Asia Vital Components: Deep-Dive Research Using the Zen Horizon Framework

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?5/10

  A

  Bottom line: the ceiling is genuinely high, but it needs to be split honestly into "two layers"

  AVC's market ceiling (TAM) is indeed high, and its foundation is physics rather than hype. But its essence is "securing a position in an existing pie that is expanding at a steepening slope (data center cooling migrating from air cooling to liquid cooling), and earning the step-up in value content plus share," rather than creating a brand-new market from scratch. Demand for this pie is also derivative of NVIDIA GPU shipment cadence plus North American cloud capex, so it is highly cyclical and can amplify volatility. Measured with Baillie's LTGG yardstick, the slope of the track is steep enough, and AVC is the largest-scale player, which is its hardest strength. But it would be dishonest to force a "fivefold in ten years" growth narrative by calling it an acyclical, ceilingless "new-market creator." This question gets "strong, but with discounts for derivative demand and cyclicality."

  Layer one: the physical foundation is real, and penetration is in the steep-ramp phase

  This round of cooling upgrades has a solid physical basis, not just a theme: the physical limit of air cooling is around 700W, while NVIDIA GB200 single-GPU TDP has reached about 1,000–1,200W, GB200 NVL72 full-rack power consumption is around 120–140kW, the reference design ships with liquid cooling preinstalled and has no pure air-cooled version, and GB300 moves even more fully to liquid cooling. Once power consumption crosses the ceiling of air cooling, liquid cooling moves from "optional" to "must-have," an irreversible structural migration.

  Penetration is in the steep-ramp phase: TrendForce puts AI data center liquid-cooling penetration at about 11–14% in 2024 → about 24–33% in 2025 → an estimated about 47% in 2026, driven precisely by single-chip TDP crossing the roughly 700W physical limit of air cooling. Add that Blackwell already accounts for more than 80% of NVIDIA's high-end GPU shipments, order visibility extends to 2028–2029, and ASIC cooling contribution is expected to surpass GPU in 2027. This means TAM does not rely only on the NVIDIA leg; a second ASIC leg is also growing. The four major components of direct liquid cooling (DLC) are cold plates + CDU + manifolds + quick disconnects, and AVC supplies them as a one-stop provider, making it one of the few vendors able to capture the whole order book from the "cooling value-content step-up."

  Layer two (honesty point one): this is "value-content step-up plus share in an existing pie," not market creation from thin air

  This must be stated plainly: the substance of the liquid-cooling TAM is data center cooling, a job that has "always existed," shifting from air cooling to liquid cooling with unit prices several orders of magnitude higher. Servers have always needed cooling. What changes is the value content per watt cooled and the share AVC can capture, not a market appearing from nothing. This is exactly where AVC's execution shows up: liquid-cooling revenue share rose within two years from less than 2% in 2023 to about one quarter in 2025 (22–28%), the company's target is to challenge 50% in 2027, FY2025 revenue was pushed directly to about 139.6 billion (up about 95% YoY), 2026Q1 revenue reached 49.038 billion (up 110% YoY), and server + networking already reached 66.4% of the mix. In other words, the value-content step-up has landed in the financial statements as real money, not PPT.

  But under Baillie's standard, the ceiling of "expanding an existing pie + taking share" is ultimately constrained by two limits: ① total shipments of servers/AI compute across the whole market (AVC makes the cooling layer inside it, a derivative layer rather than the source); ② AVC's own share. It has strong-share, quasi-chokepoint positions in cold plates/NVLink trays/system integration, but it is not an exclusive monopoly. Auras and other second suppliers are taking orders, and lower-barrier segments such as manifolds face red-ocean pressure. So TAM is large, but the slice AVC can steadily take deserves a discount for "share being eaten into."

  Layer three (honesty point two): derivative demand = high cyclicality, and the "transition period" is its thermometer

  AVC's demand is not an independent variable. It is a dependent variable of NVIDIA GPU shipment cadence + North American cloud capex. That means the TAM slope is steep, but the path is not smooth: once upstream slows, or a platform transition leaves an air pocket, cooling vendors will see amplified volatility. The current evidence is the "transition period" in 2026 Q2. The company itself acknowledged that end customers are reallocating resources for new platforms in the second half, April revenue was already down 13.24% MoM, and reports that NVIDIA Rubin vapor chambers may change from two pieces to one piece and remove gold plating directly pressure cooling-component ASP. None of this negates the long-term ceiling, but it reminds us that large TAM ≠ linear growth. Derivative demand that follows NVIDIA's cadence has cyclical volatility as an intrinsic trait, and that must be priced into risk. This is also part of why the report gives a Watch rating and treats the current price of NT$2,710 / latest around NT$2,565 (2026-06-05 intraday, high beta, once down about 5% from the previous close in early trading, with large intraday swings) as "expensive price, weakening momentum with cracks," with a reasonable buying ceiling around NT$2,100.

  Baillie lens: what has the market not fully realized?

  What Baillie is really pressing on here is years 3–10. Honestly, the market has already become quite aware that this track has a "high ceiling" for AVC. Foreign broker target prices have reached above 3,300, and PE has expanded from a historical roughly 12 times to above 40 times, so optimistic expectations are mostly priced in. What the market has not fully priced is not "how high the ceiling is," but "how much share AVC can steadily take in this steeply rising pie, and how much the cyclical swings of derivative demand will bend the path". Those two points are the real swing factors for returns in years 3–10, and they are what later questions on Q4 moat, Q7 customer stickiness, and Q9 valuation must continue to answer.

  One-sentence framing: on the TAM dimension, AVC is a "strength". The physical foundation is solid, penetration is rising steeply, and it is the largest-scale player. But this is "positioning in an expanding existing pie," not creating a new market, and demand is derivative and cyclical. Discounting the growth narrative is the honest thing to do; it should not be elevated into a ceilingless, acyclical great story.

  Jun 5, 2026
• QCan its revenue at least double over the next five years? Is growth mainly driven by volume, price, or new businesses?7/10

  A

  Bottom line: doubling revenue over the next five years is not a high hurdle for AVC; it has even been partly delivered already. But it is the result of resonance among three engines, "volume × price × new business," with better quality than pure volume growth. The cost is high dependence on NVIDIA and cloud capex cadence, a non-linear path, and an inevitable natural slowdown in growth after the revenue base breaks 100 billion.

