Compiled from AI, AIGPU, Semiconductor, Restocking themes + all reports

AI Core — MACRO CONTEXT

Japan's AI infrastructure boom is driven by >$26B hyperscaler investment (AWS $15B, Oracle $8B, Microsoft $2.9B), TSMC Kumamoto Fab 2 upgraded to 3nm (Apr 2026), and Rapidus 2nm targeting 2027. AI chip testing demand at Advantest surged 51% YoY. Government quadrupled chip + physical AI budget in late 2025. Key risk: US export controls on China (~30% of TEL/equipment revenue).

AI GPU Supply Chain — MACRO CONTEXT

Global AI GPU demand is reshaping Japan's semiconductor materials and component industry. NVIDIA GB200 Blackwell requires 30,000 MLCCs, Ibiden FC-BGA substrates, Resonac bonding materials, and Fujikura optical cables per server rack. Japan controls 90% of global photoresist (TOK), dominates CMP slurries (Resonac), and supplies critical packaging substrates (Ibiden). Total hyperscaler capex exceeds $600B in 2026. HBM demand growing 5-10x. Nidec (6594.T) removed due to active accounting investigation.

Semiconductor — MACRO CONTEXT

Japan dominates critical semiconductor supply layers: 90% coater/developer share (TEL), 70% batch ALD (Kokusai), 70%+ dicing (DISCO), 100% EUV mask inspection (Lasertec), 30% silicon wafers (Shin-Etsu). Government committed >¥2T in subsidies: TSMC Kumamoto (Fab 1 operational, Fab 2 upgraded to 3nm), Rapidus 2nm (pilot Apr 2025, mass production 2027), Micron HBM. JIC has taken JSR and Shinko Electric private to prevent foreign acquisition — consolidation pattern.

ALL COMPANIES (26 unique stocks)

3110.T — Nitto Boseki

Thesis: 90% monopoly in T-glass cloth for AI chip packaging substrates (ABF/BT). Every Nvidia GB200/300 requires T-glass. CCL prices +15-30%. 3x capacity expansion by 2027.

Risk: NE/NER glass losing M9 CCL to Asahi Kasei Q Glass. Glass core substrates long-term threat.

Trigger: T-glass capacity tripling timeline. TSMC CoPoS glass panel pilot June 2026.

Supply Chain:

• Nvidia: T-glass cloth for GB200/300 AI GPU substrates (confirmed) [https://www.tomshardware.com/tech-industry/artificial-intelligence/glass-cloth-could-be-the-next-great-ai-shortage]
• Apple: Glass cloth for iPhone PCB substrates (confirmed) [https://www.tomshardware.com/tech-industry/artificial-intelligence/glass-cloth-could-be-the-next-great-ai-shortage]
• Resonac Holdings: T-glass cloth for CCL production (confirmed) [https://www.digitimes.com/news/a20260304PD237/gases-ccl-ic-materials-price-resonac.html]
• TSMC: T-glass cloth for CoWoS/InFO advanced packaging substrates (confirmed) [https://www.trendforce.com/news/2026/04/13/news-tsmc-advances-panel-level-packaging-copos-pilot-line-reportedly-set-for-june-completion-2028-29-ramp-eyed/]

Segments:

• Glass Fiber (T-glass/NE-glass): Major (90%+ global T-glass) — Ultra-low CTE glass cloth for AI chip packaging substrates
• General Glass Fiber: Secondary (N/A) — Standard glass fiber for construction and industrial use

3407.T — Asahi Kasei

Thesis: Q Glass (quartz, Df 0.0005) wins M9 CCL specification for Nvidia Rubin-era products. PE 13.4x massively undervalued for M9 winner.

Risk: Conglomerate dilution. Q Glass scale-up uncertain.

Trigger: M9 CCL mass production ramp H2 2026.

Supply Chain:

• CCL manufacturers (EMC, Lianmao): Q Glass quartz fabric for M9-grade CCL (probable) [https://www.digitimes.com/news/a20260402PD227/asahi-kasei-ai-chip-fiberglass-cloth-materials-expansion-nittobo.html]
• PCB/CCL for telecom: High-function glass cloth for 5G/AI networking (probable)

Segments:

• Material (incl. Glass Cloth): Major (#2 Japan glass cloth) — Specialty glass cloth for high-speed PCBs, separators, fibers
• Homes: Major (N/A) — Residential housing construction
• Health Care: Secondary (N/A) — Medical devices, pharmaceuticals

3436.T — SUMCO

Thesis: Leveraged recovery play on silicon wafer upcycle. Currently loss-making but 300mm wafer demand from AI is structurally growing. +161% 1Y run reflects market pricing 2026 inflection.

Risk: No product diversification (100% wafers). Loss-making with negative FCF. ¥353B debt with -1.2x interest coverage. If upcycle delays, cash burn severe.

Trigger: Quarterly wafer shipment volume guidance. TSMC long-term supply agreement announcements.

Supply Chain:

• TSMC: 300mm silicon wafers (confirmed) [https://waferpro.com/top-5-silicon-wafer-manufacturing-companies/]
• Samsung / SK Hynix: Wafers for NAND, DRAM, HBM (confirmed) [https://waferpro.com/top-5-silicon-wafer-manufacturing-companies/]
• Intel: Silicon wafers (probable)

Segments:

• Silicon Wafers (single segment): ¥410B (100%) — 300mm polished, epitaxial, SOI wafers

4004.T — RESONAC HOLDINGS CORPORATION

Thesis: Global #1 in ceria CMP slurries and etching gases — core materials for HBM hybrid bonding and CoWoS processes. Won TSMC 2025 Excellent Performance Award for advanced packaging materials. Established local production for TSMC Kumamoto. Every HBM chip stack uses Resonac bonding materials — the picks-and-shovels play for the memory AI boom.

Risk: 1) PE 84.1x — one of the most expensive stocks in our universe. 2) ROE 4.0% and OpMar 3.5% are weak — turnaround not yet proven. 3) HBM demand could be front-loaded if AI capex cycle peaks. 4) Samsung and SK Hynix could develop alternative bonding materials.

Trigger: If operating margin exceeds 8% at next earnings (turnaround accelerating), ADD to 9%. If PE stays above 80x for 2 more quarters without margin improvement, TRIM to 5%.

Supply Chain:

• TSMC: CMP slurries, etching gases, and advanced packaging materials for CoWoS (confirmed) [https://www.resonac.com/news/2025/12/01/3669.html]
• SK Hynix: CMP slurries and bonding materials for HBM production (probable) [https://www.resonac.com/news/2025/12/01/3669.html]
• TSMC Kumamoto (JASM): Local production of high-purity gases and semiconductor materials (confirmed) [https://www.resonac.com/news/2025/12/01/3669.html]
• Applied Materials / Tokyo Electron: Joint development of panel-level interposer technology (confirmed) [https://www.trendforce.com/news/2025/09/04/news-resonac-teams-up-with-applied-materials-tel-and-other-chip-giants-on-panel-level-interposers/]

Segments:

• Semiconductor Materials: ¥499B (35%) — SiC substrates, CMP, photoresist
• Chemicals + Other: ¥854B (65%) — Petrochemicals, mobility, industrial

4062.T — Ibiden

Thesis: Primary FC-BGA substrate supplier for NVIDIA data center GPUs via CoWoS packaging. TSMC has taken a stake in Ibiden. ¥500B three-year capex plan starting FY2026 to expand IC substrate capacity. AI server substrate revenue ~3x YoY in FY2024. Every advanced AI GPU needs an Ibiden substrate — this is a physical bottleneck in the NVIDIA supply chain.

Risk: 1) PE 67x is extremely expensive — vulnerable to any demand slowdown. 2) FCF -¥8B (negative) from massive capex. 3) Forward PE 74.8x > trailing — market expects earnings to dip. 4) If TSMC diversifies CoWoS substrate sources, Ibiden loses pricing power.

Trigger: If IC substrate revenue growth decelerates below 20% YoY, TRIM to 5%. If FCF turns positive despite capex, ADD to 10%. If TSMC increases stake or signs exclusivity deal, ADD to 10%.

Supply Chain:

• NVIDIA: FC-BGA substrates for data center GPUs (H100, Blackwell) (confirmed) [https://www.allelcoelec.com/news/nvidia-ai-chip-substrate-supplier-ibiden-accelerates-production-expansion-to-cope-with-surging-deman.html]
• TSMC: IC package substrates for CoWoS advanced packaging (confirmed) [https://www.digitimes.com/news/a20250102PD221/ibiden-ic-substrate-demand-market-intel-nvidia.html]
• Intel: FC-BGA substrates for server processors (confirmed) [https://www.allelcoelec.com/news/nvidia-ai-chip-substrate-supplier-ibiden-accelerates-production-expansion-to-cope-with-surging-deman.html]
• AMD: FC-BGA substrates for data center GPUs (confirmed) [https://www.allelcoelec.com/news/nvidia-ai-chip-substrate-supplier-ibiden-accelerates-production-expansion-to-cope-with-surging-deman.html]

Segments:

• Electronics: ¥343B (86%) — ABF IC package substrates (NVIDIA, AMD, Intel), PCBs
• Ceramics: ¥55B (14%) — DPF (diesel), insulation, graphite

4063.T — Shin-Etsu Chemical

Thesis: Most defensive AI infrastructure play — #1 global silicon wafer maker (Shin-Etsu + SUMCO = 35% of 12-inch market). Fortress: ¥1.25T net cash, D/E 5%. Electronics Materials segment grew +5.9% YoY (9M) driven by AI server wafer demand. CEO Saito explicitly repositioning as 'AI stock.' New ¥83B Isesaki photoresist plant operational (first domestic plant in 56 years). GaN-on-QST: IMEC world record 800V breakdown on Shin-Etsu 300mm substrates. Total shareholder yield 5.9% (1.6% div + 4.3% buyback). FY2026 guidance ¥2.4T revenue (+24%) is conservative vs ¥2.57T TTM run-rate.

Risk: Q3 OP -14.8% YoY on flat revenue — PVC/Asia pricing deterioration is the main drag. PE 26.7x not cheap for 11.4% ROE. China wafer capacity push threatens commodity segments. FX headwind if yen strengthens (60%+ revenue from overseas).

Trigger: May 8: FY2026 full-year results + FY2027 guidance — watch Electronics Materials margin trajectory and PVC recovery. 300mm wafer ASP disclosures. Shintech $3.4B expansion progress.

Supply Chain:

• TSMC: 300mm silicon wafers, EUV photomask blanks, photoresists (confirmed) [https://www.shinetsu.co.jp/en/]
• Samsung: Silicon wafers for NAND + DRAM + logic + HBM (confirmed) [https://www.shinetsu.co.jp/en/products/semiconductor-silicon-business/]
• SK Hynix: Advanced 300mm wafers for HBM4 (confirmed) [https://www.shinetsu.co.jp/en/products/semiconductor-silicon-business/]
• Reshoring US fabs (TSMC Arizona, Samsung Texas, Intel Ohio): Wafers + photoresists for domestic US fab capacity (confirmed) [https://www.shinetsu.co.jp/en/]

Segments:

• Electronics Materials: ¥750.3B (9M) (~39%) — Silicon wafers (#1 global), photoresists (20-30% global share), EUV photomask blanks — +5.9% YoY, ~34.8% OP margin
• Infrastructure Materials (Shintech): ~¥550B (~29%) — PVC resin, caustic soda — US operations (Shintech, Louisiana). $3.4B expansion to 2030. Insulated from import tariffs.
• Functional Materials: ~¥400B (~21%) — Rare earth magnets, silicones, LED, encapsulants. New Zhejiang silicone plant (Feb 2026).
• Processing & Services: ~¥200B (~11%) — Wafer processing, specialty substrates, GaN-on-QST development

4182.T — Mitsubishi Gas Chemical

Thesis: Invented BT resin — the dominant core material for IC packaging substrates. Irreplaceable in advanced semiconductor packaging. Also makes CCL.

Risk: Chemical conglomerate with multiple segments diluting BT resin upside.

Trigger: BT resin demand tracking with advanced packaging volume.

4186.T — Tokyo Ohka Kogyo

Thesis: Japan controls ~90% of global photoresist supply. TOK is a top player — every AI chip at 5nm and below requires EUV photoresist. Record FY2025: revenue +17.9% to ¥237B, operating income +43.2%. ¥20B Korea plant expansion (3-4x capacity). Strategic partnership with Irresistible Materials for next-gen EUV. Operating margin 20.0% — high-quality materials monopoly.

Risk: 1) PE 34.7x is not cheap. 2) Korea plant expansion (¥20B) has execution risk. 3) Export controls on photoresist to China could limit TAM. 4) If EUV adoption slows (unlikely), demand plateaus.

Trigger: If Korea plant comes online and revenue accelerates >25% YoY, ADD to 8%. If export controls tighten and revenue guidance cut, TRIM to 4%.

Supply Chain:

• TSMC: EUV and ArF photoresists for advanced node fabrication (5nm, 3nm, 2nm) (confirmed) [https://www.trendforce.com/news/2025/11/06/news-japan-ramps-up-photoresist-investment-for-2nm-chips-tokyo-ohka-kogyo-jsr-lead-the-charge/]
• Samsung Electronics: Photoresists for memory and logic fabrication (confirmed) [https://www.trendforce.com/news/2025/11/06/news-japan-ramps-up-photoresist-investment-for-2nm-chips-tokyo-ohka-kogyo-jsr-lead-the-charge/]
• SK Hynix: Photoresists for HBM and advanced memory packaging (confirmed) [https://www.trendforce.com/news/2025/11/06/news-japan-ramps-up-photoresist-investment-for-2nm-chips-tokyo-ohka-kogyo-jsr-lead-the-charge/]
• Intel: Next-gen sub-2nm photoresists (joint development) (confirmed) [https://www.trendforce.com/news/2025/11/06/news-japan-ramps-up-photoresist-investment-for-2nm-chips-tokyo-ohka-kogyo-jsr-lead-the-charge/]

4975.T — JCU Corporation

Thesis: Near-monopoly in via-filling plating chemistry essential for high-density interconnects in advanced packaging. Operating margin 41.3% — highest in our entire universe. Benefits from a complexity multiplier: as AI chips get more complex (more layers, finer pitch), JCU sells more product per chip. Directly levered to Ibiden/Shinko capex cycles. Small cap (¥154B) with massive operating leverage to AI substrate buildout.

Risk: 1) Small cap ¥154B — low liquidity, volatile. 2) Niche product concentration — if alternative via-filling chemistry emerges, moat evaporates. 3) Customer concentration — heavily dependent on Ibiden and former Shinko Electric. 4) Limited diversification beyond plating chemicals.

Trigger: If Ibiden announces further capex expansion, ADD to 7% (JCU directly benefits). If operating margin drops below 35%, reassess moat. If a competitor enters via-filling market, EXIT.

Supply Chain:

• Ibiden: CU-BRITE via-filling acid copper plating chemicals for FC-BGA substrates (confirmed) [https://iamfabian.substack.com/p/jcu-corporation-4975t-an-overlooked]
• Shinko Electric Industries (Fujitsu subsidiary): Via-filling plating chemicals for IC packaging substrates (confirmed) [https://iamfabian.substack.com/p/jcu-corporation-4975t-an-overlooked]
• NVIDIA (indirect via Ibiden/Shinko): Critical plating chemistry enabling AI GPU substrates (inferred) [https://iamfabian.substack.com/p/jcu-corporation-4975t-an-overlooked]

5384.T — Fujimi Inc

Thesis: Leading CMP slurry maker. Gaining HBM market share from Soulbrain at Samsung/SK Hynix. PE 18x, 2.3% yield.

Risk: Cyclical. Competition from Cabot Microelectronics.

Trigger: HBM CMP slurry share gains. Semiconductor cycle turning.

5803.T — Fujikura Ltd

Thesis: The hottest AI infrastructure play in Japan — up 1,400% in 2 years on insatiable datacenter optical cable demand. ¥300B investment to triple optical fiber/cable capacity under Japan-US framework agreement. PE 11.0x with 27.7% ROE is extraordinary value for this growth trajectory. FY3/25 revenue +22.5% to ¥979B. Supplies high-count optical cabling for AI datacenter interconnects (GPU-to-GPU communication).

Risk: 1) Stock up 1,400% in 2 years — momentum reversal risk if AI capex cycle peaks. 2) ¥300B capacity investment is massive — execution risk on tripling production. 3) Competition from Furukawa Electric and Sumitomo Electric in optical cables.

