ABF Manufacturer CAPEX plans & Machinery We have established that ABF substrate floor space is fundamentally non-fungible. The cutting-edge ABF substrates required for top-tier chips (like NVIDIA’s Blackwell and upcoming Rubin platforms) feature large package sizes, 16–20+ layers, and line widths as narrow as 15 µm. Manufacturing these requires highly specialized, high-end vacuum lamination, high-aspect-ratio laser drilling, and advanced inline optical inspection tools. Lead times for this equipment are currently sitting at an extendedl 18 to 24 months. Eternal Precision Mechanics 長廣 7795.TW is an equipment manufacturer which holds an effective monopoly at ≈ 95% global share in mid-high end vacuum laminators. ABF has to be laminated onto the substrate with extremely tight specifications and tolerances. Each build up layer (16-20 presently) creates another lamination step or process so EPM benefits form the increasing size, complexity and layer count of cutting edge ABF architectures. How Eternal reportedly allocates scarce equipment According to Goldman Sachs priority goes to customers with: - Larger scale. - Longer-standing relationships. - Secured ABF film supply. Immediate positive implications for Ibiden and Unimicron. → What can we take-away from the CAPEX estimates in Chart 1 ? (This does not indicate equipment spending alone or machinery allocations). Kinsus 景碩 3189.TW - gradual increase Unimicron 欣興 3037.TW - strong, sustained increase Nan Ya PCB 南亞電 8046.TW - small increase Ibiden イビデン 4062.JP - peaks in 2027 Semco 삼성전기 009150.KW - largest spending contributor AT&S AT&S Austria Technologie ATS.VI - declines by 2028 Unimicron and Samsung keep increasing expenditure, while Ibiden and AT&S reduce spending in 2028. Table 2 outlines spending vs confirmed equipment orders (that are publicly disclosed). Nikko machinery refers to EPM's Japanese subsidiary so can be thought of as the same entity. 🚨 In addition to the NT 1.76B announced by unimicron to date, I've discovered that today, October 27, Unimicron authorized a further NT$50.9 billion of long-lead equipment procurement. That is substantial evidence of advance procurement planning, but the announcement does not identify suppliers or delivery years, and execution remains adjustable. Source: m.moneydj.com/f1a.aspx?a=dd1… Ok, so what about SEMCO, they are the largest contributor according to the image 1, how are they equipping their substrate expansion plans ? I address this in my next post.

ABF Manufacturer CAPEX plans & Machinery We have established that ABF substrate floor space is fundamentally non-fungible. The cutting-edge ABF substrates required for top-tier chips (like NVIDIA’s Blackwell and upcoming Rubin platforms) feature large package sizes, 16–20+ layers, and line widths as narrow as 15 µm. Manufacturing these requires highly specialized, high-end vacuum lamination, high-aspect-ratio laser drilling, and advanced inline optical inspection tools. Lead times for this equipment are currently sitting at an extendedl 18 to 24 months. Eternal Precision Mechanics 長廣 7795.TW is an equipment manufacturer which holds an effective monopoly at ≈ 95% global share in mid-high end vacuum laminators. ABF has to be laminated onto the substrate with extremely tight specifications and tolerances. Each build up layer (16-20 presently) creates another lamination step or process so EPM benefits form the increasing size, complexity and layer count of cutting edge ABF architectures. How Eternal reportedly allocates scarce equipment According to Goldman Sachs priority goes to customers with: - Larger scale. - Longer-standing relationships. - Secured ABF film supply. Immediate positive implications for Ibiden and Unimicron. → What can we take-away from the CAPEX estimates in Chart 1 ? (This does not indicate equipment spending alone or machinery allocations). Kinsus 景碩 3189.TW - gradual increase Unimicron 欣興 3037.TW - strong, sustained increase Nan Ya PCB 南亞電 8046.TW - small increase Ibiden イビデン 4062.JP - peaks in 2027 Semco 삼성전기 009150.KW - largest spending contributor AT&S AT&S Austria Technologie ATS.VI - declines by 2028 Unimicron and Samsung keep increasing expenditure, while Ibiden and AT&S reduce spending in 2028. Table 2 outlines spending vs confirmed equipment orders (that are publicly disclosed). Nikko machinery refers to EPM's Japanese subsidiary so can be thought of as the same entity. 🚨 In addition to the NT 1.76B announced by unimicron to date, I've discovered that today, October 27, Unimicron authorized a further NT$50.9 billion of long-lead equipment procurement. That is substantial evidence of advance procurement planning, but the announcement does not identify suppliers or delivery years, and execution remains adjustable. Source: m.moneydj.com/f1a.aspx?a=dd1… Ok, so what about SEMCO, they are the largest contributor according to the image 1, how are they equipping their substrate expansion plans ? I address this in my next post.