  Start with the path AVC has just traveled. The "credibility" of a doubling is almost proven by actual results, not by forecast extrapolation. Revenue went from about 47.3 billion in 2021, 56.0 billion in 2022, 59.2 billion in 2023, about 71.8 billion in 2024 (YoY +21%), to about 139.6 billion in 2025, nearly doubling within one year (YoY about +95%). Operating gross profit reached 36.0 billion, up more than 113% YoY, both setting new historical highs. Entering 2026, this momentum has not broken: 2026Q1 revenue was 49.038 billion, YoY +110%, quarterly EPS was 20.17, a historical high, and gross margin rose to 29.77%; cumulative revenue for 1–4 months was 64.669 billion, cumulative YoY +99.34%. In other words, AVC's current annualized run rate itself is on a path that is "approaching a doubling year by year". Baillie's low hurdle of doubling in five years looks more like a baseline than a challenge for it.

  More importantly, the growth has three drivers that are independent yet mutually reinforcing, which determines the "quality" of the doubling:

  • Volume (AI server ramp): server + networking combined has risen to 66.4% of the mix, and the core business is latched onto AI compute infrastructure, the thickest demand curve. This is the base power of scale.
  • Price/mix (sharp rise in high-value liquid cooling): liquid-cooling revenue share rose from <2% in 2023 to about 22–28% in 2025, and the company's target is to challenge 50% in 2027. Liquid-cooling modules (cold plates + rack-side systems) carry far higher unit value than traditional air cooling. Even if the number of shipped "pieces" is unchanged, product mix moving upward can itself drive stepwise revenue increases. This is the key difference between AVC and pure contract-manufacturing volume growth, and it is also why gross margin can jump to 29%+ from the air-cooling era.
  • New business (ASIC cooling as the second engine): as cloud providers' in-house ASICs adopt water-cooling designs and enter volume production, the company guides that ASIC cooling contribution is expected to surpass GPU in 2027. This means that beyond the NVIDIA GPU curve, a new incremental curve driven by CSP in-house chips is emerging, broadening the source of the "doubling path" from a single customer to the entire AI accelerator ecosystem.

  Order visibility also supports the path. Management's visibility has extended to 2028 and beyond. With liquid-cooling penetration still in the steep-ramp phase and the ASIC engine still climbing, revenue additions over the next 1–3 years are "visible and tangible," exactly the kind of "intrinsic volume-price quality" Baillie values most.

  But three cracks need to be stated honestly, so a strength is not portrayed as risk-free:

  First, cyclicality makes the doubling path non-linear, with a short-term transition-period crack. AVC's demand is essentially derivative demand from GPU shipment cadence and cloud capex, and it can be amplified by upstream changes. This already appeared in April 2026: monthly revenue was 15.631 billion, MoM −13.24% (though still YoY +71.62%), and the company clearly called Q2 a "transition period." Replacement windows, customer stocking cadence, and platform transitions can all create pullbacks in monthly/quarterly numbers. The five-year doubling is highly likely to be "visible, but bumpy."

  Second, customers and demand are highly concentrated. The growth engine is concentrated in NVIDIA + a few CSPs. Once cloud capex narrows, or AI capital spending enters a digestion phase, AVC will feel destocking later and more violently than terminal chipmakers. In addition, if next-generation platforms such as Rubin move toward single-piece/integrated designs, there is structural risk of compressing cooling ASP per unit, which could offset part of the "price" dividend.

  Third, base effect. The revenue base has broken 100 billion. The near doubling from 2024→2025 was a "perfect year" created by a low base plus liquid cooling jumping from 0; it is hard to repeat every year. Growth over the next five years will naturally slow, and "doubling in five years" will more likely be achieved through a curve that is steep in the early stage and slower later, not through linear, even growth.

  Overall judgment (Baillie lens): question 2 gets "strong." The probability of doubling revenue in five years is high, and the driver structure is healthy (three engines of volume + price + new business, with gross margin rising at the same time rather than exchanging price for volume). Part of the target has already been pre-delivered by 2025 results and the 2026Q1 run rate. But this "strong" is tied to the external engine of AI capex. Cyclicality and customer concentration mean it is not a smooth straight line. What must truly be validated in years 3–10 is whether liquid-cooling penetration can move toward 50% on schedule, whether the ASIC engine can smoothly take over from GPU in 2027, and whether the company can shift growth from "volume ramp" to "value content and system integration" after the base becomes larger.

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

  A

  Second curve: the candidates are concrete, but the substance looks more like "customer diversification of the main curve" than a fully independent second engine — neutral to positive

  Bottom line: AVC does have a second-curve candidate far more concrete than most companies' "blank checks," but it shares the same TAM and the same growth engine as the core business. The report rating is Watch, the current price is about 2,565 (2026-06-05 intraday, base price 2,710 / 2026-06-04 close), and the core business is AI GPU server liquid cooling/cooling. Baillie asks "who takes over after five years, and does this curve exist today?" AVC's answer is: the successor is already forming in the financial statements today (ASIC cooling, liquid-cooling value-content upgrade, Fositek quick disconnects), which is positive; but they look more like "extensions and diversification" of the main curve than another physically independent engine, so independence must be discounted.

  Strongest candidate: ASIC (cloud in-house chip) cooling, with contribution expected to surpass GPU in 2027. This is the most real and most visible second curve for AVC. The company has entered liquid-cooling solutions for cloud providers' in-house ASICs (Google TPU, AWS Trainium, Meta MTIA, etc.). These ASIC platforms have been gradually adopting water cooling from the second half of 2025, and the company has stated clearly that the revenue contribution from ASIC water-cooling solutions is expected to surpass GPU for the first time in 2027, with order visibility extending to 2028–2029. The value of this curve is that it shifts AVC from "betting on the cadence of a single NVIDIA platform" to a multi-customer structure of "GPU + multiple cloud providers' in-house chips," hedging the customer concentration risk named in the report (one May rumor about Rubin single-piece design triggered a 20% drop over 7 days). But it must be said honestly: ASIC cooling and GPU cooling both belong to the same large TAM of "AI data center compute cooling" and are driven by the same variable of North American cloud capex. This is "customers expanding from NVIDIA to Google/AWS/Meta," while the cooling components themselves (cold plates, manifolds, CDUs, quick disconnects) are almost the same product set and production lines. So it is more like a horizontal extension/customer diversification of the main curve, not a physically independent second engine that can keep running by itself if the core business stalls. If North American cloud capex slows overall, GPU and ASIC cooling will come under pressure together, so it cannot be called an "independent handoff."