Trigger: If AI datacenter capex guidance from hyperscalers (AWS/MSFT/GOOG) slows >20%, TRIM to 8%. If optical cable order backlog extends beyond 2 years, ADD to 15%. Watch quarterly revenue growth — if <15% YoY, reassess.

Supply Chain:

• Alphabet (Google): Optical fiber cables for AI datacenter interconnects (confirmed) [https://www.dividendjapan.com/p/fujikura-japan-connectivity-ai-dividend-2025]
• Apple: Optical connectivity components (confirmed) [https://www.dividendjapan.com/p/fujikura-japan-connectivity-ai-dividend-2025]
• U.S. Government (White House): Optical-fibre cables for AI infrastructure (confirmed) [https://www.silicon.co.uk/cloud/datacenter/fujikura-ai-demand-628113]
• Toyota Motor: Wiring harnesses and connectivity components (confirmed) [https://www.dividendjapan.com/p/fujikura-japan-connectivity-ai-dividend-2025]

6146.T — DISCO Corp

Thesis: Near-monopoly in precision dicing/grinding (~70%+ global share). Every advanced AI package (CoWoS, HBM, chiplet) requires DISCO equipment. Operating margin 41.7% reflects pricing power. No credible at-scale competitor.

Risk: Forward PE 45x is expensive. Small company (5,256 employees) — capacity scaling constraints. FCF data unavailable.

Trigger: TSMC CoWoS capacity additions (each line needs DISCO tools). HBM unit shipment volumes.

Supply Chain:

• TSMC: Dicing for CoWoS packaging (confirmed) [https://finance.yahoo.com/news/disco-corp-precision-toolmaker-quietly-115707225.html]
• SK Hynix / Samsung: HBM die singulation (confirmed) [https://finance.yahoo.com/news/disco-corp-precision-toolmaker-quietly-115707225.html]
• ASE / Amkor: OSAT advanced packaging (inferred)

Segments:

• Dicing Saws: ¥234B (55%) — Die singulation for advanced packaging
• Grinding/Polishing: ¥127B (30%) — Wafer thinning for 3D IC stacking
• Blades/Consumables: ¥64B (15%) — High-margin recurring

6525.T — Kokusai Electric

Thesis: ~70% global share in batch ALD — essential for GAA transistors and 3D DRAM. 5 confirmed GAA POR wins (CY2024). Won TSMC 2025 Excellent Production Support Award AND Intel 2026 EPIC Supplier Award. Building US Demo Center (Sep 2026). IPO'd Oct 2023 from KKR. FY2027 target: ≥20% revenue growth.

Risk: China ~37% of revenue (down from 47% via active de-risking). ¥14B DRAM/NAND orders deferred to FY2027. US/Japan export control escalation + US Section 232 tariff review. KKR selling overhang. Beta 2.12.

Trigger: May 13 FY2026 results + FY2027 guidance (≥20% revenue growth?). METI export control updates. China revenue quarterly disclosure. US tariff review outcomes.

Supply Chain:

• TSMC: Batch ALD/CVD for logic and advanced nodes (confirmed) [https://www.kokusai-electric.com/en/news/document/2025122600]
• Samsung Electronics: DRAM/logic deposition tools (confirmed) [https://www.kokusai-electric.com/en/]
• Micron Technology: Memory deposition tools (confirmed) [https://www.kokusai-electric.com/en/]
• Intel: Next-gen diffusion/deposition for Intel GAA nodes (confirmed) [https://www.kokusai-electric.com/en/news/document/2026032500]

Segments:

• Batch Film Deposition: ¥164B (69%) — LP-CVD, batch ALD, oxidation, diffusion (70% global share)
• Parts & Services: ¥73B (31%) — Aftermarket, maintenance

6526.T — Socionext

Thesis: Highest-margin fabless way to play custom AI silicon in Japan. Designs SoCs on TSMC 5nm/3nm for hyperscaler customers. 48.7% gross margin on design-only revenue. Near-zero debt. As hyperscalers design proprietary ASICs, Socionext's services grow in relevance.

Risk: Revenue -14.8% in FY2025 from customer inventory digestion. Revenue concentration in 1-2 large unnamed customers. Design win cancellation risk.

Trigger: New multi-year design contracts or tape-out starts at TSMC N3. Tape-out revenues are 2-3 year leading indicators.

Supply Chain:

• Fujitsu: Custom SoC design (legacy parent) (confirmed) [https://en.wikipedia.org/wiki/Socionext]
• Major US hyperscaler: AI inference SoC / SmartNIC (probable)
• Japanese automotive Tier-1s: Radar sensor SoC for ADAS (probable)

Segments:

• SoC Design (single segment): ¥210B (100%) — Custom SoCs: networking, automotive radar, AI inference, imaging

6723.T — Renesas Electronics

Thesis: De facto embedded intelligence for cars globally — auto digitization requires more Renesas MCUs per vehicle. R-Car X5H is industry's first 3nm automotive SoC. Expanding into AI data center power (800V GaN, confirmed NVIDIA partnership). Operating FCF ¥364B despite trailing IFRS loss.

Risk: ~55% automotive revenue = sensitive to global auto production. Trailing loss from Intersil/IDT/Dialog acquisition goodwill impairment — IFRS item, not operating.

Trigger: R-Car Gen 5/X5H design wins at Tier-1s (Bosch, Denso). These signal 3-5 year revenue visibility.

Supply Chain:

• Toyota / Toyota Group: RH850 MCU for ECUs (confirmed) [https://www.renesas.com/en/about/press-room/renesas-toyota-and-denso-bring-autonomous-driving-vehicles-market-faster]
• Honda / Stellantis / BMW: R-Car ADAS SoC (confirmed) [https://www.renesas.com/en/about/newsroom/honda-and-renesas-sign-agreement-develop-high-performance-soc-software-defined-vehicles]
• NVIDIA: 800V DC AI data center power (GaN) (confirmed) [https://www.renesas.com/en/about/newsroom/renesas-powers-800-volt-direct-current-ai-data-center-architecture-next-generation-power]

Segments:

• Automotive: ¥770B (55%) — RH850 MCU, R-Car ADAS SoC, body control
• Industrial/IoT: ¥630B (45%) — RA/RX MCU, analog, power management, DDR5 RCD

6762.T — TDK CORP

Thesis: World #1 MLCC maker. AI servers consume 30,000 MLCCs each vs 1,000 per smartphone — 30x volume multiplier. Murata raised AI server MLCC CAGR forecast to 30%. Exploring price hikes on AI server MLCCs due to tight supply. New ¥47B MLCC factory completed April 2026. Price hikes confirmed effective April 1. At 52-week high — demand is pulling the stock up.

Risk: 1) 15% US tariff impact = -24% net income hit. 2) Forward PE 22.7x > trailing 17.9x — earnings expected to dip. 3) Front-loaded H1 demand could create H2 cliff. 4) Smartphone MLCC demand weak offsets AI strength.

Trigger: If AI server MLCC revenue exceeds 15% of total at earnings, ADD to 10%. If US tariffs escalate beyond 15%, TRIM to 5%. If price hikes stick and margins expand, HOLD.

Supply Chain:

• Apple: iPhone batteries (via ATL subsidiary) and MLCCs (confirmed) [https://www.tdk.com/en/news_center/press/20251008_02.html]
• NVIDIA: High-current inductors and power modules for AI GPU servers (GB200) (confirmed) [https://intuitionlabs.ai/articles/nvidia-gb200-supply-chain]
• AI datacenter operators: Compact thin-film inductors for optical transceivers (confirmed) [https://www.tdk.com/en/news_center/press/20251008_02.html]

Segments:

• Passive Components: Major (#2 Japan) — Ceramic caps, inductors, ferrites, film caps
• Sensor Application: Secondary (N/A) — MEMS sensors, temperature sensors
• Energy Application: Major (N/A) — Rechargeable batteries (consumer, EV)

6787.T — Meiko Electronics

Thesis: Leading AI server PCB maker. High-layer HDI boards for AI servers. ROE 15.6%. Stock surged from 4,545 to 25,510 in 52 weeks.

Risk: Already priced in — massive run-up. High-layer PCB competition from Taiwan (Unimicron).

Trigger: AI server PCB order backlog. Margin sustainability after re-rating.

6806.T — Hirose Electric

Thesis: High-speed board-to-board and fiber optic connectors for AI server interconnects. PE ~22x, ROE ~8%.

Risk: Competition from Amphenol, TE Connectivity. Connector is commoditizing.

Trigger: High-speed connector design wins for AI racks.

6857.T — Advantest

Thesis: Monopoly-level share in SoC testers for AI accelerators. NVIDIA est. ~40%+ of revenue. FY2025 guidance raised 3rd time to ¥1.07T revenue, 42.4% op margin. Capacity expansion to 5,000 systems/yr by Mar 2027 (from 3,000), ahead of schedule. Won TSMC Excellent Performance Award. Memory tester revenue surging on HBM4 ramp (+30% QoQ in Q3). ¥100B convertible bond for capacity + R&D.

Risk: >40% NVIDIA revenue concentration. P/B 29.3x extreme. NEW: Teradyne qualifying for merchant GPU testing H1 2026 — direct competitive threat. SK Hynix developing in-house HBM4 system-level testers. China export controls tightening (20-25% of revenue).

Trigger: Apr 27 earnings: FY2026 guidance, Teradyne comments, capacity targets. Watch NVIDIA quarterly capex. If Teradyne confirmed on NVIDIA line, reassess monopoly thesis.

Supply Chain:

• NVIDIA: V93000 SoC testers for AI GPUs (H100/H200/Blackwell) (confirmed) [https://tspasemiconductor.substack.com/p/advantest-leading-the-ai-testing]
• TSMC: Logic chip test systems for advanced nodes (2nm/3nm) (confirmed) [https://www.advantest.com/en/news/2025/20251224.html]
• SK Hynix / Samsung: HBM memory test systems (HBM4) (confirmed) [https://www.advantest.com/en/news/2025/20251224.html]

Segments:

• SoC Test Systems: ¥165.2B (Q3) (~60%) — V93000 EXA Scale — dominates AI chip testing
• Memory Test Systems: ¥57.3B (Q3) (~21%) — T5835 for HBM KGD testing, surging on HBM4 ramp
• Mechatronics + Services: ¥51.3B (Q3) (~19%) — Handler systems, support, maintenance

6861.T — Keyence

Thesis: Most profitable industrial company globally — 49.9% operating margin. Q3 standalone revenue accelerated +11.4% YoY (Americas +15%, Asia +18.9%). Five major product launches Q1 2026: VS-G (AI vision + full-image storage), GX-1000 (3D printer — new market). ¥1.33T net cash, zero debt, ROIC 28.75%. Dividend +57% YoY. Fabless model insulates from tariff impact.

Risk: Forward PE 32.5x expensive (PEG ~4x). ~15% China revenue at risk from tariff escalation. Chinese vision players gaining domestic share.

Trigger: Apr 24-29: FY2025 full-year + FY2026 guidance. China revenue breakdown. Cognex quarterly competitive read.

Supply Chain:

• Toyota / major auto OEMs: FA vision and sensor systems (confirmed) [https://www.keyence.com/]
• Semiconductor fabs (TSMC, Samsung): VS-G vision systems for wafer/PCB inspection (probable)
• US reshoring factories: Full FA sensor/vision/measurement suite (confirmed) [https://www.keyence.com/]

Segments:

• Sensors/Switches: ~¥250B (9M) (~30%) — IV4 with built-in AI (99.8% detection accuracy)
• Vision Systems: ~¥209B (9M) (~25%) — VS-G (AI + full-image storage), XG-X (edge AI 10K images/sec)
• Laser Markers: ~¥167B (9M) (~20%) — CO2/fiber/UV lasers
• Measurement: ~¥125B (9M) (~15%) — LJ-X8000 2D/3D laser profiler, VK-X4000 microscope
• Safety / 3D Printing: ~¥84B (9M) (~10%) — Light curtains + GX-1000 3D printers (new market entry Apr 2026)

6920.T — Lasertec

Thesis: Pure monopoly — sole provider of actinic EUV mask blank inspection. Non-negotiable step for every chip at 5nm and below. 48.7% operating margin, 38% FCF margin. Co-dependent with ASML.

Risk: Forward PE 45x higher than trailing 43x — growth expectations front-loaded. TSMC N2 delay would defer orders. Scorpion short-seller report tail risk.

Trigger: TSMC N2 ramp schedule. ASML EUV tool shipment volumes (each needs Lasertec support).

Supply Chain:

• ASML: Co-dependent EUV ecosystem (confirmed) [https://fortune.com/asia/2024/03/21/lasertec-chip-stock-asia-future-japan-semiconductors-testing-shares/]
• TSMC: EUV reticle inspection for 3nm/2nm (confirmed) [https://fortune.com/asia/2024/03/21/lasertec-chip-stock-asia-future-japan-semiconductors-testing-shares/]
• Intel / Samsung: EUV mask inspection (inferred)

Segments:

• Semiconductor Inspection: ¥226B (90%) — EUV actinic blank + mask inspection (monopoly)
• LED/Display Inspection: ¥25B (10%) — Non-semiconductor inspection

6981.T — Murata

Thesis: AI revenue already 10% of total and growing 25-30% CAGR. New uPOL power modules specifically for AI GPU power delivery. MLCCs and passive components for datacenter power systems. PE 37.9x is expensive but AI revenue acceleration could justify re-rating. Diversified base (batteries, sensors, storage) provides stability while AI segment scales.

Risk: 1) PE 37.9x with 8.0% ROE — overvalued on current fundamentals. 2) AI is only 10% of revenue — takes years to become dominant. 3) Battery segment facing EV headwinds. 4) Competition from Murata in AI server passives.

Trigger: If AI segment exceeds 15% of revenue at next earnings, ADD to 7%. If uPOL wins design slots at NVIDIA/AMD, ADD to 7%. If PE expands beyond 45x without AI acceleration, TRIM to 3%.

Supply Chain:

• NVIDIA / AI server OEMs: MLCCs and uPOL power modules for AI GPU servers (confirmed) [https://www.trendforce.com/news/2025/11/06/news-japan-ramps-up-photoresist-investment-for-2nm-chips-tokyo-ohka-kogyo-jsr-lead-the-charge/]
• Apple: MLCCs for iPhone and consumer electronics (confirmed) [https://macro-ops.com/murata-manufacturing-6981-the-best-way-to-play-mlcc-demand-surge/]
• Hyperscale datacenter operators (AWS, Google, Microsoft): High-capacitance MLCCs and passive components for server power systems (probable)

Segments:

• MLCC (Capacitors): ¥694B (40% global) — Multi-layer ceramic capacitors for smartphones, auto, AI servers
• Communication Modules: Secondary (N/A) — WiFi/Bluetooth modules, SAW/BAW filters

6988.T — Nitto Denko

Thesis: PE 15.5x with 12.8% ROE and 18.8% OpMar — best value play in AI materials. Supplies optical films for displays AND flexible circuits/materials for HBM packaging. FCF ¥49B provides self-funding. Dividend yield 1.9%. Less hyped than Fujikura/Ibiden but similarly positioned in the AI material supply chain. Dual exposure: consumer electronics (stable) + AI/HBM (growth).

Risk: 1) HBM/AI materials revenue not yet dominant — still a diversified materials company. 2) Display business faces cyclical pressure. 3) Less direct AI exposure than Fujikura or Ibiden. 4) Korean competition in optical films.

Trigger: If AI/semiconductor materials segment exceeds 25% of revenue at earnings, ADD to 8%. If display segment drags total growth below 5%, HOLD at 6%.

Supply Chain:

• Apple: Polarizing films for iPhone/iPad displays and flexible printed circuits (confirmed) [https://www.nitto.com/]
• Samsung Display: Optical films (polarizers, retardation films) for OLED panels (probable)
• Semiconductor manufacturers (TSMC, Intel, etc.): Back-grinding tape, dicing tape, and encapsulating materials for semiconductor packaging (probable)

7735.T — SCREEN Holdings

Thesis: #1 globally in wafer cleaning. Advanced packaging (CoWoS, HBM stacks) requires intensive cleaning steps. Forward PE 18.7x — cheapest semi-cap equipment name vs peers at 30-45x. 20% capacity expansion (S3-6 facility, Feb 2025). ROE 19.8%.