Why is Unimicron 欣興 3037.TW our top ABF pick ? An illustrative Economic Concept & Example: Ther indexed ASP chart shows 2027 roughly: - Unimicron ≈ 3.6x 2018 - Kinsus ≈ 2.4x - NY PCB ≈ 1.8x This does not mean Unimicron charges twice what Nan Ya or Kinsus does for ABF substrates, It tells us that the product mix is different and that Unimicron is increasingly selling a different substrate. Unimicron appears to be getting more exposure to: - very large body AI substrates / larger - higher layer counts / more layers - advanced GPU/ASIC packages / more complex - higher yield burden - harder power delivery architecture - better T-glass allocation (according to JPM) - customer reserved premium capacity - more expensive equipment So Unimicron is selling the same product as the other 2 Taiwanese players- ABF substrate, but is receiving a radically different dollar value per unit of effective capacity. This matters A LOT. And this is a concept that carries through our passive component ecosystem research- It's another form of economic density. For ABF: Substrate volume increases, but the value per unit of substrate capacity increases even faster. Nomura adds an interesting dimension to the research: → Advanced substrate capacity is NOT economically fungible We must distinguish between different tiers of ABF quality but ALSO newer architectures: - large reticle size 2.5D substrates - EMIB-T - embedded silicon / TSV structure - two sided power delivery networks - specialized high end equipment A conventional ABF production line cannot necessarily be redirected into these newer ABF architectures and products simply because the nominal capacity exists. This means (and this is important) that industry ABF capacity is becoming an increasingly misleading indicator and that we must work much harder to uncover the underlying dynamics including who is capable of producing what and at what scale. We must instead be thinking in architecture-specific qualified capacity, which is constrained by: - T-glass allocation - ABF CCL alloaction - vacuum laminator availability - increasingly large substrate bodies - more layers - yield losses - customer qualification - specialized tooling This results in an increasingly widening gap between tier 1 installed capacity and tier 2 nominal capacity which is important for our tier 1 technological leaders like Unimicron and Ibiden and underscores why I write about choosing the tier 1 technological leaders as primary positions- Their moat is much larger. The result of this is that manufacturers with existing tier 1 production capacity will be rewarded, with immediate positive implications for Ibiden & Unimicron. Outer Ninth Research ABF Timing base thesis: ▶️ ▶️ ▶️ 2027 is where ABF moves decisively from tight to allocation constrained, and 2028 gets worse rather than better. Our Investment research throughout the entire passive component mapped ecosytstem must answer: → Which company owns the highest-value qualified effective capacity and can actually monetize the shortage without paying too much for future growth ? Unimicron and other ABF producers can announce a lot of capacity today but a very large portion of it does NOTHING for 2027. The demand clock is accelerating now and the supply clock comes in later in 2028. From this we can also infer (IMPORTANT): 2026 price increases are largely a mix of raw material inflation + product mix + selective local tightness 2027 onwards should increasingly be: genuine scarcity capture + allocation + pricing power This matters a lot because a cost driven 15% ASP increase and a scarcity driven 15% ASP increase have completely different margin implications- the first is more of a cost pass through and the second passes through more directly to margins. The latter is what can produce the very large earnings revisions which we are starting to see 👀 Unimicron 欣興 3037.TW Ibiden イビデン 4062.JP Nan Ya PCB 南亞電 8046.TW Kinsus 景碩 3189.TW #ABF #欣興 #PCB #AI $NVDA

ABF Substrate Utilization to Breach 100% in 2027: J.P. Morgan Maps AI Supply Deficits Across Here is another banger post by @OptionKing666 and the contents from JPM are what I've telling you guys this week and all summer for that matter. It's almost as if MS & JPM have been reading Outer Ninth Research 🧐 These components are all interrelated and the backbone of the AI supply chain: "Structural shortages are opening simultaneously across four interconnected supply chains: carrier substrates, high-speed copper clad laminates (CCL), multi-layer ceramic capacitors (MLCCs), and semiconductor test equipment." My research not only focuses on these bottlenecks and their physical, technological and structural importance in the supply chain but