  Second candidate: continued liquid-cooling value-content uplift (mix upgrade, not a new curve). Liquid-cooling revenue share moved from less than 2% in 2023 → about one quarter in 2025 (22–28%) → the company's target to challenge 50% in 2027. Water-cooling gross margin of about 35% is higher than 3D VC's 25–30%, lifting overall gross margin from about 26% in 2025 to 29.77% in 2026Q1. This is a real and already-delivered profit-structure improvement, but strictly speaking it is a "product mix upgrade toward higher value" inside the core business, not an independent second curve. It does not open a new demand source for the company; it only thickens unit value inside the same batch of AI server demand. Treating it as a second curve would overstate its "independence."

  Third candidate: Fositek (6805.TWO) quick disconnects + foldable-phone hinges, the most independent of the candidates. Fositek is about 17% owned by AVC and consolidated. 2025 revenue was 12.414 billion, up 51.6% YoY. It makes AI server liquid-cooling quick disconnects (UQD, the part of the liquid-cooling system most prone to leakage and with the highest technical threshold and compensation requirements, sharing the market with Danfoss), and also foldable-phone hinges. The hinge leg is genuinely decoupled from AI compute capex. It follows foldable-phone penetration and the consumer-electronics cycle, making it the only curve in AVC's map that does not rely on data centers, and it has the highest independence among the candidates. Quick disconnects, however, still belong to the AI liquid-cooling theme. But Fositek's size within AVC's consolidated statements is limited (about 8.5% of revenue in 2026Q1), and even with high growth it cannot carry the weight of "taking over from the core business" in the short term. It is more of a marginal enhancer and moat reinforcement (owned quick disconnects make AVC a "single-window" integrator).

  Honest ruling under the Baillie lens: rank the candidates by "independence": Fositek hinges (truly independent, but small) > ASIC cooling (highest visibility and largest scale, but same theme and same driver as the main curve) > liquid-cooling mix upgrade / CDU system integration (essentially an extension of value content and breadth inside the core business). AVC's second curve "does exist today and is already ramping in the financial statements," which is far better than companies that only draw slides, and it is a real positive. But the largest and most visible one (ASIC cooling) is not another physically independent engine. It is "the same AI cooling engine with a few different customers," its delivery waits until around 2027, and it faces the same risks as the core business: Rubin single-piece design pressuring ASP, Auras taking orders, and cloud capex cyclicality. Therefore this question gets a restrained neutral-to-positive judgment: it is true that the second curve is "tangible and concrete," but "independent handoff" is discounted by the fact that it has the same theme and same driver. It is more about making the main curve wider, longer, and thicker, rather than truly diversifying into a second growth pole. This also echoes the core of the report's Watch rating: quality and curve depth deserve a premium, but uncertainty is concentrated in the next 1–2 quarters of momentum confirmation, while the real ramp of the second curve (ASIC surpassing GPU) still needs to wait until 2027.

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

  A

  Bottom line: AVC's moat is "medium-wide, and still trending upward over the next three to five years, but manufacturing-execution-based rather than proprietary-technology-based." Scale and the vertically integrated "single window" are real barriers and the hardest to replicate in the short term, but the "width" of the moat depends heavily on renewal of certification for each generation and continued liquid-cooling penetration, while it is already facing narrowing pressure from Auras order-splitting and liquid-cooling red-ocean competition. It should not be elevated into "wide and unbreachable."

  From strongest to weakest, layer by layer, here is whether the moat widens or narrows:

  ① Scale (the hardest, and still widening). AVC is the world's largest system-cooling vendor. FY2025 revenue was about 139.6 billion, roughly 6 times that of direct rival Auras (3324.TWO, 2025 revenue about 23.279 billion). CSP liquid-cooling adoption highly values mass-production scale and supply stability, and monthly water-cooling-plate capacity is expanding from 200,000 to 1,000,000 sets (+4 times). The scale advantage is being widened further by this expansion. This layer will probably widen over the next three to five years and is the least easily eroded part of AVC's moat.

  ② Vertically integrated "single window" (most differentiated, the main source of width). AVC makes cold plates + manifolds + chassis + CDUs in-house, and through subsidiary Fositek (6805.TWO) fills the quick-disconnect (UQD) gap, the component most prone to water leakage. It simplifies what customers previously needed to manage across 5–6 suppliers into a single window, and positions itself as a "cooling supply-chain integrator." This is the structural advantage that differentiates it from pure cold-plate vendors and is the hardest to replicate in the short term: competitors need time and acquisitions to fill out this chain. But honestly, third parties have clearly pointed out that this advantage "comes more from execution excellence than proprietary technology." It is a moat of manufacturing execution + integration breadth, not an irreplicable single-point technology. It will slowly widen as AVC continues to thicken the chain, but in substance it can be approached by rivals with capital and time.

  ③ Customer certification stickiness (upward, but needs renewal each generation). AVC has obtained NVIDIA GB200/GB300 certification, received next-generation Vera Rubin water-cooling-plate orders, and begun small-volume shipments. CSP/NVIDIA certification cycles are long and switching costs are high, creating real stickiness. But this is stickiness that must be won again for every product generation. Certification is not a one-time lifetime asset; every platform transition (GB→Rubin→future) must pass again. This is the root of the moat's "unstable width."

  ④ Leakage compensation threshold (logically favors large vendors, but lacks first-hand proof). Once liquid cooling leaks, it can damage expensive AI servers, and CSPs have strict requirements for reliability and compensation capacity, which disadvantages small vendors. This must be labeled truthfully: "AVC has better compensation capacity than peers" is a logical inference without first-hand disclosure of compensation terms, so it should not be weighted as a hard barrier.

  The key honest assessment: it is not an "exclusive chokepoint" segment. By segment, the competitive landscape is: cold plates are a duopoly of AVC and Cooler Master (each estimated by corporates/media at about 30–50% share), NVLink switch tray cooling is close to exclusive, CDUs are led by Vertiv (AVC is one of Meta's main suppliers but not the global leader), manifolds are more fragmented, and quick disconnects are shared by Danfoss, Fositek and others. So AVC is a "quasi-chokepoint with strong share + the most complete integration," not an absolute monopoly in any single segment. This means the moat has width, but not absolute pricing power.

