redistribute orders to the remaining companies, with Hon Hai larger in scale and Wiwynn higher in gross margin, and could fill the gap after a capacity ramp. The products are highly standardized, and differences among ODMs are more about capacity/yield/delivery than irreplaceable technology. That is the essential difference between an important supplier and an indispensable hub.

  A comparison helps. If NVIDIA disappeared, the definer of GPUs, NVLink, and reference designs, the entire AI compute industry would seize up. If Quanta disappeared, the industry would stumble briefly and then route around it. Quanta is the number two among the people selling picks and assembling picks for miners. It is not the holder of the gold-mine key.

  Now on whether growth is sustainable and not dependent on harming society or regulation. Quanta performs cleanly here:

  • It does not rely on regulatory arbitrage or externalizing social costs: Quanta earns a real assembly spread. It is not making money through data abuse, regulatory avoidance, or harming consumers or the environment. Its products, AI compute infrastructure, serve the clear physical demand of data-center construction, with very low social controversy.
  • Governance and compliance record is positive: It has paid dividends for 32 consecutive years, major shareholders have zero pledging, and its corporate-governance evaluation is in the second tier, top 6~20%, not top-tier but within a good range.
  • The true sustainability risk is not social or regulatory, but cyclical: The report identifies the core risk as more than 80% revenue exposure to the AI capex cycle. If cloud capex growth slows or reverses, growth will lose speed. There are also geopolitical and tariff risks, such as the U.S. Section 232 investigation that may impose tariffs of up to 25% on chips/electronics and the threat of 25% tariffs on Mexican capacity. But these are external policy risks, not evidence that Quanta's own growth model is improper.

  In one sentence: if Quanta disappeared, customers would suffer short-term pain but would not miss it fatally. It is the important but replaceable global number two, and cloud providers can use peers to fill the gap after capacity ramps, since share already moves among suppliers. Its indispensability is moderate. But its growth model is clean and sustainable, not dependent on social harm or regulatory arbitrage. The real sustainability threats are the AI capex cycle and geopolitical tariffs, not business ethics.

  Jun 10, 2026
• QWhat are the unit economics of this business, including gross margin and incremental returns? Do they improve or worsen as scale grows? Where does the money it earns go?3/10

  A

  The unit economics of this business are poor, and they worsen rather than improve as scale grows. Quarterly gross margin has been diluted to 4.78%, net margin is about 2.63%, and incremental revenue brings marginal profit diluted by high-priced GPUs. Most of the money earned is swallowed by working capital, namely GPU/HBM prepayments, and distributed to shareholders through an 80% payout, rather than reinvested internally at high returns. This is the weakest dimension in the whole report, with earnings quality rated 2.0/5.

  Start with the absolute level of unit economics. It is very thin:

  • Gross margin: 2026Q1 quarterly gross margin was 4.78%, the lowest in nearly fifteen quarters, down about 1.54 percentage points QoQ and about 3.14 percentage points YoY. FY2025 full-year gross margin was about 6.98%, and operating margin was about 4.12%.
  • Net margin: 2026Q1 net margin was only about 2.63%.
  • Compared with the value chain: NVIDIA GPU gross margin is about 70%, while Quanta's rack assembly is below 5%. Most of the value is captured upstream.

  Now consider whether it improves or worsens as scale grows. It structurally worsens, which is the key point:

  • Intuitively, contract manufacturing should have scale effects, but Quanta's reality is the opposite. The reason is product-mix deterioration combined with the Buy & Sell model. As AI racks become a higher share and increasingly expensive GPU/HBM components are included in the revenue denominator, gross margin percentage is diluted. The report attributes the Q1 gross-margin collapse to product mix, about 2/3, GB300-generation transition trial-production expense, about 1/3, and memory price increases.
  • Worse, this deterioration continues by generation: Morgan Stanley estimates ODM gross margin will fall from about 2.7% for GB300 to about 1.9% for Rubin VR200. In other words, the larger the scale and the more expensive the racks sold, the lower the gross-margin percentage goes. This is a reverse scale effect: larger revenue, thinner unit margin.
  • The only relief is the introduction of Consignment from 2026 H2, under which customers supply components and GPUs do not enter the income statement. But the cost is slower revenue growth. This trades revenue for gross margin and treats the symptom, not the root cause.

  Now look at incremental returns, meaning the quality of ROIC/ROE. They look strong, but quality is low:

  • ROE looks attractive, with FY2025 year-end at 31.19% and TTM at about 36.84%, but DuPont decomposition exposes the issue: about 3% thin net margin times about 1.87x high asset turnover times high financial leverage equals high ROE. The report puts it well: this is ROE propped up by turnover and leverage, not ROE earned through pricing power and margins. Its quality is far below that of branded companies or software companies. Incremental capital produces thin profit diluted by expensive components and magnified by leverage.