also rigorous financial analysis and modelling. What moves stock prices and which and these companies can be attributed to these factors. Margin expansion, revenue and EPS expansion, positive revision suprises AND ultimately cash being returned to shareholders. This article addresses some of this: ------------------------------------------------------------ J.P. Morgan projects that global ABF substrate industry utilization will surpass 100% by 2027, driven by annual demand growth near 40% against supply additions of only ~10%. Structural shortages are opening simultaneously across four interconnected supply chains: carrier substrates, high-speed copper clad laminates (CCL), multi-layer ceramic capacitors (MLCCs), and semiconductor test equipment. Table of Contents: -Why Four Supply Chains Are Facing Concurrent Shortages -Substrate Demand: Die Sizes Expand, ABF Area Doubles in Two Years -Substrate Supply: 10% Annual Capacity Growth Pushes 2027 Utilization Past 100% -T-Glass Allocation and LTAs: How Shortages Translate into Unimicron's Margins -Glass Core and CoWoP: Two Potential Technology Headwinds to ABF Demand -What’s New Compared to the July ABF Reports -High-Speed CCL: A Threefold Market Expansion in Two Years with M8 Dominating -Yield Degradation and PTFE: Supply Bottlenecks and Overstated Substitution Risk -Glass Yarn Spillover and Substrate-Grade CCL: Elite Material's New Growth Horizons -MLCCs: AI Servers Surge >150% Annually, Crowding Out Commodity Capacity -Yageo: Low-End MLCC Price Hikes and Pricing Power in Tantalum and Resistors -Test Equipment: GPU Test Times Double, Tripling Chroma's Three Core Segments --Valuation and Price Targets: Pricing the Four Sectors Risks, Counter-Theses, and Key Metrics to Watch Over the Next Four Quarters Why Four Supply Chains Are Facing Concurrent Shortages With every generational transition in AI accelerators, die footprint, package substrate layer count, and rack-level thermal design power (TDP) scale in tandem. Nvidia's Hopper utilizes an ABF substrate body size of 3,190 mm² across 12 layers. Blackwell expands to 5,913 mm² and 14 layers, Rubin reaches 8,051 mm² and 18 layers, and Rubin Ultra doubles again to 16,102 mm² and 20 layers. Beyond that, J.P. Morgan estimates the Feynman generation will scale to approximately 18,000 mm² across 24 layers. Total substrate area per chip (footprint multiplied by layer count) escalates from 38,280 mm² on Hopper to 432,000 mm² on Feynman—more than a tenfold expansion. This physical scaling reverberates far beyond the substrate itself. Larger die assemblies require denser interconnect arrays, forcing compute trays, switch trays, and backplanes to adopt higher layer counts and lower-loss copper clad laminates (CCL). Meanwhile, power delivery demands surge from 700W on H100 to 2,300W on Rubin, requiring significantly more high-capacitance MLCCs per socket, extending system-level test times, and increasing the dollar content of power delivery networks. J.P. Morgan analyzes these four supply chains within a unified framework because they share a single independent variable: the physical specifications of individual AI processors. Across all four verticals, demand growth vastly outpaces capacity additions. ABF demand is forecast to compound at annual rates of 34%, 39%, and 40% from 2026 to 2028, while industry capacity expands by only 9.3%, 10.2%, and 12.1%. High-speed CCL demand is set to expand 63% in 2026 against a 19% capacity increase. Commodity MLCC supply will flip into deficit by 2027 as high-end AI components cannibalize production lines, and GPU system-level test (SLT) duration lengthens by roughly 60 minutes with each architectural cycle. The pivotal investment question is which suppliers can translate these structural shortages into pricing power and sustained gross margin expansion. Substrate Demand: Die Sizes Expand, ABF Area Doubles in Two Years J.P. Morgan models global ABF substrate demand by multiplying die footprint by substrate layer count and applying a complexity multiplier, denominated in billion mm². Under this methodology, worldwide ABF demand totaled 9,470 in 2024 and 11,308 in 2025, and is projected to reach 15,164 in 2026, 21,150 in 2027, and 29,565 in 2028. This represents annual growth of 19%, 34%, 39%, and 40%, respectively—nearly doubling demand in two years and scaling 2.6x by 2028 relative to 2025 levels. Virtually all incremental volume originates from AI servers and high-speed switches. This segment accounted for 2,448 billion mm² in 2024 and 4,634 in 2025, and is forecast to hit 8,335 in 2026, 13,726 in 2027, and 21,479 in 2028, expanding from 26% of total global ABF consumption to 73%. Over the same period, general-purpose