  Narrowing pressures over the next three to five years (must be stated):

  • Auras order-splitting. Direct competitor Auras has obtained NVIDIA water-cooling-plate certification and is being tested by Amazon as a potential second supplier, standing alongside Cooler Master and AVC as core cold-plate suppliers. CSPs naturally foster second suppliers for supply-chain security, so AVC faces a real risk that cold-plate share will be eaten into.
  • Lower-barrier segments turning red-ocean. Segments such as manifolds do not have high barriers and price wars have already appeared. Liquid cooling is moving from blue ocean to red ocean, which will suppress long-term gross margin (current water-cooling gross margin about 35%, 3D VC about 25–30%).
  • Rubin single-piece design pressuring ASP. NVIDIA Rubin-generation vapor chambers are rumored to change from two pieces to one piece and remove gold plating, directly weakening the unit price of cooling components. Even if share is unchanged, it erodes the moat's monetization ability from the "price" side.

  Overall judgment: this is a "medium-wide, upward-trending, but manufacturing-execution-based moat with erosion risk". Scale + vertically integrated single window are real strengths and the hardest to replicate in the short term, and the trend is still widening as capacity expands and the chain thickens. But it is not proprietary technology, nor an exclusive chokepoint segment. The sustainability of its width is tied to two exogenous variables, "certification renewal for each generation + continued liquid-cooling penetration," while it directly faces narrowing pressure from Auras order-splitting and red-ocean competition. From Baillie's LTGG preference for wide moats that self-deepen and compound for ten years, AVC reaches "medium-wide and upward," but not "wide, self-deepening, and unbreachable." The execution-based moat and red-ocean risk are its main discounts under this framework.

  Jun 5, 2026
• QIf the core business is disrupted, does it have the gene for self-reinvention? How does it handle mistakes and bad news?5/10

  A

  Bottom line: AVC's self-reinvention gene is one of its strongest strengths, with a verifiable record. It went from a small PC cooling vendor to an AI liquid-cooling leader in three leaps, repeatedly catching paradigm shifts and laying out liquid cooling ten years early. Its attitude toward bad news is relatively candid, but optimistic. Under the Baillie lens this is a positive signal, but cooling is an industry derivative of upstream platforms, so the next paradigm still needs to be proven. The judgment is positive but restrained.

  One, verifiable reinvention record: three leaps, each catching a paradigm shift. This is one of AVC's most solid strengths, not narrative packaging. The company's business evolution can be clearly divided into three stages: ① PC/NB cooling period, mainly fans, heat sinks, and CPU cooling modules, with low gross margin, a typical low-margin electronic component contract manufacturer; ② server/communications period, entering data-center server cooling and server chassis, with the customer structure moving toward cloud; ③ AI liquid-cooling period (from 2023 to today), moving from the physical-limit air-cooling solution 3D VC (three-dimensional vapor chamber, with heat-transfer efficiency more than 30% higher than planar VC and the world's largest shipment volume) all the way to the four major liquid-cooling components: cold plates, manifolds, CDU, and, through subsidiary Fositek, quick disconnects (UQD). At the Taiwan Stock Exchange theme presentation, the company described it as "ten years to sharpen one sword" and declared the "first year of liquid cooling". The key is that each leap was not passive following, but active betting on the next paradigm. Servers, 3D VC, and liquid cooling were all seeded before demand exploded.

  Two, reinvention has been delivered in financial statements, not PPT. Chairman Shen Ching-Hsing bought AVC in 1999 when annual revenue was only about 700 million. Since then, revenue has grown about 85 times and total shareholder return is 738%. The company laid out liquid cooling ten years in advance and delivered revenue above 100 billion in 2025 (FY2025 consolidated revenue about 139.6 billion, up about 95% YoY, EPS jumping to 49.17). From the quiet period when "average annual EPS from Shen Ching-Hsing's entry to 2024 was only about 3.2," to AI server ramp in 2024 and near-doubling revenue above 100 billion in 2025, strategic foresight and execution delivery are both high quality. The other side of reinvention ability is willingness to borrow external capability to fill gaps: accepting investment from Japan's Furukawa Electric to fill core heat-pipe technology, acquiring chassis maker Mingxin and fan maker Dalong to build a full cooling map, heavily expanding in Vietnam from 2020 to supply North American cloud customers nearby, and investing in Fositek's 2023 listing to fill the quick-disconnect gap. This hybrid capability-building model of "internal R&D + external sourcing + M&A" is hard evidence of a reinvention gene.

  Three, handling bad news: relatively candid, but optimistic and still awaiting validation by results. This is where the question needs an honest discount. Facing the negative possibility that NVIDIA Rubin vapor chambers may change from two pieces to one piece and remove gold plating, the company responded that gold plating was already planned to be outsourced in the short term and built in-house in the long term; if the customer does not use gold plating, "only the outsourcing business is canceled, with limited impact." It also proactively acknowledged Q2 as a "transition period" between old and new platforms and did not avoid the fact that April revenue was down 13.24% MoM. The disclosure posture is relatively candid, itself a sign of good governance. But the point is that "limited impact" is still the company's own statement and has not yet been validated by Q2 results. The market voted with its feet (target prices up, share price down; on June 5 intraday it once fell to about 2,565, about 5% below the previous close of 2,710, with large intraday volatility), showing that the market believes single-piece design does weaken cooling-component value. Candor is real, but the optimistic framing deserves a question mark and needs cross-checking with monthly revenue for 5–7 months and Q2 gross margin.

  Four, Baillie-lens judgment: positive, but do not extrapolate historical success into inevitable next success. Baillie's LTGG asks whether a company can pass through repeated paradigm shifts and keep renewing itself. On this point, AVC's record is unusually clean: three leaps, multiple paradigm shifts caught, ten-year early layout, all delivered in financial statements. This is one of the few dimensions in the Watch rating that can be scored highly without much reservation. But a structural risk must be stated: cooling is an industry derivative of upstream platforms (NVIDIA GPU shipment cadence + North American cloud capex). The company's reinvention has historically been "following upstream thermal design and executing solutions to the extreme." If the next paradigm comes from further upstream (for example refrigerant/immersion cooling becoming mainstream, GPU vendors designing cooling in-house, or ASIC routes rewriting cooling specifications), whether AVC's "manufacturing-execution" advantage, which third parties clearly say comes more from execution excellence than proprietary technology, can complete reinvention again still needs to be proven. A historical four-win streak cannot imply a fifth win by default.