  Finally, where does the money earned go? There are three destinations, none with high reinvestment return:

  1. Working capital, the largest cash sink: Under the Buy & Sell model, Quanta must first advance massive funds to purchase GPU/HBM, which sit in inventory and receivables. This directly led to negative operating cash flow for two consecutive years, about -TWD 39.5 billion in FY2024 and about -TWD 10.4 billion in FY2025, and the balance sheet moved from net cash to net debt of about TWD 118.2 billion. High growth is consuming working capital. Reported profit has not turned into free cash flow.
  2. Dividends, with an 80% payout: FY2025 cash dividend was TWD 15.6, payout ratio about 80%, and dividend yield about 3.99% at the current price. Distributing most earnings means the company is not retaining profits at scale for high-return reinvestment. That is reasonable for a mature cash cow, but for a company requiring large working capital to support high growth, paying dividends while also raising GDR capital is logically strained.
  3. External financing to fill the gap: Because internal cash flow is consumed by working capital, the company is turning to debt plus a planned GDR offering of about 6.4% of share capital, about USD 2.7 billion, to cover the working-capital gap.

  In one sentence: Quanta's unit economics are poor and deteriorate by generation as scale grows, from 4.78% gross margin toward an implied roughly 1.9% in the Rubin generation. ROE is high, but driven by turnover and leverage rather than margin. The money earned is swallowed by working capital for sky-high GPUs, causing operating cash flow to be negative for two straight years. This is a hard business where revenue gets larger while free cash flow becomes tighter, and it is the report's clearest weakness.

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

  A

  For Quanta to rise fivefold in ten years, roughly a 17.5% annualized return, three things must hold at the same time: revenue must keep growing rapidly, gross margin must structurally stabilize and recover, and valuation must not contract materially. These three conditions constrain one another, so achieving all of them simultaneously is not very realistic. At today's share price of about TWD 376.5, the market has already fully priced in an AI-server-leader re-rating, with TTM P/E at a ten-year absolute high. It embeds an optimistic expectation that growth keeps materializing and leaves little margin of safety.

  Break down the conditions required for a fivefold return in ten years:

  1. The revenue engine must not stall for a long time: The AI capex supercycle must last at least to 2028~2030. Rack shipments and value per rack must keep rising, with Rubin VR200 at about USD 7.8 million per rack, nearly double GB300's about USD 3.99 million. Quanta must defend its global number-two position and avoid further share erosion by Inventec/Wistron.
  2. Gross margin must disprove the ODM fate thesis: This is the hardest condition. Current quarterly gross margin is 4.78%, and Morgan Stanley estimates ODM gross margin declines by generation, from about 2.7% for GB300 to about 1.9% for VR200. A fivefold return in ten years requires EPS, not just revenue, to multiply several times. That means the Consignment model plus improved ASIC rack mix must lift net margin from about 2.6% in a real and durable way. This directly collides with the industry trend of generational gross-margin decline.
  3. The valuation midpoint cannot fall sharply: Current TTM P/E is about 20x, while some data sources using a higher TTM EPS definition show about 16~17x. Under either definition, it is a high level after the AI re-rating. In the pure ODM era from 2015~2022, Quanta traded for only 6.5~11x for a long time. A fivefold return in ten years requires the market to keep valuing it as a growth stock and not push it back to an ODM discount.

  Are these three conditions realistic at the same time? Not very, because they fight each other internally. Condition 1, revenue growth driven by Buy & Sell and high-priced components, pressures gross margin and offsets condition 2. The remedy for condition 2, Consignment, under which GPUs do not enter the income statement, slows revenue and offsets condition 1. Requiring all three to hold means asking revenue growth, margin recovery, and a high valuation to move optimally at the same time, even though the variables constrain one another. That is the bull case. The report's optimistic range of TWD 440~520, only about +17%~38%, still requires both sustained high valuation and EPS upgrades. A fivefold return over ten years would require several more doublings beyond that.

  Now consider what today's share price implies. Optimism is already priced in, and the margin of safety is thin:

  • The current price is about TWD 376.5, with market cap about TWD 1.44 trillion, or roughly USD 47 billion. TTM P/E is at an absolute ten-year high. The market has fully priced in the re-rating from an old notebook manufacturer to a core AI compute name.
  • Sell-side pricing also confirms limited upside: FactSet's median analyst target price is 385, only about +2~4% above the current price, with a high target of 425 and a low target of 250. The report judges that the current price sits in the middle of the neutral range, 350~410, with thin margin of safety and fairly efficient pricing, neither cheap nor obviously overvalued.
  • In other words, buying today embeds the expectation that AI growth keeps materializing and margins stop worsening. That is a neutral continuation assumption, not a discounted entry that leaves room for a fivefold return. The report's ideal buy price is about TWD 320, at a forward P/E of about 14x and a roughly 15% pullback from the current price, where margin of safety begins to appear.