server demand climbs modestly from 3,210 to 5,032, while consumer electronics demand contracts from 3,812 to 2,941 before staging a muted recovery to 3,055. By 2028, ABF will have transitioned entirely from a consumer PC cyclical into an AI-driven compute material. The table below breaks down AI server ABF demand by customer, highlighting the extreme concentration of incremental volume. AI Server ABF Demand by Customer (Complexity-Adjusted, Billion mm², J.P. Morgan Estimates) | Customer | 2024 | 2025 | 2026e | 2027e | 2028e | |---|---|---|---|---|---| | Nvidia | 1,743 | 3,655 | 5,981 | 8,299 | 13,874 | | Google | 140 | 215 | 1,087 | 2,941 | 3,505 | | AWS | 164 | 384 | 629 | 1,344 | 2,533 | | Apple ASIC | — | — | 197 | 385 | 607 | Nvidia remains the primary demand driver, projected to command over 60% of total AI server ABF consumption by 2028. However, custom silicon demand is accelerating rapidly. Google TPU ABF demand is forecast to surge 405% in 2026 and another 171% in 2027, driven by TPU v8 (5,250 mm², 24 layers) and v9 (13,685 mm², 24 layers). AWS Trainium 4 (modeled at 14,400 mm², 24 layers) more than doubles its substrate consumption in 2027. Custom ASICs now match merchant GPUs in driving substrate area requirements. (Reference Image 1 HERE) The physical catalyst for this surface area expansion is the scaling of CoWoS interposers. J.P. Morgan establishes a direct correlation between interposer size and substrate footprint: a 1.5x reticle interposer requires a 2,000–2,500 mm² substrate; 2x reticle requires 3,000–3,300 mm²; 3.5x reticle requires 5,000–6,000 mm²; 5.5x reticle requires 7,500–8,500 mm²; 9.5x reticle requires 13,500–15,000 mm²; and a 12x reticle design (co-packaging optical engines directly on the GPU substrate) demands 22,500–27,500 mm². Substrate area scales in discrete step-functions, and with each tier, net panel-level manufacturing yields drop significantly. Substrate Supply: 10% Annual Capacity Growth Pushes 2027 Utilization Past 100% Supply-side discipline remains pronounced. J.P. Morgan calculates global substrate manufacturing capacity using a standardized 30mm × 30mm, 6-layer equivalent unit: capacity totals 373 in 2025, 407 in 2026, 449 in 2027, and 503 in 2028, reflecting modest annual growth of 9.3%, 10.2%, and 12.1%. Net capacity additions stand at 35, 42, and 54 units, respectively, only matching the 2022 expansion peak of 46 units by 2027–2028. Because the 2022 capex wave triggered severe pricing erosion across 2023–2025, substrate vendors remain cautious regarding greenfield expansions. Global Substrate Capacity by Vendor (30mm × 30mm, 6-Layer Equivalent, J.P. Morgan Estimates) | Vendor | 2025 | 2026e | 2027e | 2028e | |---|---|---|---|---| | Unimicron | 79 | 92 | 102 | 117 | | Ibiden | 72 | 82 | 95 | 107 | | Nan Ya PCB | 48 | 50 | 54 | 57 | | Samsung Electro-Mechanics | 32 | 33 | 36 | 40 | | Shinko Electric | 49 | 50 | 53 | 55 | | Kinsus Interconnect Technology | 27 | 28 | 31 | 37 | | AT&S | 40 | 42 | 45 | 49 | | Others | 26 | 30 | 34 | 41 | | Total | 373 | 407 | 449 | 503 | Unimicron and Ibiden account for over half of all net capacity additions between 2025 and 2028, while tier-2 suppliers add only low-single-digit capacity. With demand expanding at 34%–40% annually against 9%–12% supply additions, the supply-demand deficit widens steadily. This drives industry-wide utilization from ~65% in 2024 to ~72% in 2025, ~86% in 2026, over 100% in 2027, and near 130% by 2028. Utilization exceeding 100% represents a physical impossibility; in practice, it signals unfulfillable demand where non-tier-1 customer orders face extended deferrals or rationing. (Reference Image 2 HERE) Upstream equipment data corroborates this multi-year deficit. Global demand for substrate stepper lithography tools (indexed to 2023 = 100) fell to ~82 in 2024 and ~40 in 2025, before rebounding in 2026. Because stepper procurement lags order intake, manufacturers spent 2025 absorbing previous tool installations. Greenfield tool orders only begin recovering in 2026, meaning this equipment cannot convert into qualified commercial output until 2027 or later—locking in the utilization bottleneck. T-Glass Allocation and LTAs: How Shortages Translate into Unimicron's Margins Beyond tool availability, raw materials represent an acute bottleneck. AI-grade substrates require low coefficient of thermal expansion (CTE) T-glass fabric, primarily sourced from Nittobo (Nitto Boseki). J.P. Morgan estimates Nittobo’s T-glass capacity index (FY23 = 100) will reach ~400 by the end of FY28—a fourfold expansion. Sales volumes are projected to grow ~33% YoY in FY26 and ~37% in FY27, accompanied by price hikes of ~28% in FY26 and ~7% in FY27. Allocation