  Overall judgment: the self-reinvention dimension is positive and supported by solid evidence, one of AVC's hardest quality pillars. It is candid but optimistic toward bad news, and Q2 is the first test of that optimism. The reinvention gene has been repeatedly validated, but its reinvention has always been a derivative function of upstream platforms. The next paradigm shift still requires watching whether it can upgrade from an "extreme executor" to a "rule definer." That is the key unresolved variable in pushing it from an "excellent reinventor" to a "great compounding machine."

  Jun 5, 2026
• QDoes management, especially the founder, have a long-term view, and are interests deeply tied to the company? Is it willing to sacrifice current profits for five to ten years out?6/10

  A

  Management: one of AVC's hardest fundamentals, but governance details deserve honest deductions

  Bottom line: management is a strength, not a question mark for AVC. The helmsman Shen Ching-Hsing combines rare strategic foresight, execution delivery, and capital-allocation discipline, and his interests are deeply tied to the company's long term, fully matching Baillie's LTGG preference for "long-term vision + rowing in the same boat as shareholders." But several governance deductions (chairman and general manager combined, opaque succession bench, relatively high pledge ratio by the controlling party, and an optimistic tone toward bad news) cap this dimension. We therefore give a positive but restrained judgment: it is a positive, not a perfect score; it supports a watch stance and does not constitute a red flag.

  One, long-term view: betting on liquid cooling about ten years early, already validated by delivery. Shen Ching-Hsing (born 1963, MBA from Michigan State University, former CFO of First International Computer) took over AVC in 1999 as chairman and corporate representative of "Star Union Investment," when annual revenue was only about 700 million New Taiwan dollars. His most important foresight was laying out liquid cooling about ten years early when air cooling was still absolutely mainstream. As AI server thermal-design power (TDP) surged in 2024–2025 and liquid cooling became a must-have, AVC became a core supplier of NVIDIA AI liquid cooling and delivered 100-billion-level revenue in 2025 (full-year revenue about 139.6 billion). This was not chasing a trend after the fact. It was using years of R&D and upfront capital to secure an industry inflection point, exactly what Baillie values most: seeing far and daring to place bets before consensus forms.

  Two, execution delivery: turning vision into numbers. Over more than twenty years since taking over, revenue has grown about 85 times and cumulative total shareholder return (TSR) is about 738% (per Harvard Business Review interview). During the same period, Shen Ching-Hsing also laid out and led Fositek (6805.TWO, about 17% owned by AVC and consolidated), building dual engines of AI quick disconnects + foldable-phone hinges (2025 revenue 12.414 billion /+51.6%). He brought both AVC and Fositek into thousand-dollar-stock territory. The loop of vision—execution—delivery is complete, not mere storytelling.

  Three, capital-allocation discipline: focused, non-dilutive, and aligned with shareholders. Three clues show management treats shareholder capital as its own: ① M&A has always focused on the cooling core business (including Fositek's related diversification, which remains within thermal management/precision mechanics), with no random acquisitions for size; ② continuous cash dividends (planned cash dividend of 21.0 for 2025), returning cash rather than stock dividends and not relying on stock dividends to dilute share capital; ③ strategic shareholder Japan's Furukawa Electric, a long-term heat-pipe technology partner and major shareholder, represents industrial synergy rather than financial-arbitrage equity. This discipline of "focus + cash + no dilution" is real evidence of long-term alignment with shareholders.

  Four, attitude toward bad news: relatively candid, but optimistic, a partial deduction. Facing concerns that Rubin platform single-piece design may compress cooling-module value content, management chose to respond candidly rather than avoid the issue and acknowledged Q2 as a "transition period." The candor is positive, but the wording is optimistic, framing a structural technology risk more as a short-term "transition." Baillie's test of management comes exactly in bad-news moments: willing to speak, but whether the downside is faced directly enough. Here the grade is "above passing, not excellent."

  Five, governance deductions that must be stated honestly (capping the upside, but not red flags):

  • Chairman and general manager combined, with strong personal color. Shen Ching-Hsing serves as both chairman and general manager, concentrating decisions in a strong founder. During an upcycle this is a source of execution, but lack of board checks on management is a standard governance watch item.
  • Succession bench information is missing. Public information shows no formal, transparent succession plan or clear second-generation/successor arrangement. For a company whose strategic foresight is tied to a single central figure, key-man risk is real.
  • The pledge ratio of controlling party Star Union Investment is relatively high (old data, needs confirmation against the latest official filings). The controlling vehicle "Star Union Investment" has a relatively high share pledge: according to publicly available director/supervisor shareholding details, Star Union Investment held about 6.964 million shares and pledged about 3.522 million shares, a pledge ratio of about 50%. But the data point is August 2023, and the latest pledge ratio must be verified on Taiwan's Market Observation Post System (mops.twse.com.tw). If it remains high, sharp share-price volatility could create forced selling / margin-maintenance pressure, a potential misalignment with minority shareholders. This point is marked with old data, without the latest confirmed value.

  Six, overall judgment (Baillie lens): positive, restrained. Combining the four hard indicators (strategic foresight, execution delivery, capital discipline, and interest alignment) with the four governance deductions (dual roles, unclear succession, controlling-party pledges, and optimistic bad-news tone): AVC's management quality is clearly above the average of Taiwan hardware peers and is one of the core reasons supporting a Watch rating rather than a bearish view. But opacity and concentration in governance details mean this dimension is strong enough, not beyond criticism, and creates a watch list for ongoing tracking (especially succession disclosure and pledge-ratio changes), not a current sell red flag.

  Jun 5, 2026
• QIf it disappeared tomorrow, how much would customers miss it? Is its growth model sustainable and not dependent on harming society or regulation?5/10

  A

  Bottom line: customer stickiness is strong and the growth model is relatively healthy, but "sustainability" is dragged down by high customer concentration and liquid-cooling red-ocean competition. Overall the judgment is neutral-to-positive and restrained. Baillie splits this question into two layers: ① can customers really not do without it; ② can the growth model last over the long term without relying on harm (not price-cutting subsidies, not harming customers or the industry ecosystem). AVC's answers to the two layers are not the same: it passes the stickiness layer, but sustainability needs an honest discount.