Risk: Revenue dipped 4.3% TTM — cyclical softness. Memory investment pause pressures near-term.

Trigger: HBM capacity additions at SK Hynix/Samsung. TSMC CoWoS expansion announcements.

Supply Chain:

• SK Hynix: Cleaning for HBM advanced packaging (probable)
• TSMC: Cleaning for CoWoS packaging (probable)
• Samsung: DRAM and HBM cleaning (inferred)

Segments:

• Semiconductor Equipment: ¥443B (75%) — Wafer cleaning (#1 global), track systems
• Graphic Arts: ¥59B (10%) — Printing equipment
• Display/Coater: ¥47B (8%) — FPD, advanced packaging coater
• PCB Equipment: ¥41B (7%) — PCB processing

8035.T — Tokyo Electron

Thesis: #3 global WFE. ~90% global share coater/developer tracks. FY26 company guidance: sales ¥2,350B, OP margin 24.3%, AI-related equipment ~40% of sales (up from 30%). CY26 WFE market expected record high despite FY26 WFE -5% near-term softness. ¥1.5T R&D committed through March 2029. Zero debt, FCF ¥398B.

Risk: US/Japan export controls: China share declining toward 30% in FY26 (from 42% FY24). BIS rule tightening + Japan photoresist curb risk. TSMC 2nm IP leak case at TEL Taiwan subsidiary — management says limited impact but monitoring. One major logic customer delayed investment to Jan-Mar 2027. New DRAM capex unlikely before 2H 2026.

Trigger: FY25 full-year + FY26 outlook May 13, 2026. If AI-related mix tracks 40%+ AND HBM/DRAM equipment orders accelerate in 2H 2026, hold/add. If China share stays above 35% OR logic capex slips further (delayed to FY27), trim.

Supply Chain:

• TSMC: Coater/developer tracks, etch, cleaning systems (confirmed) [https://www.tel.com/]
• Samsung Electronics: Logic + DRAM + HBM etching/coating equipment (confirmed) [https://www.tel.com/]
• SK Hynix: HBM etch/deposition tools (probable) [https://www.trendforce.com/news/2025/12/08/news-tokyo-electron-sees-ai-driven-sales-hitting-40-by-2026-offsetting-china-slowdown/]
• Intel Foundry: Intel 18A / 14A process equipment (probable) [https://www.tel.com/]
• Micron Technology: DRAM and HBM equipment (probable) [https://www.trendforce.com/news/2025/12/08/news-tokyo-electron-sees-ai-driven-sales-hitting-40-by-2026-offsetting-china-slowdown/]

Segments:

• Semiconductor Production Equipment: ¥2,271B (95%) — Coaters, etch, CVD, thermal, cleaning
• FPD Equipment: ¥119B (5%) — Display equipment

9984.T — SoftBank Group

Thesis: Leveraged call option on AI infrastructure. ~90% of Arm (every AI chip uses Arm IP). $64B invested in OpenAI. Project Izanagi $67B AI infra pledge. Arm royalties compound as AI chips proliferate.

Risk: CRITICAL: ¥19.8T net debt. Negative FCF (-¥1.4T). Entirely dependent on asset monetization. Credit/liquidity event possible if AI sentiment reverses. Not a traditional operating company.

Trigger: Arm quarterly royalty revenue growth. If below 15% YoY, AI thesis weakens. Watch SoftBank credit spreads.

Supply Chain:

• Apple: Arm architecture licensing (confirmed) [https://en.wikipedia.org/wiki/Arm_Holdings]
• NVIDIA: Arm Neoverse for Grace Hopper SoC (confirmed) [https://en.wikipedia.org/wiki/Arm_Holdings]
• OpenAI: $64B equity investment + Stargate infra (confirmed) [https://openai.com/index/announcing-the-stargate-project/]

Segments:

• SoftBank (telecom): ¥3,200B (40%) — Domestic telecom — stable cash flow engine
• Arm Holdings: ¥500B (6%) — Chip architecture licensing — every AI chip
• Vision Funds: N/A (N/A) — SVF1 + SVF2 tech investments
• Investment/Other: ¥4,300B (54%) — T-Mobile, PayPay, direct holdings

REPORT: ai_gpu_supply_chain_dd_2026-04-15.md

AI/GPU Supply Chain — Japan's Hidden Infrastructure Layer

Time: 08:31

8 stocks powering the physical layer of AI

Date: 2026-04-15

Macro Context

Global AI GPU demand is reshaping Japan's semiconductor materials and component industry. NVIDIA's GB200 Blackwell requires 30,000 MLCCs, Ibiden FC-BGA substrates, Resonac bonding materials, and Fujikura optical cables per server rack. Japan controls 90% of global photoresist (TOK), dominates CMP slurries (Resonac), and supplies critical packaging substrates (Ibiden). Total hyperscaler capex exceeds $600B in 2026. TSMC Kumamoto Fab 2 upgraded to 3nm. HBM demand growing 5-10x. The physical infrastructure layer — cables, substrates, chemicals, capacitors — is the bottleneck that determines how fast AI can scale.

Portfolio

Stock Cards

Fujikura Ltd (5803.T) — 12% [HIGH]

WHY: The hottest AI infrastructure play in Japan — up 1,400% in 2 years on insatiable datacenter optical cable demand. ¥300B investment to triple optical fiber/cable capacity under Japan-US framework agreement. PE 11.0x with 27.7% ROE is extraordinary value for this growth trajectory. FY3/25 revenue +22.5% to ¥979B. Supplies high-count optical cabling for AI datacenter interconnects (GPU-to-GPU communication).

Risk: 1) Stock up 1,400% in 2 years — momentum reversal risk if AI capex cycle peaks. 2) ¥300B capacity investment is massive — execution risk on tripling production. 3) Competition from Furukawa Electric and Sumitomo Electric in optical cables.

Trigger: If AI datacenter capex guidance from hyperscalers (AWS/MSFT/GOOG) slows >20%, TRIM to 8%. If optical cable order backlog extends beyond 2 years, ADD to 15%. Watch quarterly revenue growth — if <15% YoY, reassess.

Ibiden (4062.T) — 8% [MEDIUM]

WHY: Primary FC-BGA substrate supplier for NVIDIA data center GPUs via CoWoS packaging. TSMC has taken a stake in Ibiden. ¥500B three-year capex plan starting FY2026 to expand IC substrate capacity. AI server substrate revenue ~3x YoY in FY2024. Every advanced AI GPU needs an Ibiden substrate — this is a physical bottleneck in the NVIDIA supply chain.

Risk: 1) PE 67x is extremely expensive — vulnerable to any demand slowdown. 2) FCF -¥8B (negative) from massive capex. 3) Forward PE 74.8x > trailing — market expects earnings to dip. 4) If TSMC diversifies CoWoS substrate sources, Ibiden loses pricing power.

Trigger: If IC substrate revenue growth decelerates below 20% YoY, TRIM to 5%. If FCF turns positive despite capex, ADD to 10%. If TSMC increases stake or signs exclusivity deal, ADD to 10%.

TDK CORP (6762.T) — 8% [MEDIUM]

WHY: World #1 MLCC maker. AI servers consume 30,000 MLCCs each vs 1,000 per smartphone — 30x volume multiplier. Murata raised AI server MLCC CAGR forecast to 30%. Exploring price hikes on AI server MLCCs due to tight supply. New ¥47B MLCC factory completed April 2026. Price hikes confirmed effective April 1. At 52-week high — demand is pulling the stock up.

Risk: 1) 15% US tariff impact = -24% net income hit. 2) Forward PE 22.7x > trailing 17.9x — earnings expected to dip. 3) Front-loaded H1 demand could create H2 cliff. 4) Smartphone MLCC demand weak offsets AI strength.

Trigger: If AI server MLCC revenue exceeds 15% of total at earnings, ADD to 10%. If US tariffs escalate beyond 15%, TRIM to 5%. If price hikes stick and margins expand, HOLD.

RESONAC HOLDINGS CORPORATION (4004.T) — 7% [MEDIUM]

WHY: Global #1 in ceria CMP slurries and etching gases — core materials for HBM hybrid bonding and CoWoS processes. Won TSMC 2025 Excellent Performance Award for advanced packaging materials. Established local production for TSMC Kumamoto. Every HBM chip stack uses Resonac bonding materials — the picks-and-shovels play for the memory AI boom.

Risk: 1) PE 84.1x — one of the most expensive stocks in our universe. 2) ROE 4.0% and OpMar 3.5% are weak — turnaround not yet proven. 3) HBM demand could be front-loaded if AI capex cycle peaks. 4) Samsung and SK Hynix could develop alternative bonding materials.

Trigger: If operating margin exceeds 8% at next earnings (turnaround accelerating), ADD to 9%. If PE stays above 80x for 2 more quarters without margin improvement, TRIM to 5%.

Tokyo Ohka Kogyo (4186.T) — 6% [MEDIUM]

WHY: Japan controls ~90% of global photoresist supply. TOK is a top player — every AI chip at 5nm and below requires EUV photoresist. Record FY2025: revenue +17.9% to ¥237B, operating income +43.2%. ¥20B Korea plant expansion (3-4x capacity). Strategic partnership with Irresistible Materials for next-gen EUV. Operating margin 20.0% — high-quality materials monopoly.

Risk: 1) PE 34.7x is not cheap. 2) Korea plant expansion (¥20B) has execution risk. 3) Export controls on photoresist to China could limit TAM. 4) If EUV adoption slows (unlikely), demand plateaus.

Trigger: If Korea plant comes online and revenue accelerates >25% YoY, ADD to 8%. If export controls tighten and revenue guidance cut, TRIM to 4%.

Nitto Denko (6988.T) — 6% [MEDIUM]

WHY: PE 15.5x with 12.8% ROE and 18.8% OpMar — best value play in AI materials. Supplies optical films for displays AND flexible circuits/materials for HBM packaging. FCF ¥49B provides self-funding. Dividend yield 1.9%. Less hyped than Fujikura/Ibiden but similarly positioned in the AI material supply chain. Dual exposure: consumer electronics (stable) + AI/HBM (growth).

Risk: 1) HBM/AI materials revenue not yet dominant — still a diversified materials company. 2) Display business faces cyclical pressure. 3) Less direct AI exposure than Fujikura or Ibiden. 4) Korean competition in optical films.

Trigger: If AI/semiconductor materials segment exceeds 25% of revenue at earnings, ADD to 8%. If display segment drags total growth below 5%, HOLD at 6%.

JCU Corporation (4975.T) — 5% [MEDIUM]

WHY: Near-monopoly in via-filling plating chemistry essential for high-density interconnects in advanced packaging. Operating margin 41.3% — highest in our entire universe. Benefits from a complexity multiplier: as AI chips get more complex (more layers, finer pitch), JCU sells more product per chip. Directly levered to Ibiden/Shinko capex cycles. Small cap (¥154B) with massive operating leverage to AI substrate buildout.

Risk: 1) Small cap ¥154B — low liquidity, volatile. 2) Niche product concentration — if alternative via-filling chemistry emerges, moat evaporates. 3) Customer concentration — heavily dependent on Ibiden and former Shinko Electric. 4) Limited diversification beyond plating chemicals.

Trigger: If Ibiden announces further capex expansion, ADD to 7% (JCU directly benefits). If operating margin drops below 35%, reassess moat. If a competitor enters via-filling market, EXIT.

Nidec (6594.T) — 5% [LOW]

WHY: Market leader in AI server liquid cooling equipment. Partnered with Lenovo on AI datacenter cooling. Targets ¥1T annual scale for cooling business long-term. Investing ¥5B to boost production. PE 40.7x but stock well off highs (¥2,217 vs ATH ¥3,296) — potential catch-up trade as AI server cooling demand accelerates with NVIDIA Blackwell power requirements.

Risk: 1) ROE 3.5% and OpMar 1.6% are very weak — turnaround dependent. 2) PE 40.7x for these margins is expensive. 3) ¥1T cooling target is long-term and unproven. 4) Competition from Daikin (Chilldyne) and Furukawa Electric in DC cooling.

Trigger: If cooling segment revenue exceeds ¥50B at next earnings, ADD to 7%. If OpMar stays below 3% for 2 more quarters, EXIT. If Lenovo partnership generates disclosed revenue, ADD to 7%.

REPORT: ai_supply_chain_gaps_2026-04-21.md

AI Supply Chain Gap Analysis — 6 Missing Layers Identified

Time: 07:30

Date: 2026-04-21

Type: Explore — supply chain completeness audit

Stocks added: 3110.T, 3407.T, 4182.T, 6787.T, 5384.T, 6806.T

Why We Were Missing Critical Layers

Our AI/semiconductor coverage (20 stocks) focused on chip-making equipment and materials but missed the layers BETWEEN chip fabrication and the final AI server. The full AI supply chain has 12+ layers — we covered 6 well but had gaps in 6 others.

Supply Chain Map — Before vs After

BEFORE (gaps marked with X)

`

Layer 1: Wafer Materials [OK] Shin-Etsu, SUMCO

Layer 2: Fab Equipment [OK] TEL, SCREEN, Kokusai, Lasertec, DISCO

Layer 3: Fab Materials [OK] TOK, Resonac, Nitto Denko, JCU

Layer 4: Chip Design [OK] Renesas, Socionext

Layer 5: Photomask [X MISSING]

Layer 6: CMP Slurry/Pads [X MISSING]

Layer 7: IC Substrates [OK] Ibiden (but missing BT resin supplier)

Layer 8: Glass Cloth/CCL [X MISSING] ← Nittobo, Asahi Kasei

Layer 9: PCB Manufacturing [X MISSING] ← Meiko Electronics

Layer 10: Connectors [X MISSING] ← Hirose Electric

Layer 11: Passive/Power [OK] Murata, TDK, Fujikura

Layer 12: Cooling [partial] Nidec in other theme

Layer 13: AI Platform [OK] SoftBank, Keyence

`

AFTER (6 gaps filled)

`

Layer 1: Wafer Materials Shin-Etsu (4063), SUMCO (3436)

Layer 2: Fab Equipment TEL (8035), SCREEN (7735), Kokusai (6525), Lasertec (6920), DISCO (6146)

Layer 3: Fab Materials TOK (4186), Resonac (4004), Nitto Denko (6988), JCU (4975)

Layer 4: Chip Design Renesas (6723), Socionext (6526)

Layer 5: Photomask [Still gap — Tekscend 429A.T recommended]

Layer 6: CMP Slurry NEW: Fujimi (5384) ← gaining HBM share

Layer 7: IC Substrates Ibiden (4062) + NEW: MGC (4182) ← BT resin inventor

Layer 8: Glass Cloth/CCL NEW: Nittobo (3110) T-glass 90% + Asahi Kasei (3407) Q Glass M9

Layer 9: PCB Manufacturing NEW: Meiko Electronics (6787) ← AI server PCBs

Layer 10: Connectors NEW: Hirose Electric (6806) ← high-speed server connectors

Layer 11: Passive/Power Murata (6981), TDK (6762), Fujikura (5803)

Layer 12: Cooling [In other themes — Nidec, Furukawa Electric]

Layer 13: AI Platform SoftBank (9984), Keyence (6861)

`

New Additions Detail

3110.T Nitto Boseki — Glass Cloth (T-glass) — HIGH

90% monopoly in T-glass for AI chip packaging. Every Nvidia GB200 needs it. 3x capacity expansion. CCL prices +15-30%. BUT NEZ glass lost M9 to Asahi Kasei Q Glass.

3407.T Asahi Kasei — Glass Cloth (Q Glass) — HIGH

Q Glass (quartz, Df 0.0005) wins M9 CCL specification for Nvidia Rubin. PE 13.4x is dirt cheap. Market hasn't priced in M9 win due to conglomerate structure.

4182.T Mitsubishi Gas Chemical — BT Resin/CCL — MEDIUM

Invented BT resin — the irreplaceable core material for IC packaging substrates since 1980s. Every advanced chip package uses BT laminate. Cheap conglomerate.

6787.T Meiko Electronics — AI Server PCBs — MEDIUM

Japan's leading high-layer HDI PCB maker for AI servers. ROE surged to 15.6%. Competes with Taiwan's Unimicron for AI server board orders.

5384.T Fujimi Inc — CMP Slurry — MEDIUM

Leading CMP slurry maker gaining HBM market share from Soulbrain at Samsung/SK Hynix. PE 18x, 2.3% yield. Essential for every wafer polishing step.