  In one sentence: a fivefold return over ten years requires three mutually constraining things to occur together: high revenue growth, structural gross-margin recovery, and no valuation contraction. That is not very realistic, which is the core reason the report assigns Hold rather than Buy. Today's TWD 376.5 share price already fully prices in the AI leader re-rating, with TTM P/E at a ten-year high and analyst upside of only 2~4%. It embeds optimistic expectations that growth keeps delivering and leaves almost no safety margin for a fivefold outcome. If making this bet, the more sensible pitch would be a pullback to about 14x forward P/E, around TWD 320, or waiting until gross margin has stabilized for two consecutive quarters.

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

  A

  Frankly, the market has basically understood and seen Quanta. This is not a mispriced or ignored stock. Its growth, position, and earnings weaknesses have all been widely discussed and priced in, with TTM P/E at a ten-year high and the median analyst target price implying only +2~4% upside. The real disagreement is not about what benefit the market has missed, but whether to value it on TTM or forward earnings. The narrative inflection point is gross margin. It will determine whether the stock gets a Davis double play or a double kill.

  First, correct the premise embedded in the question. The question assumes that the market has not realized some underappreciated truth. Quanta's case is the opposite: it is one of the most thoroughly researched and efficiently priced names, and there is no obvious cognitive-arbitrage dividend from the market not understanding, looking down on it, or failing to look far enough:

  • It is not that the market does not understand: Quanta's business, NVIDIA rack assembly, is simple and straightforward. The whole market understands it. Its financial statements, monthly revenue, and investor conferences are highly transparent, and the 2026Q1 gross-margin collapse to 4.78% was clearly explained by the company itself at the investor conference.
  • It is not that the market looks down on it: Quite the opposite, the market has given it a full valuation. Its TTM P/E of about 20x is an absolute ten-year high, versus a long-term 6.5~11x in the pure ODM era, and its P/B of about 6.86x is the highest among Taiwan peers. The market is not looking down on it; it has re-rated Quanta from an old notebook manufacturer into a core AI compute name.
  • It is not that the market cannot look far enough: Among 21 FactSet analysts, 17 rate it Buy, 3 Neutral, and 1 Sell, and order visibility to 2027 has been widely discussed. Both bulls and bears have a fairly full understanding of forward growth.

  So where is the real disagreement? It is about which denominator to use for valuation:

  • Bears look at TTM: For a contract manufacturer with gross margin below 5%, why should it enjoy a 20x tech-stock valuation? This looks like a cyclical top. Goldman Sachs has long maintained a Neutral view on Quanta and is cautious on the old AI segment.
  • Bulls look at forward earnings: 2026E forward P/E is about 15~16x and 2027E about 13x, which does not look expensive for the world's second-largest rack-scale ODM with triple-digit revenue growth.

  This tug-of-war between expensive on TTM and cheap on forward is exactly visible in the recent share-price arc: after the 2026-05-15 investor conference disclosed a 4.78% gross margin, the stock fell about 8.5% the next day to about 306, then rebounded to nearly 417 as the AI trade warmed, and closed at 376.5 on June 8, down -3.59% that day. The sharp one-month move from 306 to 417 to 376.5 is the market voting back and forth between growth and earnings quality. It proves the market sees the issue clearly; it simply has not resolved the unknown of whether gross margin will stabilize.

  What will become the narrative inflection point? The answer is highly concentrated in one variable: gross margin.

  • Upside inflection, triggering a Davis double play: The Consignment model starts to work in 2026 H2, GB300 full-production trial costs are amortized, and quarterly gross margin stabilizes and rises for two consecutive quarters to 5%+ or even 6%+. Once the market confirms that the ODM fate thesis has been disproved and net margin can hold, low forward valuation of 13~16x and EPS upgrades will work together. This is the most important inflection signal repeatedly emphasized by the report.
  • Downside inflection, triggering a Davis double kill: Gross margin stays below 5% or makes new lows, ODM gross margin declines further to about 1.9% in the Rubin generation, dragging EPS below expectations; or cloud AI capex guidance is cut and order visibility narrows; or more sell-side firms follow Goldman into a conservative view. In that case, the 20x peak valuation will contract toward the historical midpoint of 9~12x.
  • Key observation windows: 2026 Q2 and Q3 investor conferences, to test Consignment adoption and GB300 full-production gross margin; monthly revenue around the 10th of each month; and capex guidance from NVIDIA and the four major cloud providers.

  In one sentence: Quanta is not a mispriced stock that the market has failed to recognize. Its strengths and weaknesses are fully visible and fully priced, with TTM valuation at a ten-year high and analyst upside of only 2~4%. The real disagreement is whether to value it on TTM or forward earnings, and the narrative inflection point that resolves this disagreement is concentrated in gross margin. Whether gross margin can stabilize and rise for two consecutive quarters in the second half will determine whether it gets a Davis double play or double kill. That is also the fundamental reason to Hold and watch the margin inflection closely rather than chase at the current high price.

  Jun 10, 2026