  First layer (stickiness/indispensability): in the short term, CSPs really cannot do without it, and stickiness is real. Four thresholds combine to create high switching costs in the short to medium term:

  • Certification stickiness. AVC has obtained NVIDIA GB200/GB300 certification, received next-generation Vera Rubin water-cooling-plate orders, and begun small-volume shipments. CSP/NVIDIA cooling certification cycles are long, and switching suppliers requires rerunning validation. During a platform mass-production window, changing suppliers temporarily is highly risky. This is the hardest lock.
  • Vertically integrated "single window" (most differentiated). AVC makes cold plates + manifolds + chassis + CDUs in-house, and supplies the quick disconnects most prone to leakage through subsidiary Fositek (6805.TWO), simplifying work that customers previously had to coordinate across 5–6 suppliers into a single window. This is its stickiest source of differentiation versus pure cold-plate vendors and the hardest to replicate in the short term. Once a customer hands the whole liquid-cooling set to it, the coordination cost of unbundling and subcontracting again is high.
  • Leakage compensation threshold. If liquid cooling leaks, it can damage expensive AI servers. CSPs have stringent requirements for supplier reliability and compensation capacity, naturally keeping small vendors out and favoring large vendors with scale and compensation capacity. (Note: "AVC has better compensation capacity than peers" is a logical inference, without first-hand compensation terms disclosed as proof.)
  • NVLink switch tray cooling is close to exclusive, and with a cold-plate duopoly position, this makes it a "strong-share quasi-chokepoint" in NVIDIA full-rack reference designs.

  This layer holds. Within the current GB/Rubin platform production cycle, leading customers indeed cannot quickly switch away from it.

  Second layer (whether the growth model is sustainable): the growth model itself is healthy, but "sustainability" needs two honest discounts.

  Start with the healthy side: AVC's growth is not subsidy-style. It wins orders through scale (the world's largest system-cooling vendor, FY2025 revenue above 100 billion and about 6 times Auras), the breadth of vertical integration, and certifications generation by generation. Gross margin rises rather than falls (overall gross margin rose from about 26% in 2025 to 29.77% in 2026Q1, water-cooling gross margin about 35%). It is not gaining share through price cuts, nor by harming customers or the industry ecosystem. This is the "clean growth" Baillie would credit.

  But sustainability is dragged down by two points:

  • ① High customer concentration, with derivative demand amplified by single points. It depends heavily on the NVIDIA system + leading CSPs. In essence it is derivative demand from GPU shipment cadence + North American cloud capex (cyclical). A design change or shipment cadence change by a single customer can be amplified. In May, a rumor that NVIDIA Rubin vapor chambers would change from "two pieces to one piece and remove gold plating" triggered a roughly 20% share-price drop over 7 days (April 27 high of 3,010, May 6 intraday limit-down to 2,435). It is true that customers cannot do without it, but it also cannot do without these two customer groups. The bargaining scale is not fully in its own hands.
  • ② Red-ocean competition in low-barrier segments + second-supplier order splitting + Rubin single-piece design pressuring ASP. Direct competitor Auras (3324.TWO) has obtained NVIDIA water-cooling-plate certification and is being tested by Amazon as a potential second supplier. Its liquid-cooling share is rising faster and valuation is cheaper. Lower-barrier segments such as manifolds are facing price wars, and liquid cooling is moving from blue ocean to red ocean. With NVIDIA Rubin-generation vapor-chamber single-piece design pressuring the unit price of cooling components, whether "high growth + high gross margin" can both be sustained over the long term is in doubt. The moat itself also leans toward "manufacturing execution" rather than proprietary technology, so sustainability depends heavily on certification renewal for each generation and continued penetration.

  Overall judgment (neutral-to-positive, restrained): AVC passes the stickiness layer with meaningful weight. Certification stickiness, the single window, and leakage compensation thresholds create three locks, CSPs really cannot do without it in the short term, and the growth model is relatively clean and not based on harm. This is positive. But what Baillie truly asks, "can it remain sustainable for the long term," must be discounted: customer concentration makes growth derivative demand that can be punctured by a single point, while red-ocean competition and second-supplier order splitting erode the stability of "high growth + high gross margin." Therefore question 7 gets neutral-to-positive. Stickiness adds points, but customer concentration and red-ocean competition are unresolved sustainability question marks that must be stated honestly. This also echoes the report's overall Watch rating: "good company, expensive price, cracks where momentum is weakening."

  Jun 5, 2026
• QWhat are the unit economics of this business (gross margin, incremental returns)? Do they improve or deteriorate as scale increases? Where does the money earned get spent?5/10

  A

  Bottom line: unit economics are AVC's most underestimated real strength today, and also the part of it that looks least like a typical growth stock under Baillie's LTGG framework. It is ramping volume quickly while gross margin rises structurally, and the balance sheet is net cash. Scaling up has indeed improved unit economics. But whether this "bigger gets better" pattern continues depends on liquid-cooling penetration and whether unit prices hold. Looking forward, capex depreciation and ASP mean reversion are two hidden risks, so the judgment is "currently very strong, watch going forward," positive but restrained.

  One, current facts: scaling has indeed brought better unit economics, in the same direction Baillie wants. AVC is not a cash-burning growth stock buying growth with subsidies. It is a rare case where scale rises and gross margin structurally rises as well. Overall gross margin rose from about 26% in 2025 to 29.77% in 2026Q1. FY2025 full-year operating gross profit was 36.0 billion, up 113.6% YoY (far faster than revenue growth of about 95%, meaning gross margin expanded rather than diluted). The driver is product mix tilting toward high-value liquid cooling: water-cooling plates have gross margin of about 35%, 3D VC about 25–30%, and liquid-cooling revenue share rose from less than 2% to about one quarter within two years. The further it moves toward liquid cooling, the higher unit gross profit becomes. Add operating leverage: 2026Q1 operating margin was 24.51%, net margin about 16–17%, EPS jumped from 21.21 in 2024 to 49.17 in 2025, and 2026Q1 alone was already 20.17. Fixed costs are being diluted by fast-growing high-value revenue, so profitability scales with scale. Although exact official ROIC is unavailable, at the company level it has overall net cash (net interest-bearing debt-to-equity ratio about -51%) and rough full-year ROE of 40%+, meaning business expansion does not rely on leveraged cash burn and can roll forward on internal cash generation. This is exactly the kind of "unit economics improve after scaling" sample Baillie favors.