6806.T Hirose Electric — Connectors — MEDIUM

High-speed board-to-board and fiber optic connectors for AI server interconnects. PE 22x. The physical connections inside every AI rack.

Remaining Gaps

Lesson Learned

Our original AI DD focused on the "chip-making" layers (equipment + materials + design) but missed the "chip-packaging" and "server-building" layers. The supply chain bottleneck has shifted downstream: from wafers/equipment (2023-2024) to packaging materials, substrates, and PCBs (2025-2026). Glass cloth shortage is the proof — it's not a chip problem, it's a packaging material problem.

Future DD rounds should audit the full 13-layer chain, not just fab equipment.

REPORT: ai_supply_chain_layers10-13_japan_moat_2026-04-18.md

AI Semiconductor Supply Chain: Layers 10-13

Japan vs Taiwan vs Global — Irreplaceable Edges Analysis

Date: 2026-04-18

Executive Summary: Japan's Moat Ranking

Bottom line: Layer 11 (Passive Components) is Japan's TRUE monopoly position in these layers. The rest are competitive or catching up.

Layer 10: Connectors / High-Speed Interconnects

Market Structure (2025-2026)

• Global market size: USD 5.37B (2025), growing to USD 12.77B by 2036 (8.2% CAGR)
• AI data center connectors are the fastest-growing segment

Global Leaders (Japan is NOT dominant here)

Japan's Position: Hirose Electric (6806.T)

• Revenue: 9-month sales to Dec 2025: JPY 156.5B (~USD 1.0B); TTM ~USD 1.32B
• Segments: Automotive JPY 49.25B (growing 7% YoY), Smartphone JPY 45B (declining 3%)
• R&D: JPY 14B for FY2025-26
• Verdict: Hirose is a quality niche player but lacks the scale and AI-server dominance of Amphenol/TE. Not an irreplaceable position.

Optical Fiber / Transceivers (Japan's Stronger Sub-Niche)

Fujikura (5803.T) — Japan's hidden AI winner:

• Stock surged 160% in 2025; market cap exceeded USD 33B
• Revenue up 22.5% to JPY 979B (FY ending Mar 2025)
• 75% of optical fiber output exported (Google/Alphabet is a major customer)
• Announced plan to triple optical fiber production as part of USD 2B investment
• World's first optical fiber developer (1959)
• Supplies fiber cables for AI data center interconnects

Japan Pure Chemical — ultra-niche:

• Supplies gold-plating chemicals for optical transceiver PCBs
• Gold's conductivity/corrosion resistance makes it hard to substitute
• Currently ~5% of revenue but growing rapidly

Optical Transceiver Module Market (Japan is WEAK here)

Global leaders are US/China: Coherent (~16%), Innolight (China), Lumentum, Broadcom. Japan has ~6% of the transceiver market. China dominates with 64% market share by geography.

Layer 10 Verdict

Japan's edge: MODERATE in fiber optic cables (Fujikura), WEAK in connectors and transceiver modules.

• Fujikura is a strong #2-3 globally in fiber cables but competes with Corning (US) and Prysmian (Italy)
• Connectors are dominated by US companies (Amphenol, TE, Molex)
• Optical transceivers dominated by US/China

Layer 11: Passive Components + Power

THIS IS JAPAN'S TRUE MONOPOLY LAYER

MLCC (Multi-Layer Ceramic Capacitors)

Key facts:

• Japanese companies collectively control ~65-70% of global MLCC production
• In high-end AI server MLCCs: Murata + Samsung control 80%+ (Murata's 0402-size high-cap products are industry standard)
• Murata projects AI server MLCC demand to grow at 30% CAGR through 2030
• AI server MLCC demand expected to 3.3x by 2030 vs 2025 levels
• Each AI server uses 5,000-10,000 MLCCs vs 1,000 for a standard server

Inductors

Japan controls ~70% of global chip inductor production. This is near-monopoly territory.

Ferrite Materials

• TDK (originally Tokyo Denki Kagaku = Tokyo Electric Chemical) was literally founded on ferrite technology
• Japan dominates high-frequency ferrite cores used in power conversion

Power Semiconductors (Japan is Competitive but NOT Dominant)

Critical weakness: Japan's five power semi makers each hold <5% global share. Technology gap with Chinese competitors has narrowed to:

• Silicon chips: 1-2 years
• SiC chips: ~3 years maximum

GaN bright spot: Japanese entities hold >1/3 of GaN power-device patents since 2023, suggesting defensible IP depth even if market share is fragmented.

Layer 11 Verdict

IRREPLACEABLE in passive components (MLCC + inductors). Japan (primarily Murata, TDK, Taiyo Yuden) controls 65-85% of critical passive components that EVERY AI server requires thousands of. No credible near-term substitute exists.

Competitive but fragmented in power semiconductors. Not a chokepoint; Infineon/ON Semi are stronger.

Layer 12: Cooling / Thermal Management

Market Context

• Data center liquid cooling market: USD 6.65B (2025) -> USD 8.17B (2026)
• Goldman Sachs forecasts liquid-cooled AI servers: 15% (2024) -> 54% (2025) -> 76% (2026)
• Total market approaching USD 7B by 2029

Taiwan DOMINATES Thermal Modules (~70% Global)

Taiwan's four leaders have jointly partnered to develop cooling for Nvidia's Vera Rubin/Rubin Ultra chips (200-700% more power than current gen).

Taiwan accounts for ~70% of global thermal management module production.

Japan's Position: Niche Supplier, Not Leader

Nidec (6594.T) — CDU specialist:

• Supplies in-rack CDUs to Supermicro, Dell, HPE
• Developed OCP-compliant Project Deschutes CDU (Google spec), showcased at SC25
• Partnership with Fujitsu + Supermicro for integrated liquid cooling
• Strong in motor/pump technology (core competency)
• Market position: Significant in CDU sub-segment but not overall thermal market leader

Furukawa Electric (5801.T) — Heat pipe / vapor chamber:

• Product range: traditional heat pipes -> planar vapor chambers -> 3D vapor chambers -> liquid/immersion cooling
• Long history in thermal solutions
• Expanding into AI data center cooling

Fujikura — Also produces heat pipes and vapor chambers (separate from fiber business)

Japan vs Taiwan Comparison

Layer 12 Verdict

Taiwan dominates thermal module assembly. Japan has niche strengths in CDU pumps (Nidec) and heat pipe IP (Furukawa/Fujikura) but is NOT irreplaceable. Multiple alternatives exist for every Japan-sourced cooling component.

Layer 13: AI Platform / Infrastructure

Japan's AI Infrastructure Initiatives (2025-2026)

Government commitment: NEDO has earmarked ~JPY 1 trillion (~USD 6.3B) in AI support over 5 years from FY2026.

Major corporate moves:

NTT IOWN — potentially disruptive long-term:

• Shift from electronic to photonic data transmission
• Targets: 100x power reduction, 125x capacity, 200x latency improvement
• Still pre-commercial for AI inference; more of a 2028-2030 story

Preferred Networks (MN-Core):

• MN-Core L1000: custom processor for generative AI inference
• Joint venture PFCI launching AI cloud service in early 2026
• Niche player; cannot compete with Nvidia/AMD on training scale

Japan vs US Hyperscalers

Layer 13 Verdict

Japan is NOT competitive in cloud AI platform/infrastructure today. US hyperscalers dominate by 10-100x. Japan's differentiation is:

1. Physical AI / robotics (SoftBank consortium targeting factory automation)

2. Photonic networking (NTT IOWN — long-term disruptive potential)

3. Sovereign AI (Japanese-language models trained on Japanese data)

None of these represent irreplaceable chokepoints today.

Global Competitive Summary: Japan's TRUE Moat

Definitive Chokepoint Rankings (Layers 10-13)

#### TIER 1: TRUE MONOPOLY / IRREPLACEABLE

• Chip Inductors: Japan ~70% global production (Murata, TDK, Taiyo Yuden)
• MLCC (high-end): Japan ~65% global, ~45% AI server segment (Murata alone)
• Ceramic capacitor materials/processes: Decades of accumulated know-how in dielectric powder formulation

#### TIER 2: STRONG BUT NOT SOLE SOURCE

• Optical fiber cables: Fujikura is top-3 globally, but Corning (US) and Prysmian (Italy) are alternatives
• Heat pipe / vapor chamber IP: Furukawa, Fujikura have strong patents but Taiwan (AVC, Auras) competes
• CDU pumps/motors: Nidec is major but Vertiv, CoolIT, and others exist

#### TIER 3: COMPETITIVE / NO MOAT

• Connectors: US companies (Amphenol, TE) dominate; Hirose is niche
• Power semiconductors: Fragmented, Chinese catching up (1-3 year gap)
• Optical transceivers: US/China dominate module market
• AI platform/cloud: US hyperscalers dominate 10-100x

#### TIER 4: POTENTIAL FUTURE MOAT (NOT YET REALIZED)

• NTT IOWN photonic computing: Revolutionary if commercialized (~2028-2030)
• Physical AI foundation models: SoftBank consortium starting from behind
• GaN power IP: Japan holds >1/3 of patents but hasn't translated to market share yet

Cross-Cutting Japan Semiconductor Chokepoints (For Context)

These are NOT Layers 10-13 but show Japan's broader supply chain leverage:

Investment Implications

Strongest Japan plays in Layers 10-13:

1. Murata (6981.T) — MLCC + inductor monopoly, direct AI server demand (30% CAGR)

2. TDK (6762.T) — #2 inductors, ferrite, diversified passive portfolio

3. Fujikura (5803.T) — Fiber optic + heat pipe dual AI exposure (stock +160% in 2025)

4. Taiyo Yuden (6976.T) — #3 MLCC, purest passive component play

5. Nidec (6594.T) — CDU/liquid cooling for AI servers

Avoid overpaying for:

• Hirose (6806.T) — quality company but no AI server moat vs Amphenol/TE
• Power semiconductor names — fragmented, Chinese competition closing fast

Key Takeaway

In Layers 10-13, Japan has ONE truly irreplaceable position: passive components (MLCC + inductors). This is a real chokepoint — every AI server needs thousands of these, Japan controls 65-85% of production, and the technology gap (decades of ceramic material science) cannot be replicated in less than 10 years. Everything else is competitive or dominated by other geographies (US for connectors/transceivers, Taiwan for thermal modules, US for AI platforms).

Sources

• Global Growth Insights - High Speed Connector Market
• Amphenol at OCP 2025
• TrendForce - Murata AI Server MLCC Demand
• China MLCC Makers Reach 10% Market Share
• Lombard Odier - Why Liquid Cooling Will Dominate 2026
• Nidec Project Deschutes CDU
• Fujikura Stock Surges 160%
• Fujikura to Triple Optical Fiber Production
• Japan AI Foundation Model Company
• Microsoft $10B Japan AI Investment
• NTT IOWN Innovation
• TrendForce - Japan Power Chip Lead Shrinks
• Japan Times - Semiconductor Supply Chain Chokepoints
• CERADIR - Passive Component Supply Chain Analysis
• Taiwan Thermal Module Industry
• DIGITIMES - Taiwan AI Server Thermal Providers
• Hibiki Path Advisors - Japan Pure Chemical
• Preferred Networks MN-Core
• Connector Supplier - Changing Landscape
• Institut Montaigne - Japan Semiconductor Plans

REPORT: dc_sovereign_ai_dd_2026-04-14.md

DC Ecosystem & Sovereign AI — Deep DD

Time: 02:03

10 stocks powering Japan's AI infrastructure buildout

Date: 2026-04-14

Method: Claude Opus (all research + judgment). Data: J-Quants + StockAnalysis cross-validation. Triggered by SoftBank sovereign AI JV announcement (Apr 13) and Tomakomai DC groundbreaking.

Macro Context

April 13-14, 2026: SoftBank/NEC/Sony/Honda founded "日本AI基盤モデル開発" for sovereign AI (1T parameters). Government ¥1T/5yr support. SoftBank Tomakomai DC ¥65B (300MW→1GW). IDC Frontier absorbed Apr 1. Total hyperscaler+SoftBank capex in Japan exceeds $40B.

Portfolio

Stock Cards

Fujitsu (6702.T) — 10% [HIGH]

WHY this stock: Best risk/reward in the DC ecosystem. PE 12.9x with 22.7% ROE and ¥390B FCF is genuinely cheap for a sovereign AI infrastructure play. Started NVIDIA Blackwell server manufacturing at Kasashima plant (March 2026). Fugaku supercomputer pedigree. Both an AI compute provider AND an IT services compounder.

WHY 10%: 10% as HIGH conviction Core. PE 12.9x is the cheapest quality name in the DC group (Daikin 22.7x, SoftBank Corp 16.4x). ROE 22.7% is best-in-class. FCF ¥390B provides massive self-funding capacity. D/E 0.59x = clean balance sheet. Discount vs peers reflects market still pricing Fujitsu as legacy IT services, not sovereign AI infrastructure.

Risk: 1) AI compute/DC is still small relative to total IT services revenue — may take 2-3 years to move the needle. 2) NVIDIA dependency for Blackwell servers — supply allocation risk. 3) Forward PE 13.8x slightly higher than trailing 12.9x — market expects modest earnings growth, not acceleration.

Trigger: If NVIDIA Blackwell server revenue appears as breakout segment in FY2027 earnings (May 2026), ADD to 12%. If sovereign AI government contracts announced, ADD to 12%. If AI/cloud revenue growth below 15% YoY, HOLD at 10%.

Market misses: Fujitsu started manufacturing NVIDIA Blackwell AI servers at Kasashima plant in March 2026 — this makes them a physical node in the sovereign AI supply chain, not just an IT services company. The ¥390B FCF means they can self-fund DC expansion without dilution.

Meidensha Corporation (6508.T) — 12% [HIGH]

WHY this stock: Direct transformer bottleneck play. Global transformer lead times at 128+ weeks. ¥37B+ hyperscaler capex committed to Japan (AWS ¥2.26T, Microsoft $12.9B, Oracle $8B). Meidensha's 1.5x capacity expansion gives it pricing power in a supply-constrained market. Operating margin trajectory (3.1% → 7.2% in 2 years) shows structural improvement, not cyclical bounce.

WHY 12%: 12% as HIGH conviction Core holding. PE 17.1x (validated, StockAnalysis) is reasonable for this quality trajectory. ROE 15.6%, D/E only 0.32x — clean balance sheet. Outperforms SWCC (5805.T) on balance sheet quality and margin expansion rate, though SWCC has higher revenue growth. Forward PE 18.95x reflects near-term earnings dip from ¥16B capex absorption — acceptable for multi-year capacity build.

Risk: 1) Forward PE (18.95x) > trailing (17.1x) — analysts forecast 6.5% EPS decline in FY2026 as capex absorbs. 2) ¥5.8B gain on asset sale in TTM inflates current earnings — stripping it out, PE is ~21x. 3) No confirmed data center power contracts disclosed by name — thesis relies on industry-level demand evidence.

Trigger: If FY2026 earnings (May 2026) show operating margin declining below 6.0%, TRIM to 8%. If Numazu capacity expansion on schedule and new DC grid contracts announced, ADD to 15%. Watch transformer order backlog at earnings.

Market misses: Operating margin has doubled in 2 years (3.1% → 7.2%) but the stock is priced for the near-term earnings dip, not the structural margin expansion. The VCB capacity doubling (SF6-gas-free products) opens an export growth vector to North America/Europe that barely appears in current revenue.

SoftBank Corp (9434.T) — 8% [MEDIUM]

WHY this stock: The direct DC operator play. IDC Frontier absorbed April 1 2026. Tomakomai DC 50MW Phase 1 (2026) scaling to 300MW then 1GW. Infrinia AI Cloud OS differentiates vs pure co-location. ¥124B FCF + 3.9% dividend yield = income while waiting for DC ramp. Sovereign AI infrastructure anchor via parent SoftBank Group.

WHY 8%: 8% as MEDIUM conviction Core. PE 16.4x is fair (not cheap) for a telecom transitioning to DC operator. ROE 14.4% and OpMar 17.0% are solid. D/E 3.06x is the concern — leveraged balance sheet. FCF ¥124B and 3.9% DY provide income. Held at MEDIUM because DC revenue is still small vs telecom base and forward PE 19.5x shows market expects earnings dilution from DC capex cycle.

Risk: 1) D/E 3.06x — leveraged, DC capex will add more debt. 2) Forward PE 19.5x > trailing 16.4x — earnings expected to decline as investment cycle weighs. 3) DC revenue still small relative to telecom — may take 3+ years to become material.