is the primary profit arbiter. J.P. Morgan estimates that Unimicron receives 35%–40% of Nittobo’s T-glass allocation, while Ibiden secures 30%–35%. Tier-2 competitors collectively split the remaining 25%–33%. Upstream material allocation dictates who can process high-layer-count, large-body AI substrates. Even if secondary vendors like Nan Ya PCB or Kinsus possess nominal floor space, insufficient T-glass access prevents them from running high-end lines at full capacity. This allocation structure explains why sell-side margin revisions are concentrated almost entirely in the top two suppliers. Material scarcity has also transformed commercial contract structures. J.P. Morgan categorizes long-term agreements (LTAs) into five types: customer prepayments, customer-purchased equipment, customer-funded capacity, capacity reservation fees, and super-expedited orders (hot lots). Evaluated across financial returns, operational flexibility, and contract irrevocability, capacity reservation fees emerge as the superior model (scoring 5/5, 4/5, and 4/5, respectively). Customer-funded capex offers maximum irrevocability but minimum operational flexibility, whereas prepayments offer the weakest financial upside. Under a reservation fee model, customers pay non-refundable fees to secure forward capacity allocations without fixing specific order pricing, providing suppliers upfront liquidity while preserving spot pricing flexibility. Broad adoption of these terms extends substrate revenue visibility from quarters to years. Pricing power directly drives operating margins. J.P. Morgan models next-generation AI accelerators (GPUs, TPUs) securing blended ABF ASP premiums exceeding 50% over prior architectures due to higher layer counts, larger footprints, and stricter co-packaging tolerances. Benchmarking Unimicron's 2018 unit ASP at 1.0, its pricing index peaked at ~3.3 in 2022, trough-tested at 1.9–2.0 across 2023–2025, and is projected to rebound to ~2.6 in 2026 and ~3.6 in 2027, exceeding previous cycle highs. By comparison, Nan Ya PCB is projected to reach only ~1.8 in 2027, while Kinsus reaches ~2.4—a divergence reflecting product mix and T-glass access. (Reference Image 3 HERE) Simultaneous increases in ASP and fab utilization generate significant operating leverage. J.P. Morgan projects Unimicron's ABF segment operating margin to climb from ~28% in 1Q25 to ~30% in 1Q26, ~38% in 2Q26, ~53% in 1Q27, and ~62% in 4Q27. Consolidated corporate operating margin expands from ~13% in 1Q25 to ~31% in 3Q26 and ~47% by 4Q27, with the HDI division holding steady at ~15%. Consequently, Unimicron's net profit is forecast to surge 357% in 2026 and 132% in 2027, as consolidated gross margins expand from 13.9% in 2025 to 28.0% in 2026 and 43.0% in 2027. (Reference Image 4 HERE) Nvidia $NVDA Google $GOOG Amazon $AMZN Apple $AAPL Unimicron 欣興 3037.TW Ibiden イビデン 4062.JP || $IBIDY Nanya PCB 南電 8046.TW Samsung Electro-Mechanics 삼성전기 009150.KW Kinsus 景碩 3189.TW AT&SAustria Technologie & Systemtechnik Aktiengesellschaft $ATS.VI Nittobo 日東紡 3110.JP Full substack article: 404kresearch.substack.com/p/…

Summary of Unimicron (3037.TT) – Initiate BUY, TP NT$1,820
77% EPS CAGR amid AI ABF supercycle • AI ABF substrate TAM set to hit US$13bn / $18bn in 2027E/28E (+100%/41% YoY)
• Unimicron poised to capture >30% share
• ABF capacity +18%/35% in 2026E/27E → ABF revenue +66%/90% to NT$105bn/199bn
• Supply shortage expected to persist into 2029 (18%/16% shortage in 27/28)
• Product mix shift (high-end ABF + mSAP + LPDDR) drives GPM to 28.8% / 41.7% / 44.6%
• EPS: NT$28.6 / 60.8 / 89.3 (vs consensus 25/55/80)
• TP based on 30x 2027E P/E #Unimicron #ABF #AI #Semiconductors #TaiwanTech

CORNING $GLW Q3’26 EARNINGS HIGHLIGHTS 🔹 Revenue: $5.08B (Est. $5.02B) 🟢; +19% YoY 🔹 Adj. EPS: $0.91 (Est. $0.88) 🟢; +36% YoY 🔹 Core Gross Margin: 40.5%; +150 bps YoY 🔹 Core Operating Margin: 22.7%; +310 bps YoY Q4 Guide: 🔹 Revenue: $5.2B-$5.3B (Est. $5.21B) 🟡 🔹 Adj. EPS: $0.96-$1.00 (Est. $0.94) 🟢 Segment Net Revenue: 🔹 Optical Communications: $2.4B; +44% YoY 🔹 Glass Innovations: $1.5B; -3% YoY 🔹 Automotive: $470M; +4% YoY 🔹 Solar: $436M; +37% YoY 🔹 Life Sciences & Emerging Growth: $290M; +1% YoY Announcements: 🔹 Secured four long-term, multibillion-dollar customer agreements, including with Verizon and AT&T 🔹 Solar revenue stream progressing toward more than $3B Other Q3 Metrics: 🔹 GAAP Gross Margin: 37.3% 🔹 GAAP Operating Margin: 18.8% 🔹 Operating Cash Flow: $1.1B 🔹 Adjusted Free Cash Flow: $661M Comments: 🔸 “faster adoption of our innovations helped us double our Gen AI sales year over year in the quarter, outpacing GPU unit growth.” 🔸 “we expect our outstanding performance to continue.”