  Two, forward concerns: three forces may discount "bigger gets better," and must be stated honestly. First, aggressive capex raises the depreciation base: 2026Q1 capex rose 169% YoY to 3.657 billion, full-year guidance is about 15.0 billion, and monthly water-cooling-plate capacity is moving from 200,000 to 1,000,000 sets (+4 times). This is a heavy bet on liquid-cooling penetration, but it also raises depreciation and utilization leverage to a high level. If demand or unit prices miss expectations in the second half, the operating leverage amplifying profit today will reverse and amplify pressure on net margin. Second, mean-reversion pressure on unit price: NVIDIA Rubin-generation vapor chambers are rumored to move from two pieces to one piece and remove gold plating, with cooling-component unit prices rumored to fall notably. But this risk needs an honest boundary: latest market interpretation tends to point to high-end vapor-chamber specialists (such as Jentech) as the direct victims of this change, while treating AVC and Auras, which provide full-system solutions of water-cooling plates + manifolds + CDUs, as relatively less affected and even structurally benefiting. Even so, red-ocean competition in lower-barrier segments such as manifolds and direct order-splitting by Auras remain long-term downward pressure on overall gross margin. Third, the moat is manufacturing-execution-based rather than proprietary-technology-based, meaning high gross margin depends more on certification renewal and mix upgrade each generation than on an irreplicable single-point technology. The "sustainably high" level of unit economics needs continuous validation.

  Three, overall judgment (Baillie lens): positive, restrained. On the question of "do unit economics improve or deteriorate as scale increases," AVC's answer is clearly improving, and in a healthy direction: structural gross-margin rise, operating leverage, net cash, and high ROE. This is the opposite of most dilution-style growth stocks and fully matches Baillie's yardstick of "better after scaling," one of the core supports for the report's "high quality" assessment. But for the improvement to continue, watch two thermometers: whether second-quarter gross margin can hold around 29% (a direct read on ASP/red-ocean pressure), and whether aggressive capex can be fully absorbed by liquid-cooling volume ramp (a read on depreciation drag). If both hold, unit economics will continue moving toward Baillie's ideal. If Rubin single-piece design + Auras order-splitting structurally shift unit prices lower while utilization is insufficient, current high gross margin faces mean-reversion risk. Precisely because of this tension between "currently strong" and "future ASP/utilization uncertainty," the report's overall rating remains Watch (base price 2,710, latest about 2,565). Unit economics are the firmest bullish cornerstone for AVC, but not a free-pass item that can be extrapolated unconditionally.

  Jun 5, 2026
• QFor it to rise fivefold over ten years, what conditions must hold simultaneously? Are those conditions realistic? What expectations are embedded in today's share price?3/10

  A

  Bottom line: AVC is a good company that deserves a "reasonable" valuation, but at the current price of about 2,565–2,710 (PE-TTM about 42–44 times, PB about 22–24 times, cash yield below 1%), the margin of safety is clearly insufficient. A fivefold gain in ten years (about 2,600 → about 13,000) is not absurd, but the hurdle is high. It requires both strong EPS compounding and "multiples not compressing sharply." Today you are starting from a historical valuation high of 44 times, and from a position where demand is highly cyclical, so the risk of multiple downside is real. One sentence: good company, but the current price has already prepaid most of the deposit for optimistic growth in 2026–2027. What is missing is not the story, but the price buffer.

  First, why quality can support a "reasonable" valuation. AVC is the world's largest system-cooling vendor and a core supplier of NVIDIA AI liquid cooling. The track has high certainty, and earnings are truly accelerating rather than being a story: 2026 Q1 EPS alone reached 20.17 (up about 146% YoY), full-year analyst median FY2026E EPS is about 93.75 (FactSet survey raised to 93.75, target price 3,140), and Morgan Stanley further estimates 96.5 in 2026, 145.03 in 2027, and 191.8 in 2028, raising its target price to 3,333. If this growth curve is delivered, the current price implies forward FY2026E PE of about 27–29 times and FY2027E only about 18–19 times. For a company whose earnings more than double in two years, these forward multiples are not expensive. This is the basis for the report placing intrinsic value in the "reasonable" range of 2,300–2,900 and foreign broker target prices at 3,140–3,333: the quality and track deserve the word "reasonable."

  But "good quality" is not the same as "good price," and this is what this question must face honestly. At the current price, TTM EPS of about 61 implies PE of about 42–44 times (StatementDog shows PE is high and above more than 20% of peers), PB about 22–24 times, and cash yield below 1% (about 0.78% based on proposed cash dividend of 21.0). All three are at the company's historical highs. Before the AI trade, AVC's PE was only about 12 times for years; this AI demand cycle pushed the multiple above 40 times, and the PE river chart has also landed in the red zone indicating a high level. The report's reasonable buying ceiling is about 2,100 (corresponding to FY2026E forward PE about 22 times), about 18–22% below the current price. In other words, although the current price is still inside the "reasonable" range, it hugs the upper edge and leaves almost no buffer for "forecast misses."

  Now lay out the math for "fivefold in ten years." From about 2,600 to about 13,000 requires about 17–18% annualized. Break it into two parts:

  • Earnings side: this is AVC's most defensible part. If 2027E EPS of 145 is delivered on schedule, earnings growth alone would push the current price's forward PE down to less than 19 times. The growth engine itself is strong enough, and a fivefold gain driven by long-term EPS compounding is not fantasy.
  • Multiple side: this is the risk. You are starting today from a historical valuation high of 44 times, while AI cooling demand is clearly cyclical. Three reverse signals overlap: ① if Rubin single-piece design advances, it will lower cooling-component ASP per unit; ② entering the Q2 platform transition period, water-cooling plates for new GPU platforms only begin small-volume shipment at the end of the second quarter; ③ order-splitting by peers such as Auras (Auras) + red-ocean competition will squeeze gross margin and share over the long term. If any of these develop, the high 44 times multiple could revert sharply toward historical averages, and multiple compression directly eats away the return contribution from earnings growth. This is the core reason the "fivefold in ten years" hurdle is high.