Trigger: If DC segment revenue exceeds 10% of total at next earnings, ADD to 10%. If Tomakomai Phase 1 on schedule (2026), maintain. If D/E exceeds 4.0x, TRIM to 5%.

Market misses: IDC Frontier absorption (April 1 2026) + Tomakomai + Infrinia AI Cloud OS = SoftBank Corp is becoming Japan's largest DC operator, not just a telecom. The 1GW potential at Tomakomai alone is massive.

SWCC Corporation (5805.T) — 8% [MEDIUM]

WHY this stock: High-voltage cable incumbent for Japan's DC grid connections. Earnings accelerating (+59.7% EPS YoY). Showa Furukawa Cable full acquisition eliminates JV friction and enables manufacturing rationalization. Medium-Term Plan 2030 implies operating income nearly doubling. Superconducting cable demo with BASF provides future optionality for hyperscale DC power distribution.

WHY 8%: 8% as MEDIUM conviction Satellite holding. PE 25.4x is expensive relative to Meidensha (17.1x) despite similar quality trajectory. ROE 19.1% is best-in-class for the DC power group, but negative TTM FCF and one-time items (¥7.2B asset sale gain, ¥7.6B equity investment loss) distort earnings quality. FY2026 guidance revised upward (OP +24.2%, NI +40.3%) — earnings are accelerating, not declining as J-Quants forward PE suggested.

Risk: 1) Negative TTM FCF — heavy capex from Showa Furukawa integration. 2) One-time items inflate/distort TTM earnings (¥7.2B gain, ¥7.6B loss). 3) PE 25.4x is expensive for a cable company — vulnerable to de-rating if grid investment cycle slows before ¥40B OP target reached.

Trigger: If FY2027 (May 2026 earnings) shows OpMar below 9% or Showa Furukawa integration costs higher than expected, TRIM to 5%. If operating income reaches ¥30B+ run-rate, ADD to 10%. Watch FCF turning positive as integration capex completes.

Market misses: Medium-Term Plan 2030 targets ¥40B+ OP (nearly doubling from ¥25.3B) — this implies 12%+ OpMar achievable through Showa Furukawa consolidation and pricing power in supply-constrained cable market. The superconducting cable demo with BASF at Totsuka is a world-first that could become relevant for hyperscale DC power in 3-5 years.

NEC Corporation (6701.T) — 7% [MEDIUM]

WHY this stock: Co-lead developer (with SoftBank) of Japan's sovereign AI foundation model (1T parameters). NEC's cotomi LLM provides a head start. ¥1T/5yr government support flows primarily through NEC and SoftBank. Targeting ¥50B generative AI revenue over 3 years. Analyst target ¥6,088 (current ~¥4,011 = +52% upside).

WHY 7%: 7% as MEDIUM conviction Satellite. PE 27.3x is expensive, but sovereign AI JV membership gives preferential access to government contracts worth hundreds of billions yen. FCF ¥202B is strong. Held at MEDIUM because AI revenue is still bundled into IT Services (not broken out) and NEC has historically been slow to commercialize R&D.

Risk: 1) AI revenue buried in IT Services — hard to track growth. 2) PE 27.3x is not cheap for 8.8% ROE. 3) NEC historically slow to monetize R&D — sovereign AI may follow the same pattern. 4) If sovereign AI costs balloon without revenue, margins compress.

Trigger: If AI/generative revenue broken out at earnings and growing >30% YoY, ADD to 10%. If sovereign AI government contracts announced with NEC as primary contractor, ADD to 10%. If no AI revenue disclosure by FY2027, TRIM to 5%.

Market misses: NEC's role as co-lead in the sovereign AI JV gives it the inside track on ¥1T government AI spending. The cotomi LLM + existing government IT relationships create a moat that pure-play AI startups cannot replicate. Analyst consensus +52% upside signals under-ownership.

Sony (6758.T) — 6% [MEDIUM]

WHY this stock: Negative ROE (-6.4%) and FCF (-¥344B) are ACCOUNTING ARTIFACTS from the Oct 2025 SFGI financial services spin-off (¥1.4T non-cash loss). Continuing operations ROE is ~15%, OpMar 13.6%. Once FY2026 reports normalize, screener-based investors will discover a clean entertainment/tech conglomerate. 50%+ global CMOS image sensor share = direct link to autonomous vehicle AI. Sovereign AI co-founder for gaming/entertainment/robots.

WHY 6%: 6% as MEDIUM conviction Satellite. PB 2.28 is fair (not cheap) for 13.6% OpMar. The spin-off distortion creates a temporary window — once normalized metrics appear in FY2026, passive/quant funds that screened Sony out on negative ROE will re-enter. Sovereign AI adds gaming/entertainment optionality. Image sensor moat (50%+ global) is the strongest in the group. Held at MEDIUM because post-normalization upside may already be anticipated by informed investors.

Risk: 1) Spin-off noise may take 2-3 quarters to fully clear from databases. 2) PS5 price hike ($100-150, April 2026) risks demand destruction. 3) Image sensor business is cyclical (smartphone demand). 4) Kadokawa alliance (¥50B) has uncertain ROI.

Trigger: If FY2026 Q1 results show normalized ROE >12% and positive FCF, ADD to 8% (spin-off noise clearing). If image sensor revenue growth >15% YoY on autonomous vehicle demand, ADD to 8%. If PS5/gaming segment declines >10% YoY, TRIM to 4%.

Market misses: The -6.4% ROE headline is scaring away every screener-based investor, index fund, and quant strategy. Real continuing ROE is ~15%. When this normalizes in FY2026 reports, passive buying pressure will return. The SFGI spin-off actually makes Sony a cleaner, higher-growth story.

Honda (7267.T) — 7% [MEDIUM]

WHY this stock: PB 0.39 — trading at 39% of book value with ¥10T+ net assets. Sovereign AI co-founder using 1T-parameter model for autonomous driving (ASIMO OS). Honda 0 Series EVs launching H1 2026. DY 5.5%. Honda-Nissan merger dead (Feb 2025) but Honda pivoting independently. The market prices Honda as legacy auto in decline — PB 0.39 implies negative value for the operating business above tangible assets.

WHY 7%: 7% as MEDIUM conviction Tactical. PB 0.39x is the cheapest stock in our entire universe — even cheaper than JFE (0.45x). But OpMar 2.9% is razor-thin and EV transition costs are massive. Forward PE 5.5x from J-Quants may be aggressive (other sources show 7-8x). Sovereign AI membership gives autonomous driving data advantage. DY 5.5% pays you to wait. Tactical because this is a deep-value bet on EV/AI transformation, not a quality compounder.

Risk: 1) OpMar 2.9% — thin margins with EV transition capex ahead. 2) Honda-Nissan merger dead — no cost-sharing partner. 3) China retreat + global EV competition from BYD/Tesla. 4) Forward PE 5.5x may be too optimistic (7-8x more realistic).

Trigger: If Honda 0 Series pre-orders exceed expectations at H1 2026 launch, ADD to 10%. If OpMar drops below 2%, TRIM to 5%. If ASIMO OS licensed to other automakers, ADD to 10%.

Market misses: PB 0.39x assigns negative value to Honda's operating business above tangible assets. The sovereign AI JV gives Honda access to a 1T-parameter model for autonomous driving training — a massive advantage over building alone. ASIMO OS could become a robotics/SDV operating system that transcends cars.

DAIKIN INDUSTRIES (6367.T) — 5% [LOW]

WHY this stock: Global #1 HVAC maker with Chilldyne acquisition (negative-pressure liquid cooling) — the real moat for AI DC cooling. Every hyperscaler DC in Japan needs HVAC + liquid cooling for GPU racks. Revenue ¥4.84T, but DC cooling is still a tiny emerging segment. This is a long-duration bet on the physical infrastructure layer.

WHY 5%: 5% as LOW conviction Satellite. PE 22.7x is not cheap for 8.1% ROE. The DC cooling thesis is real (Chilldyne liquid cooling acquisition) but DC revenue is tiny relative to ¥4.84T total. This is a quality compounder where DC is optionality, not the core thesis. Better value in Fujitsu (12.9x, 22.7% ROE) and Meidensha (17.1x, transformer bottleneck).

Risk: 1) DC cooling is tiny vs total revenue — may never become material enough to re-rate the stock. 2) PE 22.7x is fair for HVAC quality but not cheap. 3) Chilldyne integration risk — liquid cooling is different from traditional HVAC.

Trigger: If DC cooling segment revenue disclosed and growing >30% YoY, ADD to 8%. If no DC-specific revenue breakout at earnings, maintain 5% as HVAC compounder.

Market misses: Chilldyne acquisition (negative-pressure liquid cooling) gives Daikin a technology moat in AI DC cooling that competitors lack. Every NVIDIA GPU rack generates 70kW+ of heat — liquid cooling is becoming mandatory, not optional.

Hokkaido Electric Power (9509.T) — 5% [MEDIUM]

WHY this stock: PE 3.4x, PB 0.47 — absurdly cheap direct power supplier for SoftBank Tomakomai DC (300MW → 1GW) and Rapidus 2nm fab (Chitose). Tomari Unit 3 nuclear restart (912MW, mid-2027 target) is the big catalyst — governor approval secured Dec 2025. Renewable energy surplus in Hokkaido for DC power. But trailing PE inflated by one-time items (forward PE 8.1x is more realistic).

WHY 5%: 5% as MEDIUM conviction Tactical. Even at forward PE 8.1x this is very cheap for a utility with nuclear restart catalyst. PB 0.47 provides deep value cushion. But D/E 4.17x is dangerously high for a utility — balance sheet risk is real. Small tactical position to capture Tomari restart + DC power demand optionality.

Risk: 1) Trailing PE 3.4x vs forward PE 8.1x — current earnings inflated by one-time factors, real PE is ~8x. 2) D/E 4.17x — highest leverage in all our portfolios. 3) Tomari nuclear restart could be delayed (NRA process). 4) Utility earnings are regulated — upside capped.

Trigger: If Tomari Unit 3 passes NRA safety review and restart confirmed for mid-2027, ADD to 8%. If D/E exceeds 5.0x or dividend cut, EXIT. If SoftBank Tomakomai DC power contract announced, ADD to 7%.

Market misses: SoftBank Tomakomai DC will eventually need 300MW-1GW of power — Hokkaido Electric is the monopoly supplier. Plus Rapidus fab needs power. Plus Tomari restart adds 912MW of nuclear baseload. The stock at PB 0.47 prices in none of this demand growth.

SoftBank Group (9984.T) — 5% [LOW]

WHY this stock: Architect of Japan's entire AI infrastructure — sovereign AI JV, Stargate ($500B), ARM (90% stake ~¥23T), OpenAI investor, Izanagi AI chips. UBS NAV ¥42.5T vs market cap ~¥20T = 30% discount. PE 5.1x is unreliable (holding company mark-to-market swings). FCF -¥4.6T is deliberate capital deployment into Stargate/AI, not operating weakness. This is a concentrated bet on the AI infrastructure buildout.

WHY 5%: 5% as LOW conviction Tactical ONLY. The NAV discount (30%) and AI thesis are real, but: concentration risk is extreme (ARM = ~70% of NAV), FCF -¥4.6T is terrifying, D/E 2.03x with massive commitments ($40B OpenAI loan, Stargate). If AI capex cycle peaks, everything unravels. Small tactical position captures optionality without portfolio-level risk.

Risk: 1) ARM is ~70% of NAV — if ARM corrects 30%, NAV collapses. 2) FCF -¥4.6T from Stargate/OpenAI commitments — refinancing risk if AI cycle slows. 3) PE 5.1x is meaningless for a holding company (swings with mark-to-market). 4) D/E 2.03x with massive future cash commitments ($40B OpenAI, Stargate multi-year).

Trigger: If ARM market cap drops >25% from peak, EXIT (concentration risk too high). If Izanagi chips ship on schedule (late 2026) and gain traction, ADD to 7%. If NAV discount narrows to <15%, TRIM (re-rating complete).

Market misses: 30% NAV discount implies market assigns zero value to everything beyond ARM — Stargate, OpenAI stake, Vision Fund, sovereign AI JV leadership, Izanagi chips. SoftBank's Izanagi chips (Graphcore + Ampere + ARM integration) could create a vertically integrated AI computing stack rivaling NVIDIA.

REPORT: deep_evolve_2026-04-19.md

Deep Evolve — Fuji Electric, Tokyo Electron, TDK

Time: 17:45

Date: 2026-04-19

Type: evolve (deep research — not metric-only)

Stocks evolved: 3 HIGH-conviction anchors (oldest data_date 2026-04-12)

Themes touched: nuclear, ev, ai, semiconductor

Why these three

All three are HIGH-conviction cross-theme anchors with 2026-04-12 data_date (oldest in

the universe). Each has had material news flow since April 12 that changes either the

thesis strength or the near-term catalyst timeline. This is a deep-evolve — not a

metric refresh — covering FY26 guidance, new supply chain links, and catalyst dates.

1. Fuji Electric (6504.T) — HOLD, nuclear + EV dual theme

Metric changes (2026-04-12 → 2026-04-19)

Data caveat: J-Quants PE (27.3x) is meaningfully above StockAnalysis (20.6x).

Per our data pipeline notes, J-Quants annualizes cumulative EPS (e.g., 3Q×4/3) while

StockAnalysis uses rolling TTM. Using StockAnalysis as display value. ROE drop

(11.8%→8.2%) reflects the latest filing annualization gap — will resolve at FY25 close.

What's new (material findings)

1. FY26 guidance raised Oct 30, 2025 — Sales ¥1,185B, OP ¥128.5B, NI ¥89B (all above

prior forecasts). Interim dividend raised to ¥91 (+21% YoY from ¥75).

2. Dec 19, 2025 — Bosch SiC co-development — Package-compatible SiC power modules for

EV inverters. Mechanical compatibility enables faster design cycles and sourcing

flexibility for automakers. This is MATERIAL — it broadens Fuji's addressable European

OEM base beyond Honda.

3. ¥200B capex FY24-26 in semiconductors — Targeting 50x SiC capacity vs 2022 baseline

by end of FY2026. This is the most aggressive SiC ramp among Japanese power-semi

players.

4. Analyst target raised — ¥11,650 → ¥11,920 (17-analyst consensus). Price ¥11,960

roughly at consensus.

5. 30% payout ratio commitment reiterated in FY26 medium-term plan.

Thesis status: HOLD

Core thesis (nuclear I&C monopoly + power-semi dual theme) fully intact. The Bosch JV

is net positive — expands TAM without threatening Honda relationship. Watch Q4 FY25

results (May 12) for:

• SiC capacity utilization (ramp path to 50x)
• Nuclear I&C segment quarterly order flow (restart pipeline)
• Data-center power electronics (Japan AI capex readthrough)

New supply chain links (5 total)

Added: J-Power / Kansai Electric / TEPCO aftermarket (nuclear I&C). Bosch SiC updated

with Dec 2025 press release URL.

Key dates refreshed

• 2026-05-12: FY25 full-year results
• 2026-05-12: Monitoring trigger — SiC/IGBT segment backlog + nuclear I&C order flow
• 2026-06-30: Semiannual rebalance
• 2026-08: Q1 FY26 results (NEW)
• 2026-10: Interim dividend announcement — watch 30% payout ratio progress (NEW)
• 2026-12: Bosch SiC JV first package-compatible module samples expected (NEW)

2. Tokyo Electron (8035.T) — HOLD, AI/semi anchor

Metric changes

What's new (material findings)

1. FY26 company guidance (Dec 2025 reaffirmed) — Sales ¥2,350B, OP margin 24.3%,

AI-related equipment rising to ~40% of sales (from 30% FY25). ¥1.5T R&D committed

through March 2029. ¥240B capex for FY26.

2. WFE market softness — FY26 WFE -5% vs CY26 expected record high. This is a

timing/bifurcation signal: near-term softness, CY26 inflection.

3. China sales trajectory — 42% FY24 → ~35% FY25 → ~30% FY26 target. AI-related

equipment growth is offsetting China decline per management.

4. DRAM/HBM inflection — Large-scale new DRAM equipment investment unlikely until

2H 2026. HBM etching tool demand strong now (SK Hynix, Samsung). This is the key

FY27 upside driver.