早安！10/27 外電綜合整理 - 2327國巨: 美系重申正向 美系大行預期3Q 毛利率40.4%、OPM 29.1%，高出市場共識。展望4Q，預期漲價與AI需求預期持續推動營收成長及利潤率擴張。在即將到來的法說會上，券商會關注非AI市場的需求能見度，還有各產品漲價及供需狀況，重申正向。 - 2345智邦: 美系重申正向 將辦理現金增資，美系大行表示稀釋幅度最高2%，影響可控。重申正向。 - 2360致茂：美系重申正向 美系大行預期半導體及光子技術業務持續帶動2H26成長；預期明年有將是強勁成長的一年。預期3Q毛利率60.5%，季持平；OPM季增至39.3%。重申正向。 - TPCA 2026: 美系大行表示CCL是否會用PTFE會在明年中決定，在此同時也在試其他方法如氫樹脂（HC）基材、HC＋PTFE混合材料，以及苯並環丁烯（BCB）基材的CCL。 此外，觀察薄型E-glass仍是短缺，將既有產線轉換為LDK1／2及T-glass玻纖布，以滿足強勁的AI需求。 最後看到ABF的壓合設備交期至少兩年，意即全新建廠的ABF載板產能，最早只能從2029年開始逐步放量。明年載板尺寸又放大，券商持續正向看待載板產業。總結：券商對覆蓋裡的台燿、台光電、金居、欣興、南電、金像電等持正向看法不變。 #每日外電

Nan Ya Plastics 南亞 1303.TW As a follow-up to the "mini article" that I wrote at the bottom of this post. I have not yet performed detailed financial analysis on the company but I did uncover some REALLY interesting information on the corporate structuring of the Nan Ya trio of companies. → I want to quickly point out that I see barely any mentions if any at all about this company which for me is bullish. Nan Ya Plastics is becoming a lot more interesting 🚨 Nan Ya Plastics 南亞 1303.TW currently owns: - ≈ 67% of Nan Ya PCB 南亞 8046.TW - ≈ 30% of Nanya technology 南亞科 2408.TW - 100% of PFG Fiber Glass, which makes electronic grade glass yarn - 100% owned electronic materials subsidiaries producing combinations of CCL, copper foil, fiberglass fabric and epoxy So 1303 owns/controls: - epoxy/resins - glass yarn/fabric - copper foil - prepreg/CCL 67% of Nan Ya PCB - ABF substrates - PCB - BT 30% of Nanya Technology - DRAM Materials + substrates/PCB + memory all inside one economic group with vertical integration- that's powerful. Vertical Integration here is VERY important: Glass shortages → they make glass Copper Foil shortages → they make CF Resin/formulation scarcity → they formulate resin CCL price inflation → they sell CCL ABF/BT substrate shortage → 67% owned nan Ya PCB sells substrates Memory recovery → 30% owned Nanya Tech. participates That's a remarkable amount of the value chain under one corporate umbrella. They have constant access to most of the scarce upstream materials and the drastic price increases pass right through directly to their margin. Nan Ya's electronic material segment reached roughly 58% of revenue in Q3 and OPM reached 30.9%. It is already mass producing M7/M8, has M9/M10 under qualification, and is developing its own HC resin as well as HVLP 4/5/6 copper foil 👀 👀 👀 1303 you have my attention ! The remaining question becomes- How does Nan Ya Plastics stand up to Outer Ninth's rigorous financial analysis ? We intend to answer that question later this week.

Elon Musk says “robot RAM needs will exceed everything else on Earth” and believes superintelligence will eventually drive even more demand through deep space compute. To put that into perspective $TSLA producing 10M Optimus robots a year would require ~720 petabytes of memory at 72GB per robot. Pretty wild for $MU and $SKHY as AI expands beyond data centers although robots use LPDDR rather than HBM so this could become a much bigger volume story.

南亞搶先推出M11材料案子 2027年在CSP客戶斬獲良好 M9的案子在自研晶片也有不錯的進度 #TPCA

Nittobo's Yuhiro Mikami Interview 📰 Nitto Boseki Co. 日東紡 3110.JP Company Positioning - Global leading supplier of high-performance semiconductor packaging substrates and high-end glass cloth for electronic equipment motherboards. - Special glass (NE/T glass, etc.) holds overwhelming market share and industry standard status. Current Market Conditions - AI server demand remains robust, with excellent demand for T glass (low CTE, warp-resistant, used for GPU/CPU) and NE/NER glass (low dielectric/low loss, used for high-speed signals). - Supply is severely tight; the company is working to improve yields to increase shipments. Demand Structure - CCL base materials for thin FCCSP and FCBGA are mainly from Japanese manufacturers. - CSP for memory/modules is stronger from Korean manufacturers. Investment Plans - Continuing large-scale investments domestically and internationally, advancing according to schedule. - Original yarn expansion: Centering on Taiwan to add melting furnaces; T glass yarn expected to contribute to mass production starting in fiscal 2027, aiming to ship as early as possible. - Domestic plans to decommission E glass melting furnaces to expand special glass production capacity. - Glass cloth expansion, with weaving processes in collaboration with Nan Ya Plastics. - Due to upward demand revisions, investment amount revised upward to 120 billion yen by fiscal 2027 (an increase of 40 billion yen from original), targeting T glass capacity at 3 times that of fiscal 2025. Competition and Differentiation - Known competitors entering the T glass market, potentially leading to a slight drop in market share, but revenue scale continues to grow. - Emphasizing over 30 years of proven track record in stably supplying high-quality T glass, upholding "Nittobo quality." Future Demand Outlook - Inference AI driving an increase in CPU volumes, with growth in dedicated ASICs/CPUs from service providers. - Packaging substrates becoming larger in size and more multi-layered at the core, with core thickness potentially doubling to 2.4–3 mm. - Special glass demand expected to remain strong in the short term; currently evaluating demand for 2029–2030 and the next wave of investment. Next-Generation Products - Recommending NER/NEZ (excellent low-Dk/Df), with partial mass production already underway. - Next-next-generation "DX II" at quartz glass level, targeting sample shipments around fiscal 2028. - Next-generation T glass "Vlex" under development (low CTE + low loss), expected to officially ship around 2028. - Emphasizing the need to pair with new resins and collaborate on substrate design to deliver high value-added products. 🔽 🔽 🔽 The interview directly confirms several of our assumptions: - T-glass and NE/NER demand remains extremely tight - Nittobo is trying to increase output through capacity additions AND yield improvement. - Taiwan glass yarn melting capacity is being expanded - E-glass furnaces in Japan are being retired/reallocated toward specialty glass (scarcity cascades 👀) - FY 2024-27 total CAPEX was raised by ¥40B to ≈120B - T-glass capacity is targeted at roughly 3x FY25 levels by FY27 - Next gen VLEX T-glass is targeted around 2028 - next gen low loss products include NER/NEZ and later DX II Key Takeaway: Even though Nittobo is undertaking a ¥120B four year CAPEX program, demand has repeatedly been revised upwards faster than their expansion plans. Another theme we are seeing throughout the passive component ecosystem- demand is increasing faster than supply, even with aggressive capacity expansion.