  The hardest reverse signal is the market voting with its feet. Foreign brokers have raised target prices one after another, but the share price has fallen rather than risen. In early June 2026 it plunged about 5% in one day (historical quotes show a one-day -5.08%, -145 selloff), and foreign investors sold about 8,540 lots net over the latest 20 trading days. The combination of "positive target prices + share price selloff + continuous foreign selling" often means good news has been priced in and smart money is taking profits at high levels. Add that this is a high-beta stock with daily moves often at ±5–9%, and if you buy at the current price, the first thing you must bear is downside volatility under a valuation with no buffer.

  Back to the intent of Baillie's question: from today, can it fivefold in ten years, and is there a margin of safety?

  • Can it fivefold? It cannot be ruled out. The earnings engine is strong, forward valuation (under the delivery assumption) is not absurd, and the long track supports long-term compounding. But it depends on "substantial EPS compounding" and "multiples not compressing sharply" holding simultaneously. The latter condition is tight when starting from 44 times and with cyclical demand.
  • Is there a margin of safety? No. The current price is about 18–22% above the report's reasonable buying ceiling (about 2,100), equivalent to prepaying most of the deposit for optimistic growth in 2026–2027. If Rubin pressures ASP, transition-period shipments are delayed, or order-splitting intensifies, the valuation without a buffer will come under pressure first.

  Final judgment: AVC is a good company that makes one want to hold it long term, but "good company" and "good price" are two different things. At the current price, there is not much cheapness left. What it offers is a premium for certainty, not a margin of safety. For capital that truly wants to hold for ten years, the more rational approach is to pull the entry point back to the lower edge of the reasonable range (around 2,100, corresponding to forward PE of about 22 times), or wait for a cyclical pullback to create a valuation buffer, rather than chase at a historical high of 44 times while foreign investors are taking money off the table. A fivefold gain over ten years is not impossible, but be clear: one must first endure a period of downside risk with no valuation buffer before talking about that fivefold.

  Jun 5, 2026
• QWhy has the market not realized all this yet? Is it because it does not understand, looks down on it, or cannot look far enough? What will become the &quot;narrative inflection point&quot;?3/10

  A

  Bottom line: AVC is not Baillie's typical "great growth stock the market has not yet realized." The long-term AI liquid-cooling theme is not only fully recognized by the market; at stages it has been over-recognized. PE expanded from a historical roughly 12× to 40×+ (current price about 44×, above the peer average), the share price ran from about 600 to a high of 3,010 within one year, foreign brokers raised targets all the way to 3,140 (FactSet consensus) and Morgan Stanley 3,333. The "first year of liquid cooling" narrative is already consensus in Taiwan equities, so this is not a case of "not understanding / looking down on it / not looking far enough." Honestly, if Baillie asks "why has the market not realized this company can be great," the answer is: the market has realized it, and has prepaid most of the deposit for long-term growth.

  The real "perception gap" is not in the long term, but in short-term momentum. The market is fiercely debating whether Q2 is a "transition period" or an "inflection point." The most counterintuitive and intriguing signal right now is a divergence: analyst target prices keep rising, but the share price falls instead, and foreign investors sold about 8,540 lots net over the latest 20 days (CMoney Huang Shih-Ming: target price rises sharply but the stock does not). This shows that the market is not euphoric in the near term; it is cautious. It is voting with its feet on an unresolved question: second-quarter monthly revenue pulled back (April already down 13.24% MoM), and Rubin vapor chambers are rumored to move from two pieces to one piece (pressuring cooling-component unit prices). Is this merely a temporary air pocket during the transition between old and new platforms, or the starting point of structural weakening in growth momentum? This divergence is the stock's real pricing conflict today, and it is the root cause of the 20% drop over 7 days among the three cooling leaders in early May and another selloff in early June.

  Why is this short-term divergence amplified? Because the long-term narrative is already priced in, and valuation has no buffer. When a company's PE-TTM stands at 44×, yield is below 1%, and long-term growth expectations have been fully included in the price, any doubt about "momentum over the next one or two quarters" is amplified into violent volatility. Good news has been overdrawn, while bad news can still kill the valuation. This is the price of "market fully recognizes it": AVC has lost the "undervalued safety cushion" and gained high sensitivity to short-term marginal changes (on June 5 intraday it once fell about 5% from the previous close, then recovered intraday, with large single-day volatility). So rather than saying the market "cannot look far enough," it is more accurate to say the market "has already looked very far, but cannot see the next one or two steps clearly, and is unwilling to pay more."

  The narrative inflection point is two-sided, and it sits directly on actual results over the next 1–2 quarters:

  • Bullish inflection (upward): second-quarter results disprove the fear that the "transition period turns into weakness," monthly revenue returns to sequential growth, and April's MoM decline is proven to be only a platform-transition air pocket rather than a trend; Vera Rubin water-cooling plates move from small volume to ramp; 2027 ASIC cooling contribution materializes as a second growth curve. If any of these are delivered, the "transition vs inflection" debate will tilt bullish, and the suppressed gap between target prices and share price may narrow. The argument that system integrators (owned CDU, manifolds, quick disconnects) are less affected by single-piece design than pure vapor-chamber vendors is also a structurally bullish point.
  • Bearish inflection (downward): Rubin single-piece design truly translates into systematic cooling-component ASP decline, Auras materially takes cold-plate and manifold share, and second-quarter revenue/gross margin (whether it holds around 29%) clearly deteriorates. Then the "transition period" is confirmed as an "inflection point," the narrative shifts from "quarterly growth" to "momentum peaking + red-ocean competition," and the 44× PE-TTM faces a double hit from valuation and earnings. If high-valuation AI stocks in Taiwan suffer systematic derating as well, this high-beta stock lacking dividend and low-valuation buffers will be among the first hit.

  One-sentence close: Baillie's framework asks "why has the market not realized it." The honest answer for AVC is: the market has already realized the long-term theme, and has even over-priced it; there is no cognitive dividend from "not understanding / looking down on it / not looking far enough." The only real perception gap today is the short-term momentum game of "transition vs inflection," expressed in the divergence of "target prices up, share price down, foreign investors selling." This is not the story of a "great growth stock the market has not discovered." It is the story of "growth already fully believed, with the only dispute left being cadence and valuation buffer." The inflection point is not in narrative discovery, but in second-quarter results revealing whether the issue is "temporary vs structural." Before that is revealed, a valuation with no buffer will only make volatility more violent.

  Jun 5, 2026