5. TSMC 2nm IP leak at TEL Taiwan subsidiary — CEO Kawamoto says limited impact on

TSMC relationship. Watching but not thesis-changing.

6. One major logic customer delayed investment plan to Jan-Mar 2027. This is the

source of FY26 WFE -5% softness.

7. Japan photoresist export curb rumored — China targets 40% self-sufficiency by

2026. Secondary risk to watch.

Thesis status: HOLD

CY26 WFE record + AI mix to 40% of sales + DRAM/HBM 2H 2026 inflection supports premium

multiple. But FY26 WFE -5% and delayed logic capex mean near-term earnings bumpy. Not a

time to add weight until the logic customer timing clarifies.

New supply chain links (5 total)

Added: SK Hynix (HBM etch — probable), Micron (DRAM — probable). Upgraded TSMC and

Samsung detail with 2nm leak and HBM context.

Key dates refreshed

• 2026-05-13: FY25 full-year results + FY26 guidance update
• 2026-05-13: Monitoring — AI mix %, China share %, FY26 WFE outlook
• 2026-06-30: Semiannual rebalance
• 2026-07: TSMC CY26 capex readthrough from Q2 earnings (NEW)
• 2026-08: Q1 FY26 results (NEW)
• 2026-Q4: DRAM/HBM capex inflection (Samsung/Hynix/Micron order flow) (NEW)

3. TDK Corp (6762.T) — HOLD, EV + AI-server HDD

Metric changes

Note: Both PE and Fwd PE moved up — stock price up materially. Market cap rose

~13% in a week.

What's new (material findings)

1. Q3 FY26 earnings (Feb 2, 2026) — RECORD 9M results: revenue ¥1,858.6B (+11.3% YoY),

OP ¥230.7B (+10.4%). Q3 specifically: net profit +26.5%.

2. FY26 guidance RAISED — Sales ¥2,370-2,470B (+12% YoY), OP ¥245-265B (+18.2%). This

is a meaningful guidance raise from prior ¥2,120B.

3. Magnetic Application segment OP +380% YoY — AI-server nearline HDD demand inflection.

HDD head and suspension assembly volumes +8% guided. This is the single biggest

positive surprise and a NEW narrative.

4. Sensor Application +17% growth — Magnetic sensors + MEMS microphones for ICT

(smartphones / AirPods). Apple supply chain confirmed strong.

5. Q4 FY26 outlook WEAK — Guided -8% to -11% sequential decline. Energy segment

-16% to -19% (BEV demand slowdown hit). Sensor -6% to -9%. Magnetic Application

still +6-9% growth.

6. TDK-Bosch JV for Apple sensors in US (Jan 2026) — Advanced sensors and ICs

produced in US for Apple. This is a MATERIAL customer disclosure (previously no

named customer for sensor segment).

7. Restructuring costs recorded for aluminum electrolytic capacitors and film

capacitors. BEV slowdown is the headwind.

Thesis status: HOLD

AI-server HDD head inflection (+380% OP growth in magnetic segment) is a genuine new

driver that wasn't in prior thesis. Offsets BEV slowdown. Valuation gap vs growth

narrowed (fwd PE 20.9→23.7), so the "cheap optionality" angle is weaker.

Watch Q4 print (May 8) vs -8% to -11% guide. If Q4 beats (ICT/HDD offsetting BEV),

thesis strengthens further. If Hungary NdFeB plant names a European OEM in 2026, that

would complete the bull case.

New supply chain links (6 total — biggest expansion)

Added:

• Apple — MEMS mics, MLCCs, sensors; TDK-Bosch US JV for advanced sensors (CONFIRMED Jan 2026)
• AI server OEMs (unnamed) — Aluminum electrolytic caps for high-V AI servers
• Nearline HDD OEMs (Seagate/WD/Toshiba) — HDD heads + suspensions (CONFIRMED)
• Porsche (Formula E) retained; LEM retained

Key dates refreshed

• 2026-05-08: FY26 full-year results + FY27 guidance
• 2026-05-08: Monitoring — Q4 sequential vs -8 to -11% guide; AI-server HDD head ramp
• 2026-06-30: Semiannual rebalance
• 2026-08: Q1 FY27 results — first read on BEV auto recovery (NEW)
• 2026-10: Hungary NdFeB plant — watch for named EU OEM (NEW)
• 2026-11: Q2 FY27 results (NEW)

Global / macro cross-cutting observations

1. AI infrastructure still driving hardware demand — TEL AI equipment to 40%, TDK

magnetic OP +380% on AI-server HDD, Fuji Electric eyeing data-center power electronics.

All three names are AI beneficiaries through different channels.

2. BEV demand slowdown is a real 2026 headwind — TDK explicitly flagged it; EV

supply chain names should be scrutinized for exposure. Fuji Electric's SiC ramp is

still on track (Bosch JV is forward-looking, not current revenue).

3. Export controls still worth monitoring — TEL China down to 30%, but photoresist

curbs (rumored Dec 2025) could pressure broader JP semi supply chain.

4. FY25 full-year earnings window May 8-13, 2026 — All three names report within

5 business days. This is the key near-term catalyst cluster.

Errata / data caveats

Sources

• Fuji Electric — FY26 guidance + interim dividend (Oct 2025)
• Fuji Electric — Bosch SiC co-development (Dec 2025)
• Fuji Electric — SiC 50x capacity expansion
• Tokyo Electron — AI-driven sales to 40% by FY2026 (TrendForce Dec 2025)
• TEL Q1 FY2026 slides — investment patterns shift
• TDK Q3 FY2026 results — record sales, guidance raised
• TDK-Bosch US JV for Apple sensors
• StockAnalysis — Fuji Electric
• StockAnalysis — Tokyo Electron
• StockAnalysis — TDK

Methodology: J-Quants paid plan (107-column fundamentals) for raw financials +

company forecasts. StockAnalysis.com for PE/PB display values (split-adjusted, rolling

TTM). Web research via Claude for macro, press releases, IR decks. All supply chain

links graded CONFIRMED / PROBABLE / INFERRED with dated evidence.

REPORT: deep_evolve_3stocks_2026-04-14.md

Deep DD Evolution Report — 2026-04-14

Time: 13:11

Stocks covered: Kokusai Electric (6525.T), ShinMaywa Industries (7224.T), Meidensha Corporation (6508.T)

Type: Deep evolve — analyst-level research, not just metric updates

Date: 2026-04-14

1. Kokusai Electric (6525.T) — Semiconductor Equipment

Theme: Semiconductor | Conviction: HIGH | Weight: 10%

What's New

Intel 2026 EPIC Supplier Award (March 25, 2026) — HIGH materiality

Kokusai received Intel's highest supplier honor for "Excellence in Local R&D Investment to Enable Next Generation Diffusion Technology." Building a US Demo Center (completion Sep 2026). Selected from thousands; one of only 41 global honorees. Confirms Intel GAA process-of-record alignment.

Source: https://www.kokusai-electric.com/en/news/document/2026032500

FY2026 Guidance Cut — HIGH materiality

• Revenue: ¥230B (-3.7% YoY from ¥238.9B)
• Adj. Operating Profit: ¥44.4B (-23.1% YoY)
• ¥14B of China DRAM/NAND equipment orders deferred to FY2027
• China revenue share: 47% → ~37% of FY2026 total (active de-risking)
• FY2027 target: ≥20% YoY revenue growth (management guiding outperformance)

Source: https://www.investing.com/news/company-news/kokusai-electric-q3-fy263-slides-service-revenue-grows-amid-equipment-sales-decline-93CH-4501941

5 GAA Process-of-Record Wins — HIGH materiality

Confirmed POR wins at logic/foundry customers in CY2024. Combined with TSMC 2025 Award + Intel 2026 EPIC, validates dominance across all 3 major foundry/IDM customers.

US Section 232 Tariff Review — HIGH materiality

Equipment currently exempt from 25% tariff, but April 14 review deadline (today) could change this. Global 10% baseline tariff in effect until July 24, 2026.

KKR Stake Reduction — MEDIUM materiality

Principal shareholder change disclosed Feb 2, 2026. Potential selling overhang.

China Retaliatory Controls — MEDIUM materiality

China MOFCOM issued export controls on dual-use items to Japan (Jan 6, 2026). Anti-dumping probes on Japanese chipmaking materials.

Thesis Validation

Thesis status: HOLD — Structural story stronger than ever. Near-term headwinds from guidance cut, China deferrals, and tariff uncertainty. Stock down from ¥7,575 high.

Metric Changes

Key Dates

2. ShinMaywa Industries (7224.T) — Defense & Aerospace

Theme: Defense & Aerospace | Conviction: HIGH | Weight: 12%

What's New

Arms Export Liberalization — HAPPENING NOW (April 2026) — HIGH materiality

Japan's LDP is finalizing revision of Three Principles on Transfer of Defense Equipment. Implementation expected by end of April 2026. Lethal weapons exports allowed for first time. This directly unlocks US-2 sales to partner nations.

Sources: UPI Apr 7 2026, Nikkei Asia

India Wet Lease RFI (January 2026) — HIGH materiality

India MoD issued RFI for 4 fixed-wing amphibious aircraft on 4-year wet lease. Pivoted from stalled outright purchase — wet lease lowers entry barrier dramatically.

Source: Bharatshakti.in

Pentagon NDAA Seaplane Program (Dec 2025) — HIGH materiality

FY2026 NDAA authorizes SecDef and SecNav to contract commercially-operated amphibious aircraft for INDOPACOM. 3-year program. US-2 described as "near perfectly fit" and "leading candidate."

Firefighting Export Route (Feb 2026) — MEDIUM materiality

ShinMaywa identified firefighting as civil export route for US-2. Targets: Australia, Taiwan, Greece.

Q3 FY2025 Results — HIGH materiality

• Order backlog: ¥349.7B (+11.1% YoY)
• NI: ¥6.3B (+28.5%)
• Dividend raised to ¥54/share
• Revenue guidance cut ¥4B on EV weakness, but profit guidance held

Thesis Validation

Thesis status: UPGRADED — Arms export rule change is the biggest policy catalyst. India + Pentagon represent first credible demand diversification beyond JMSDF. Execution risk 2-3 years.

Metric Changes

Key Dates (NEW — previously empty)

3. Meidensha Corporation (6508.T) — Data Center Power

Theme: Data Center Ecosystem | Conviction: HIGH | Weight: 12%

What's New

Numazu ¥16B Capacity Expansion CONFIRMED — HIGH materiality

¥16B investment for testing facility at Numazu Works. 1.5x production capacity target. 25% capacity vs FY2023, 50% lead time reduction.

Source: Meidensha IR Oct 30, 2025

Meiden America US Expansion (Oct 2025) — MEDIUM materiality

$9.2M, +37,000 sq.ft. in South Carolina. 123 kV circuit breaker manufacturing. SF6-free for green grid.

Transformer Bottleneck WORSENING — HIGH materiality

Power transformers: 128 weeks. GSUs: 144 weeks. Prices up 77% since 2019. 30% shortfall projected. Competitor capacity (Hitachi, Siemens) all 2027-2028.

Japan Utility Grid ¥150B+ Investment — HIGH materiality

Osaka substations, Greater Tokyo 66 kV expansion. Direct domestic demand.

FY2026 Earnings Dip Confirmed — HIGH materiality

OP guided DOWN to ¥20.0B (from ¥21.5B). NI ¥14.0B (-24%, stripping ¥5.8B asset sale).

Dividend raised ¥123 → ¥136 (+10.6%).

Valuation Stretched — MEDIUM materiality

At ¥8,360, above analyst targets (¥7,600-7,967). Priced for FY2027 recovery.

Thesis Validation

Thesis status: HOLD — Structural thesis validated. FY2026 dip expected. FY2027 MTP (¥25B OP) is the real catalyst. Valuation full.

Metric Changes

Key Dates

Cross-Stock Observations

1. May 2026 earnings cluster: All 3 report full-year in May. Most important catalyst window.

2. Tariff exposure: Kokusai HIGH, ShinMaywa LOW, Meidensha LOW.

3. All 3 have structural tailwinds confirmed: GAA transition, defense 2% GDP, transformer shortage.

4. Near-term earnings pressure: Kokusai (OP -23%) and Meidensha (OP -7%) guide lower. ShinMaywa guides higher.

*Report generated by Claude deep evolve pipeline. All metrics cross-validated. 2026-04-14.*

REPORT: deep_evolve_dd_2026-04-14.md

Deep DD Evolution Report — 2026-04-14

Time: 20:53

Stocks Evolved

1. Fanuc (6954.T) — Robotics & Factory Automation

New Information Found

Financial Filings (Q3 FY2025, Jan 26, 2026):

• 9M revenue ¥623.3B (+6.5% YoY), operating income ¥127.7B (+15.6% YoY)
• Full-year FY2025 guidance revised upward to ¥840.7B revenue, ¥172.9B operating income
• Robot segment is now the dominant division at 44.8% of Q3 revenue (+21.9% YoY)
• China revenue: 26.8% of Q3 sales (+22.7% YoY), driven by EV supply chain automation
• Operating margin expanding: 20.5% for 9M vs 18.9% prior year

NVIDIA Physical AI Partnership (Mar 16, 2026):

• Concrete integration: NVIDIA Jetson edge modules, Isaac Sim, Omniverse into Fanuc robots
• Voice command → automatic Python code generation for robot programming
• Post-iREX (Dec 2025): 1,000+ CRX orders received, pipeline of "several thousand units"
• Major automotive manufacturers quoted: "FANUC is far ahead of other companies"

$90M US Manufacturing Investment (Mar 24, 2026):

• New 840,000 sq ft Michigan facility for paint robots + Physical AI
• Part of ~$300M deployed in US since 2019
• Strategic tariff hedge — US-made robots sidestep potential Section 232 tariffs
• Completion: late 2027

New Products (iREX Dec 2025):

• CRX-3iA: Ultra-lightweight 11kg hand-carry cobot for shipbuilding
• R-2000: Maintenance-free robot with battery-less encoders — auto OEMs requesting immediate deployment

Tariff Impact Assessment

• US revenue ~25-27% of total (~¥200B)
• Yen at ¥159/USD vs guidance ¥145 provides partial FX offset
• US factory expansion (late 2027) is structural but not near-term hedge

China Monitor: PMI 50.4 (March 2026)

• Official NBS Manufacturing PMI: 50.4 (beat expectations of 50.1)
• Caixin Manufacturing PMI: 50.8 — 4th consecutive month of expansion
• Fanuc Q3 China robot sales: +46.3% YoY, driven by EV automation
• Monitoring trigger "PMI > 52" not yet hit but trend is positive

Competitive Landscape

• SoftBank acquiring ABB Robotics for $5.4B — near-term disruption benefits Fanuc, medium-term creates stronger competitor
• JARA robot orders at record levels: ¥324.5B in Q1 2025 (+14.2% YoY)

Metric Changes

Updated Risk Assessment

Original risks confirmed: ~27% China revenue (higher than 25% estimate), capex cyclicality

New risks added:

• Section 232 investigation (May 30 deadline) — potential 25%+ tariff on robots/CNC
• 10% baseline tariff already active on US-bound shipments
• 24% reciprocal tariff suspension expires ~Jul 9, 2026

Thesis Status: HOLD

The thesis strengthens on fundamentals (guidance raised, robot segment +21.9%, NVIDIA partnership concrete) but weakens on tariff risk. Net: HOLD with increased monitoring. The April 25 earnings will be critical — will Fanuc issue FY2026 guidance with tariff assumptions?