Nittobo 日東紡 3110.JP This is a long post but PACKED full of useful info. My comments are marked by ▶️ These are initial research tidbits and breadcrumbs for you guys. - Read to the end, there's a mini article down there 🙂 I cover a lot in this piece from Nittobo and glass fiber to CCL, materials and numerous companies. -------------------------------------------------------- "And the price hike rate for E-Glass, which isn't even top-grade, is insane." ▶️ Refer to the attached table from Citi. From Dec 2025 to Oct 2026 - 1080 cloth: Rmb 5.0 → Rmb 18.3/m ≈ +266% - 2116: Rmb 4.9 → Rmb 17.7 ≈ +261% - 7628: Rmb 4.6 → Rmb 13.1 ≈ +185% ▶️ This explains why Panasonic could issue a 200% price increase on selected conventional #ccl The upstream input has effectively tripled or quadrupled in the thinnest fabric categories. Thin glass fiber cloth and CCL are facing shortages again, with prices expected to rise by 15% next month. South Asia Securities bluntly states that the market has "no other options." A shortage of upstream raw materials is triggering a new wave of price increases, and the impact is extending to copper-clad laminates (CCL). According to supply chain sources, price hikes for glass fiber cloth and CCL are imminent, with the increase estimated at around 10-15%, set to take effect from November, though the actual hike will vary by product specs and suppliers. In particular, supply of thin glass fiber fabrics like 2116, 1080, and 1078 is especially tight. The expansion speed of looms is hitting limits, making it the biggest bottleneck in resolving the short-term supply shortage. Supply chain analysis points to supply shortages as the main driver of price increases. As manufacturers shift existing production capacity toward high-end products, supply of standard glass fiber fabrics for general electronics is getting squeezed, even as demand persists, leaving shortages of certain thin glass fiber fabrics ongoing. ▶️ Premium AI demand is distorting the entire fiberglass allocation system, including ordinary E-glass. This is the scarcity cascade bottleneck dynamics that I've been talking to you guys about since summer. Production capacity is re-routed to higher end products, so we get a decreased output of the lower grade commodity products which creates price inflation in those lower grade products as they become less available AKA supply down while demand up. Nanya Plastics says the market is currently in short supply, with many competitors pivoting strategies toward high-end product lines. Yet demand for general-purpose products remains high, and supply and product lineups aren't keeping up. The shift to high-end is clear, but shortages of general-purpose items persist, and the supply-demand imbalance hasn't been resolved. Nan Ya notes that the shortages have also shifted downstream customers' purchasing strategies. In the early stages of price hikes, customers tended to partner with suppliers offering full product lineups and diverse specs, but now with supply even tighter, it's a "no choice" situation, where securing raw materials and meeting delivery deadlines has become the focus of procurement. ▶️ I spoke about this in my note about Kinsus: "Material allocation itself has become strategic." This is an important industry read-through VERY important heading into 2027 that companies have upstream allocation so they can operate at full capacity. This is exactly why we have seen Jenson locking down upstream material supply with their announcement this week. On whether capacity expansions by major mainland China glass fiber maker China Jushi can ease the shortages, industry insiders say Jushi's new capacity is mainly geared toward 7628 heavy-weight fabric, slated for mass production in 2027. This differs from the thin fabrics currently in shortest supply. So even as new capacity comes online, the relief for thin fabrics like 2116, 1080, and 1078 will remain relatively limited. ▶️ Without going too much into it, day 1 of research has materially increased confidence in Nittobo 👀 👀 👀 On pricing, supply chain surveys show that major mainland CCL maker Kingboard has raised prices 12 times this year alone, and the price of 1080-spec E-glass fiber cloth has surged about 300% from low to high, marking an astonishing increase. ▶️ Addressed this, see attached Citi table Industry sources say abnormal shortages have hit certain glass fiber cloth specs, leading the market into irregular pricing via unusual distribution channels. The root bottleneck lies in the loom supply shortage. Supply chain estimates put global annual production capacity for related looms at about 2,400 units, potentially rising to 3,600 by 2027. However, actual production volumes and delivery increases remain unclear. After equipment installation, shifting to effective supply requires setup