Sources

• Fanuc Q3 FY2025 Reference PDF
• FANUC-NVIDIA Physical AI PR
• FANUC $90M US Investment PR
• Section 232 Robotics Investigation (White & Case)
• StockAnalysis 6954.T

2. Advantest (6857.T) — AI / Semiconductor Test Equipment

New Information Found

Financial Filings (Q3 FY2025, Jan 28, 2026):

• Q3 revenue ¥273.8B (+25.5% YoY), operating income ¥113.6B (+64.0%), operating margin 41.4%
• FY2025 guidance raised for 3rd time to ¥1.07 trillion revenue, ¥454B operating income (42.4% margin)
• SoC tester revenue ¥165.2B (~60% of test systems), memory tester ¥57.3B (+30% QoQ)
• Management: "anticipated demand digestion did not materialize — AI pull-in sustained orders"

Capacity Expansion (Jan 2026):

• CEO targeting ≥5,000 SoC tester systems/year by March 2027 (from ~3,000 now)
• "5,000 is the bare minimum" — midterm plan targets 7,500 → 10,000 systems
• Capacity expansion described as "ahead of schedule"

Convertible Bond (April 2026):

• ¥100B zero-coupon convertible bonds due 2031
• Conversion price ¥36,000 (60% premium) — only 0.38% dilution
• Proceeds: capacity expansion, strategic inventory, next-gen R&D

TSMC Recognition (Dec 2024):

• Won TSMC's "Excellent Performance Award" — signals deep strategic partnership

Earnings Date Correction:

• Actual earnings date: April 27, 2026 (not April 30 as previously listed)

NVIDIA Capex — No Slowdown Visible

• NVIDIA FY2026 revenue: $215.9B (+65% YoY), data center $197.3B
• Q1 FY2027 guidance: ~$78B (continuing acceleration)
• Hyperscaler capex approaching $700B annually (Amazon $200B alone)
• Risk thesis "NVIDIA capex pullback" is NOT materializing

Competitive Threat: Teradyne

This is the most material new finding:

• Teradyne Compute segment grew 90% in FY2025, now ~50% of their SoC revenue
• Teradyne holds ~50% share of custom-ASIC chip testing (Google TPU, Amazon Trainium)
• Teradyne expects qualification on a major merchant GPU maker's production line in H1 2026 — likely NVIDIA
• If confirmed, this would be the single most significant competitive threat to Advantest's thesis

HBM4 Memory Tester Opportunity

• SK Hynix and Samsung both delivering paid HBM4 samples to NVIDIA
• Samsung planning 50% HBM capacity surge in 2026
• Memory tester revenue growing: Q2 ¥43.9B → Q3 ¥57.3B (+30% QoQ)
• Watch: SK Hynix developing own system-level HBM4 testing equipment internally

Tariff Assessment

• Section 232 tariff (25%) on semiconductor equipment effective Jan 15, 2026
• Management says "direct impact remains limited" — Advantest sells to chip makers, not data centers
• Equipment tariffs are secondary; if US chip production shifts (TSMC Arizona), tester demand follows

Metric Changes

Updated Risk Assessment

Original risks confirmed: NVIDIA concentration (~40%+), extreme valuation (P/B 29x)

New risks added:

• Teradyne qualifying for merchant GPU testing (H1 2026) — direct competitive threat
• SK Hynix developing in-house HBM system-level testers
• China export controls tightening; Chinese domestic competitors (CCTC, Huafeng) gaining mid-range share

Risk mitigated: "NVIDIA capex pullback" — currently NOT materializing; Q1 FY2027 guidance strong

Thesis Status: HOLD

Thesis strongly intact on fundamentals: 3rd guidance raise, ¥1.07T revenue, 42.4% margin, capacity expansion ahead of schedule. The Teradyne GPU qualification is the key new risk to monitor — if Teradyne enters NVIDIA supply chain, Advantest's near-monopoly in AI GPU testing narrows. Valuation remains extreme but earnings growth is supporting it.

Sources

• Advantest Q3 FY2025 Results
• Bloomberg: Advantest capacity expansion
• NVIDIA Q4 FY2026 Earnings (Fortune)
• Teradyne compute growth (Seeking Alpha)
• SK Hynix HBM4 internal tester (TrendForce)
• StockAnalysis 6857.T

3. Murata Manufacturing (6981.T) — EV / Passive Components

New Information Found

Financial Filings (Q3 FY2026, Feb 2, 2026):

• 9M revenue ¥1,370.2B (+2.9% YoY), Q3 standalone ¥467.5B (+4.3%)
• MLCC sub-segment 9M revenue ¥694.0B (+10.1% YoY) — driven by AI servers
• Operating profit 9M: ¥203.0B (-13.3% YoY) — distorted by ¥43.8B SAW filter goodwill impairment
• Underlying operating profit (ex-impairment): healthy
• FY2026 guidance: Revenue raised to ¥1,800B (+3.2%), OP lowered to ¥270B (impairment)

New MLCC Factory (April 3, 2026):

• Izumo, Shimane Prefecture: 10-story, 69,829 sqm facility completed
• Investment: ~¥47B including equipment
• Purpose: MLCC capacity for automotive, industrial, consumer demand growth

New Automotive MLCCs (April 7, 2026):

• 7 new AEC-Q200-qualified automotive MLCCs in mass production
• 5 parts for ADAS/autonomous driving IC peripheral circuits
• 2 parts for in-vehicle power line applications
• Technical breakthrough: 100uF in 1206 package (36% PCB space reduction)

Price Increases Confirmed (April 1, 2026):

• Price hikes across 4 passive component categories effective April 1
• Samsung Electro-Mechanics also weighing double-digit MLCC price hikes
• Driver: silver cost surge + AI/EV demand pressure

AI Server MLCCs — Major New Growth Vector:

• AI server (GB300 platform): ~30,000 MLCCs per server (vs 1,000 per smartphone, 3,000-5,000 per ICE vehicle)
• MLCC cost rank in AI server BOM: #3 (after GPU, memory)
• Murata forecasts AI server MLCC demand 30% CAGR through FY2030 (3.3x current demand)
• Murata holds ~45% share in AI server MLCCs (Samsung ~40%)
• Planning mass production of AI server power modules in 2026 — ¥50B revenue target by FY2027

Tariff Assessment

• 15% tariff on Japanese imports to US (reduced from initial 25% proposal)
• Management: "tariff impact not yet fully priced in" to guidance
• Net income expected to fall 24% in FY2026 partly due to tariffs
• Front-loading risk: demand "expected to be front-loaded in H1 due to tariff policy" — H2 pullback risk
• Mitigation: diversifying production from 60% Japan to targeting 50%, expanding to Thailand, India, Mexico

MLCC Market Dynamics

Bifurcated market:

• General-purpose: surplus, price pressure, 8-12 week lead times
• High-spec (automotive, AI, 5G): structural shortage, 20-30 week lead times, 5-10% price increases
• China MLCC makers now ~10% global share but concentrated in general-purpose
• Murata + TDK control ~85% of automotive MLCC market

Metric Changes

Updated Risk Assessment

Original risks confirmed: Cyclical MLCC pricing, semiconductor downturn impact

New risks added:

• 15% US tariff active — net income -24% impact
• Front-loaded demand in H1 creating H2 pullback risk
• SAW filter business under competitive pressure (¥43.8B impairment from Chinese rivals)
• Samsung closing gap in AI server MLCCs (40% vs Murata's 45%)

New thesis strengtheners:

• AI server MLCCs: 30x demand per server vs smartphone; 30% CAGR
• April 1 price hikes confirmed — pricing power is real
• New ¥47B MLCC factory completed, India operations beginning
• Capex doubled to ¥220B/year — management conviction in demand durability
• Mid-term target: Revenue >¥2T by FY2027, OP margin >= 18%

Thesis Status: HOLD

The thesis significantly strengthens with the addition of AI server MLCCs as a co-equal growth driver alongside automotive/EV. The "structural volume tailwind" now has two legs: (1) EV content growth (9,000+ MLCCs to 12,000+) and (2) AI server demand (30,000 MLCCs per server, 30% CAGR). Tariff and front-loading risks are near-term headwinds but the structural story is intact. April 30 earnings will be critical for FY2027 guidance.

Sources

• Murata Q3 FY2026 Earnings
• New MLCC Factory (Murata IR)
• New Automotive MLCCs (BusinessWire)
• Price Hikes Confirmed (TrendForce)
• AI Server MLCC Demand (Investing.com)
• StockAnalysis 6981.T

Cross-Validation Errata

Methodology note: J-Quants forward PE is computed from company guidance divided by 9-month period, creating distortion. StockAnalysis uses full TTM and analyst consensus forward estimates, which are more market-standard. We use StockAnalysis values for display.

Key Upcoming Catalysts

*Report generated: 2026-04-14 | Data sources: J-Quants (fundamentals), StockAnalysis (cross-validation), Fanuc IR, Advantest IR, Murata IR, TrendForce, Bloomberg, Reuters*

REPORT: deep_evolve_dd_2026-04-15.md

Deep DD Evolution — 2026-04-15

Time: 21:31

Type: Scheduled deep DD evolution (analyst-level research, not just metric updates)

Stocks updated: 6506.T (Yaskawa Electric), 6861.T (Keyence), 4063.T (Shin-Etsu Chemical)

Data sources: J-Quants (paid plan), StockAnalysis.com, company IR filings, web search (EN + JP)

Previous data_date: 2026-04-12 → Updated to: 2026-04-15

Verdict: All 3 stocks UPGRADED

Changes Summary

1. Yaskawa Electric (6506.T) — UPGRADED

Financial Update — FY2026 Actuals (Released April 10, 2026)

*NI decline is one-time distortion from FY2025 gain on Yantai Dongxing Materials transfer.

FY2027 guidance rationale: "Strong order intake driven by robust demand mainly from AI- and semiconductor-related markets."

Segment Deep Dive (FY2026)

Critical finding: Robotics now 45.6% vs Motion Control 44.2% — nearly equal, not the 55%/35% split previously modeled.

New Research Findings

NVIDIA Partnership — CONFIRMED (HIGH)

GTC 2026 (Mar 16): Yaskawa named "Big Four" for Physical AI. Jetson modules in controllers, Omniverse/Isaac integration, GR00T N1.7 early access.

Source: https://nvidianews.nvidia.com/news/nvidia-and-global-robotics-leaders-take-physical-ai-to-the-real-world

SoftBank Physical AI — CONFIRMED (HIGH)

MOU Dec 2025. Mar 2026: MOTOMAN NEXT validated in SoftBank logistics warehouse with VLM+VLA on AI-RAN MEC.

Source: https://www.yaskawa-global.com/newsrelease/news/178574

$180M Wisconsin Campus — IN PROGRESS (HIGH for tariff hedging)

First US high-volume robot manufacturing. 800-900K sq ft, 700+ jobs. Property closed Feb 2026.

Source: https://www.yaskawa-global.com/newsrelease/news/176879

Section 232 Tariff — PENDING (HIGH risk)

Investigation deadline May 30, 2026. Mitigants: US-Japan Framework Agreement, Wisconsin campus.

Source: https://www.whitecase.com/insight-alert/trump-administration-initiates-section-232-investigation-robotics-and-industrial

Humanoid Thesis — ADVANCING (MEDIUM)

FY2026 filing: "deepen engagement in humanoid robot domain through verification of advanced actuator technologies." "Dash 35" mid-term plan May 2026 disclosure.

Metric Cross-Validation

2. Keyence (6861.T) — UPGRADED

Financial Update — Q3 FY2025 (Released January 29, 2026)

Q3 standalone acceleration: Revenue +11.4% YoY. Americas +15.0%, Asia +18.9%, Europe +2.2%, Japan +0.3%.

New Research Findings

Five Product Launches Q1 2026 (HIGH)

VS-G (AI vision + full-image storage), GX-1000 (3D printer — new market), LJ-X8000, VK-X4000, IV4 (AI vision 99.8% accuracy). ~30% of revenue from products <3 years old.

Source: https://www.vision-systems.com/cameras-accessories/image-sensors/article/55279655/keyence-launches-new-sensor-series-with-built-in-ai

Dividend +57% YoY (HIGH): ¥350 → ¥550 per share.

Tariff Risk LOW (MEDIUM): Fabless model — no manufacturing. Main risk: customer capex delays.

Japan FA CAGR 9.2% (HIGH): $16.84B (2025) → $28.57B (2031). Labor shortage + Green Transformation + TSMC Kumamoto.

Source: https://www.mordorintelligence.com/industry-reports/japan-factory-automation-and-industrial-controls-market-industry

Moat Intact (MEDIUM): Keyence + Cognex ~50% global machine vision. 83%+ gross margin. Chinese players gaining in domestic low-tier only.

Metric Cross-Validation

3. Shin-Etsu Chemical (4063.T) — UPGRADED

Financial Update — Q3 FY2026 (Released January 27, 2026)

Electronics Materials carried growth (+5.9% YoY, ~34.8% OP margin). PVC dragged. TTM revenue: ¥2.57T — ahead of ¥2.4T guidance (conservative).

New Research Findings

CEO "AI Stock" Pivot (HIGH)

President Saito repositioning Shin-Etsu from "chemical stock" to "AI stock." ¥83B Isesaki photoresist plant operational since June 2025 — first domestic plant in 56 years.

Source: https://finance.biggo.com/news/gC7rUp0Bq7sy_YQMAiJd

GaN-on-QST World Record (HIGH — long-term optionality)

IMEC achieved 800V+ breakdown on Shin-Etsu 300mm QST substrates (Nov 2025). Targeting 1200V for AI data center power. Sample phase — pure upside option.

Shareholder Returns 5.9% Total Yield (HIGH)

¥500B buyback authorized, ¥400B completed. Payout ratio raised to 40%. 1.6% div + 4.3% buyback = 5.9%.

Source: https://www.ainvest.com/news/shin-etsu-chemical-400-billion-yen-buyback-strategic-masterstroke-investors-2505/

Shintech $3.4B US Expansion (MEDIUM — long duration)

Louisiana: +625K t/yr ethylene, +500K t/yr VCM, +310K t/yr caustic soda. US-based = tariff insulated. Complete 2030.

Source: https://www.shinetsu.co.jp/en/news/news-release/shintech-announces-capital-investment-of-3-4-billion-to-bolster-pvc-and-caustic-soda-business/

300mm Wafer Demand +7% (HIGH)

2025 global wafer shipments 13,076 MSI (+5.4%). AI/HPC wafer segment +18.94% CAGR to 2031. TSMC CapEx $52-56B = wafer pull-through. Crystal-pulling lags → ASP uplift on specialty wafers.

Source: https://www.semi.org/en/semi-press-release/semi-reports-2025-annual-worldwide-silicon-wafer-shipments-and-revenue-results

PVC Headwind Turning (MEDIUM)

US PVC +12.5% in March 2026 on Gulf supply disruption. Shintech = "one of few reliable global supply hubs."

Metric Cross-Validation

Upcoming Catalysts

*Report generated 2026-04-15 by Claude. All claims sourced. Cross-validated against StockAnalysis.com and company IR filings.*

REPORT: deep_evolve_dd_2026-04-17.md

Deep DD Evolution — 2026-04-17

Time: 14:08

Type: Scheduled deep DD evolution (analyst-level research, not metric-only updates)

Stocks updated: 6857.T (Advantest), 6954.T (Fanuc), 8035.T (Tokyo Electron)

Data sources: J-Quants (paid plan, official JPX filings), StockAnalysis.com, company IR/press releases, web search (EN + JP), industry filings (SEMI, USITC)

Previous data_date: 2026-04-12 → 2026-04-16 → Updated to 2026-04-17

Verdict: 1 UPGRADED (8035.T), 2 HOLD with qualifiers (6857.T, 6954.T)

Changes Summary

1. Advantest (6857.T) — HOLD (qualified UPGRADE)

Thesis Validation

Monopoly position confirmed. FY2025 guidance ¥1.07T revenue (raised 3x) and 42.4% op margin remain base case. Capacity ramp to 5,000 systems/yr by Mar 2027 on-track; potential 7,500-system stretch target. TSMC CoWoS packaging booked through 2026 and SK Hynix/Micron HBM supply sold out → test demand is supply-constrained through 2026.

What Changed Since Last DD (2026-04-14)

Teradyne NVIDIA GPU qualification — did NOT close H1 2026. Search through Apr 17 shows Teradyne collaborating with NVIDIA on Physical-AI / robotics, not core merchant GPU test. Risk deferred to H2 2026, not materialized.

Source: https://seekingalpha.com/article/4837312-advantest-teradyne-market-share-shifts-in-this-duopoly

SK Hynix in-house HBM4 tester — CONFIRMED (Jan 2026) but narrower than feared. SK Hynix qualified a proprietary HBM4 system-level tester Jan 2026 with TSMC. This is SiP/post-integration tester, not SoC replacement. Estimated TAM impact: -2 to -3%.

Source: https://www.trendforce.com/news/2026/02/25/news-sk-hynix-reportedly-develops-hbm4-system-level-testing-equipment-deepens-tsmc-collaboration/

NVIDIA capex — structurally bullish. Q1 FY2026 data-center rev $62B; hyperscaler capex ~$700B in 2026 (Meta $135B, Google $185B) vs ~$400B in 2025.