and mass-production ramp-up. Supply chain assessments suggest that depending on equipment deliveries, new capacity startups, and product conversion progress, the pressure from shortages could gradually ease from late 2027 into 2028. But with demand for high-end applications continuing to grow, even increased supply may leave the market tight, and rising material costs could keep getting passed on to downstream CCL and PCB makers. ▶️▶️▶️ This is IMPORTANT so i'll leave you with this- The margin outcome differs hugely on vertical position If fabric cost rises 200%, a weak laminate supplier may simply pass it through and preserve margin. A strong vertically integrated supplier may additionally make money inside the upstream material itself. This is why simply tracking the CCL ASP is insufficient, instead we must ask how much of the ASP increase becomes gross profit rather than simply compensating for raw material inflation. Using Kingboard Laminates 建滔積層板 01888.HK as a case study- The company explicitly says electronic fiberglass fabric shortages forced some competitors to reduce production, while their vertically integrated supply chain allowed them to maintain near-full utilization, raise prices and win new customer qualifications. Kingboard controls important parts of: - glass yarn -glass fabric - resin/materials - CCL This gives them upstream margin + laminate pricing power + assured supply This is why we see some analysts modelling much higher 2026 H2 margins for Kingboard. Kingboard itself says the laminate industry remains tight and that prices of fabric, yarn, copper foil and laminates continued to surge after June. I spoke about Nan Ya Plastics vertical integration in my Nan Ya business posts, but I still need to really dig into the financial structure. When including Nan Ya PCB the integrated chain is essentially: Chemical feedstocks ↓ custom resin systems ↓ glass yarn/fabric ↓ copper foil ↓ CCL/prepreg ↓ PCB ↓ BT/PP/ABF substrate This solves a remarkable amount of the supply chain under pressure Nan Ya Plastics 南亞 1303.TW Nan Ya PCB 南電 8046.TW However, I have yet to research internal transfer economics: - how much 8046 buys from 1303 - is pricing market based - is scarce material preferentially allocated internally - do upstream margins remain primarily at 1303 - does 8046 gain through lower costs or simply through supply assurance - how much consolidated AI economics ultimately pass through to 1303 shareholders Day 1 & 2 of research did not allow me to get into these details but I will do in the coming days. Meanwhile if you are interested in this, check out the 2 pieces on "Nan Ya Business" Part 1: xitter.cf/outerninth/status/2109… Part 2: xitter.cf/outerninth/status/2110… #ccl #PCB #日東紡 #建滔集团 #Tglass #Qglass

ＪＸ金属は、スマートフォンやウェアラブル端末等に使われるフレキシブルプリント基板向けに、銅原料の100％リサイクル化を実現した圧延銅箔の本格販売を開始しました。 独自の技術により、従来品と変わらない品質での安定生産を実現。資源循環とお客様のサステナビリティ推進に貢献します。 jx-nmm.com/newsrelease/2026/…

ABF substrate CCL prices are reaching ~US$1,500 per full sheet compared to US$180 for M8 laminate as end customers secure dedicated multi-year capacity agreements through 2027–2030. Sell-side consensus across House A, House B, and House F confirms that structural AI packaging shortages are driving direct customer subsidies, prepayments, and capacity bookings. House A estimates ABF substrate CCL gross margins exceed 50% and could reach 70%+, driven by expanding package sizes where a quarter sheet yields only 2–3 units by 2028, down from 6–8 units today. In server networking, House A projects Nvidia ($NVDA) Rubin Ultra switch trays starting in 2H27 will likely adopt M9/M10 CCL with Q-glass instead of PTFE, as PTFE yield runs at 10–15% versus 80–90% for M9 Q-glass. Different coverage metrics preclude a single comparable forecast range, revealing divergent focal points between pure capacity scarcity and macro headwinds. House B (dated 2026-10-21) anchors its multi-year AI substrate thesis on Unimicron securing customer capacity reservations for its Guangfu plant, where new supply releases in 2Q27 and reaches full capacity by mid-2028. Conversely, House F (dated 2026-10-22) projects that while strong FC-BGA and BGA pricing lifts Samsung Electro-Mechanics' operating margin by at least 1 percentage point to 18% (from 17%), currency headwinds trim its target price to W2.0mn from W2.4mn because a 1% won appreciation cuts operating profit by 1%+. Investors should watch the execution of 2Q27 customer-backed capacity ramps and formal Rubin Ultra tray laminate qualification. #PCB #Electronics

Goldman Sachs’ Taiwan PCB/CCL checks point to an accelerating AI hardware upgrade cycle. Customers are increasingly funding supplier capacity expansions through LTAs/prepayments, while M10 materials command ~$800/sheet vs EMC’s ~$63 blended ASP. ABF substrate CCL TAM could quadruple by 2028.