Source: https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2026

Strategic Innovation Centers (Apr 14, 2026): Sunnyvale/San Jose labs for AI-test partner collaboration. Reinforces customer lock-in.

Source: https://www.advantest.com/en/news/2026/20260414.html

Metric Cross-Validation

Updated WHY / RISK / TRIGGER

WHY: SoC ATE monopoly (~58-60% share, ~70% in AI-GPU niche) with 49% ROE and 43% op margin. Supply-constrained by TSMC/SK Hynix test demand. Teradyne qualification deferred; SK Hynix in-house tester narrow.

RISK: (1) PE 63x / PB 26x leaves no margin for NVIDIA capex deceleration. (2) Teradyne merchant-GPU qualification now H2 2026 binary event. (3) Capacity at 5K systems assumes >80% attach; any slowdown dilutes margin 300-500bps. (4) Convertible bond dilution overhang.

TRIGGER: Apr 27 FY2026 initial guidance — mid-teens+ growth = add on weakness. Flat/"digesting" language = trim. VOICE 2026 (May 18-20) new-product announcements.

Key Dates

• 2026-04-27 (earnings): FY2025 + FY2026 initial guidance — critical
• 2026-05-18 to 05-20 (catalyst): VOICE 2026, Scottsdale
• 2026-06-30 (rebalance): Semiannual rebalance
• 2026-07-15 (monitoring): NVIDIA Q2 FY2027 capex commentary
• 2026-12-31 (monitoring): Teradyne merchant-GPU qualification decision window

2. Fanuc (6954.T) — HOLD (no adds pre-May 30)

Thesis Validation

Picks-and-shovels monopoly intact. ~60% global CNC share, ~25% industrial robot share, zero-debt. FY2026 ordinary income guidance ¥2,148B (+9.2% YoY). Q3 cumulative +14.2% YoY.

What Changed Since Last DD (2026-04-16)

China +87% YoY robot sales — validated. China now >25% of robot revenue. March 2026 China mfg PMI 50.4 (expansion, from 49.0 Feb). Rebound is happening.

Source: https://global.morningstar.com/ja/stocks/fanuc-earnings-strong-china-robot-demand-supports-long-term-view-fair-value-up-5-jpy-5460

NVIDIA Physical-AI partnership LIVE (Mar 16, 2026). Jetson + Isaac Sim + Omniverse integration into Fanuc robots and ROBOGUIDE simulator. MODEX 2026 (Apr 13-16) showcased CRX-30iA on OTTO 600 AMR.

Source: https://www.prnewswire.com/news-releases/fanuc-accelerates-physical-ai-in-industrial-robotics-leveraging-nvidia-technologies-302714972.html

$90M Michigan plant (Mar 24, 2026). 840K sq ft, 225 jobs, completion late 2027. Management betting its own capital against Section 232 tail risk.

Source: https://www.prnewswire.com/news-releases/fanuc-america-announces-90-million-investment-to-create-production-ready-capacity-for-robot-manufacturing-in-the-us-302722608.html

Section 232 deadline May 30 — still binary. Commerce report on robots/CNC due May 30, 2026. 25% tariff could cost 5-10% of FY2026 EPS. No leaked Japan exemption.

Source: https://www.whitecase.com/insight-alert/trump-administration-initiates-section-232-investigation-robotics-and-industrial

Hannover Messe 2026 (Apr 20-24). Fanuc confirmed exhibitor; watch for CRX/humanoid demos during show window.

Metric Cross-Validation

Updated WHY / RISK / TRIGGER

WHY: ~60% CNC / ~25% robot global share, zero debt, 20.5% op margin. China +87% YoY validated. NVIDIA Physical-AI partnership accretive. Michigan plant = self-funded tariff hedge.

RISK: (1) Section 232 binary — May 30 Commerce recommendation + ~90-day implementation. 25% tariff = 5-10% FY2026 EPS hit. (2) Japan exemption status unknown. (3) Apple/CRX pipeline still unconfirmed in filings. (4) ROE 8.6% below cost of capital.

TRIGGER: Apr 25 earnings — FY2027 guidance + tariff-scenario language critical. May 30 Section 232 = add-or-trim inflection. Hold into May 30; do NOT add on earnings beat until tariff resolves.

Key Dates

• 2026-04-20 to 04-24 (catalyst): Hannover Messe 2026
• 2026-04-25 (earnings): FY2025 full-year + FY2026 guidance w/ tariff language
• 2026-05-30 (catalyst — BINARY): Section 232 Commerce report
• 2026-06-30 (rebalance): Semiannual rebalance
• 2026-08 (monitoring): Potential Presidential tariff implementation window
• late-2027 (monitoring): Michigan facility operational

3. Tokyo Electron (8035.T) — UPGRADED

Thesis Validation

#3 global WFE position intact, AI pivot accelerating. Coater/developer ~90% share uncontested. FY2025 revenue ¥2.43T (+32.8% YoY), OP +50.2%, NI +49.5% — all-time highs. FY2026 guidance ¥2.35T (+6.9%) with H2 expected as record. AI mix on-track 40%.

What Changed Since Last DD (2026-04-12)

TSMC Q1 2026 capex high-end $52-56B (Apr 16). CEO C.C. Wei flagged 2026 revenue >30% growth; Q2 NT$39-40.2B (+10% QoQ). TSMC = TEL's largest customer; structurally bullish for foundry equipment demand into 2026-2027.

Source: https://www.cnbc.com/2026/04/16/tsmc-q1-profit-58-percent-ai-chip-demand-record.html

Samsung/SK Hynix HBM4 ramp. Samsung targeting ~250K wafer/mo by end-2026 (+47%); HBM4 mass production Q2 2026; HBM shipments tripling in 2026. Memory equipment ~40% of TEL revenue.

Source: https://www.datacenterdynamics.com/en/news/samsung-and-sk-hynix-to-scale-up-memory-production-capacity-in-2026-to-meet-ai-demand/

SEMI 2026 WFE forecast +9.0% to $135.2B, memory equipment +12.7% to $15.7B. Stronger tailwind than prior framing.

Source: https://www.semi.org/en/semi-press-release/global-total-semiconductor-equipment-sales-forecast-to-reach-a-record-of-dollar-139-billion-in-2026-semi-reports

ASML Q1 2026 raised 2026 guidance to €36-40B; memory 51% of sales; China fell to 19%. Parallel indicator for Japanese WFE peers.

Source: https://www.asml.com/en/investors/financial-results/q1-2026

Austin RiverSouth HQ inaugurated (Apr 16). CEO Toshiki Kawai at opening; US footprint signal ahead of any BIS changes.

Source: https://www.semi.org/en/news-resources/press/tel-new-hq-austin-tx

BIS enforcement intensity up. Applied Materials $252M penalty, Cadence $95M (Feb 2026). Japanese firms historically under more lenient rules — watch METI alignment.

Source: https://www.mondaq.com/unitedstates/export-controls-trade-investment-sanctions/1772088/biss-2026-enforcement-agenda-takes-shape

Metric Cross-Validation

Share price ¥44,350 (down from Feb peak ¥46,600) — pullback creates entry.

Updated WHY / RISK / TRIGGER

WHY: #3 global WFE, ~90% coater/developer monopoly. FY2025 all-time-high rev/OP/NI. TSMC $52-56B 2026 capex + Samsung/SK Hynix HBM4 ramp = sustained demand through H2 2026 and 2027. AI equipment 40% of sales de-risks China 30% exposure. Record dividend ¥601/share +23% YoY, zero-debt.

RISK: (1) China 30% exposure still tail risk if METI aligns fully with BIS. (2) PE 41.9x / PB 10.5x leaves little room if FY2026 H1 weakness extends. (3) Memory capex tied to NVIDIA GPU demand. (4) Q3 FY2026 revenue -15.7% YoY; H2 ramp must materialize.

TRIGGER: May 12 FY2025 confirmation + FY2026 initial guidance — watch H2 ramp language and AI-equipment % disclosure. Add on post-earnings dip to ¥42-43K. Trim if China revenue moves toward 40% or TSMC capex guidance is cut.

Key Dates

• 2026-05-12 (earnings): FY2025 + FY2026 initial guidance — critical
• 2026-06 (catalyst): Annual shareholder meeting — dividend policy, capex
• 2026-06-30 (rebalance): Semiannual rebalance
• 2026-08-06 (earnings): FY2026 Q1 — first H2 ramp validation
• 2026-09 (monitoring): SK Hynix HBM4 mass production phase-2
• 2026-12 (monitoring): 2027 capex guidance pre-announcement window

Methodology Notes

• J-Quants used for raw financials. PE/PB from J-Quants cross-checked against StockAnalysis.com; StockAnalysis preferred for display (split-adjusted, rolling TTM).
• Web research executed via parallel Explore subagents searching EN + JP sources (Kabutan, Nikkei, Morningstar, company IR, SEMI, USITC, BIS).
• Thesis validation per stock: verified original WHY still holds, flagged incremental findings (confirmed/weakened/new), graded materiality (HIGH/MEDIUM/LOW).
• Per-stock update_history[] entry added to each theme JSON with thesis_status tag; theme recent_updates[] updated.

Disclaimer

This report is for research and educational purposes only. Not investment advice. Always do your own due diligence.

REPORT: hbm_memory_supply_chain_dd_2026-04-13.md

HBM & AI Memory Supply Chain: Japan Deep Dive

Time: 01:59

Date: 2026-04-13

Theme: High Bandwidth Memory (HBM), AI Server Components, Advanced Packaging Materials

Catalyst: HBM demand growing 5-10x driven by NVIDIA GPU buildout; HBM to consume 23% of DRAM wafers in 2026

Executive Summary

The AI infrastructure buildout is creating a massive demand pull through Japan's semiconductor materials and components supply chain. HBM revenue is expected to surpass 30% of all DRAM revenue by 2026, even though it accounts for only ~8% of total DRAM output. CoWoS capacity remains sold out through mid-2026. Japanese companies sit at critical chokepoints: substrates (Ibiden), photoresists (TOK), passive components (Murata, TDK), semiconductor materials (Resonac), and optical connectivity (Fujikura).

Key finding: JSR (4185) and Shinko Electric (6967) are both delisted -- JSR in June 2024 (JIC acquisition) and Shinko Electric in June 2025 (JIC/DNP/Mitsui consortium). Fujikura (5803) emerges as the top Japanese optical/AI connectivity play.

1. Ibiden (4062.T) -- FC-BGA Substrates for NVIDIA GPUs

Company Profile

IC packaging substrate and PCB manufacturer. Key NVIDIA supplier for FC-BGA substrates used in GPU/AI accelerator packaging. Commands >50% market share in server IC substrates.

Fundamentals (April 2026)

AI/HBM Revenue Exposure

• AI semiconductor substrates currently >15% of total revenue (~370B JPY total sales)
• Electronics division revenue up 15.6% YoY to 113.5B JPY; operating income +41.4% YoY
• AI share expected to increase significantly with NVIDIA Blackwell ramp
• GenAI projected to triple IC substrate sales vs FY2024 baseline

Capacity Expansion

• New substrate factory in Gifu Prefecture: 25% capacity late 2025, ramping to 50% by March 2026
• Plans to expand from 2 sites/3 plants to 3 sites/5 plants by end FY2025
• $1.2B flip-chip substrate fab in Phoenix, Arizona (targeting 30K panels/month by late 2027, with $320M CHIPS Act grants)
• Production output target: 2.5x 2024 level by FY2027

Bull Case

• Dominant market position (>50% share in server IC substrates)
• Direct NVIDIA supplier with deep relationship
• Massive capacity expansion underway with government support in both Japan and US
• AI substrate demand at full capacity; robust orders expected to continue

Bear Case

• Extreme stock run (+328% in 1 year) creates valuation risk
• P/E of 27x prices in significant growth; any demand slowdown would be punished
• Heavy capex cycle ($1.2B Phoenix fab) creates execution risk
• Customer concentration risk (NVIDIA dependence)
• Non-AI substrate demand remains weak

Sources

• Yahoo Finance - Ibiden
• Ibiden Weighs Capacity Boost - Yahoo Finance
• Ibiden Accelerates Capacity Expansion - GuruFocus
• NVIDIA GB200 Supply Chain - IntuitionLabs
• Ibiden Raises Forecast - Investing.com
• Ibiden Integrated Report 2025

2. Shinko Electric Industries (6967.T) -- DELISTED June 2025

Status: NOT INVESTABLE

Shinko Electric was delisted from the Tokyo Stock Exchange on June 6, 2025 following a tender offer by a JIC-led consortium including Dai Nippon Printing (DNP, 15% stake ~85B JPY) and Mitsui Chemicals. The company was formerly a Fujitsu subsidiary specializing in IC packaging substrates.

Why It Matters (For Supply Chain Context)

• Was among top 5 ABF substrate makers globally (with Unimicron, Ibiden, AT&S, Nan Ya PCB)
• Commissioned new plant in Osaka boosting ABF output by 30%
• AI server substrate revenue was ~3x YoY in FY2024
• Now privately held under JIC/DNP/Mitsui -- no public market access

Last Known Fundamentals (Before Delisting)

Sources

• JPX Delisting Decision - Shinko Electric
• Shinko Electric Delisting - TipRanks
• Shinko Electric Delist + DNP/Mitsui Partnership - Digitimes

3. Tokyo Ohka Kogyo / TOK (4186.T) -- EUV/ArF Photoresists

Company Profile

Japan's leading photoresist manufacturer. Photoresists are essential for patterning circuits in HBM DRAM production and advanced logic chips. TOK supplies EUV and ArF photoresists to all major foundries/memory makers.

Fundamentals (April 2026)

AI/HBM Revenue Exposure

• Semiconductor photoresists projected to grow 150% driven by AI chip demand
• Packaging materials to expand 70%; high-purity chemicals +50%
• Back-end process materials (relevant to HBM assembly) projected +30%
• FY2025: Record revenue of 237.0B JPY (+17.9% YoY), operating income 47.4B JPY (+43.2%)
• FY2026 guidance: Revenue 261.0B JPY (+10.1%), operating income 52.2B JPY (+10.2%)

Recent Developments

• Strategic investment in Irresistible Materials for next-gen EUV photoresist technology
• Developing photoresists for 2nm node (commercial target ~2030)
• Medium-Term Plan 2027 targets ROE of 14.0%

Bull Case

• Near-monopoly position in advanced photoresists (alongside JSR, now delisted)
• Every HBM chip requires multiple lithography steps using TOK's materials
• Strong ROE (15.6%) with margin expansion trajectory
• JSR delisting/privatization removes a key competitor from public markets
• Recurring revenue model -- materials consumed in production

Bear Case

• Stock trading above analyst consensus target (8,833 vs 7,671 avg target)
• FY2026 growth guidance decelerating to ~10% from 18% (still strong but slowing)
• Sensitivity to semiconductor capex cycles
• Competition from Shin-Etsu Chemical in specialty chemicals

Sources

• TOK FY2025 Record Profits - Investing.com
• TOK EUV Partnership - PR Newswire
• Japan Ramps Up Photoresist Investment - TrendForce
• Yahoo Finance - 4186.T

4. JSR Corporation (4185.T) -- DELISTED June 2024

Status: NOT INVESTABLE

JSR was delisted from the Tokyo Stock Exchange on June 25, 2024 following acquisition by Japan Investment Corporation (JIC) for ~$6.3 billion. Corporate merger finalized December 1, 2024. CEO Eric Johnson has stated intent to relist in 5-7 years.

Why It Matters

• Was one of the world's top photoresist makers alongside TOK
• Critical EUV photoresist supplier
• JIC acquisition part of Japan's national strategy to secure semiconductor supply chain sovereignty
• JSR's exit from public markets consolidates the photoresist investment thesis around TOK (4186.T)

Sources

• JIC Acquires JSR - Digitimes
• JSR Delisting Announcement - JPX
• JSR Acquisition Details - SunsetHQ

5. Nitto Denko (6988.T) -- Semiconductor Process Materials

Company Profile

Diversified materials company. Key products include semiconductor manufacturing tapes, flexible printed circuits, optical films for displays, and release technologies for advanced packaging. UV release technology supports wafer thinning for HBM stacking.

Fundamentals (April 2026)

AI/HBM Revenue Exposure

• Raised full-year FY2026 guidance to 1.027T JPY revenue on stronger-than-expected IT/smartphone materials demand
• Semiconductor process tapes are used in HBM die thinning and stacking -- UV release technology critical for advanced packaging
• Flexible PCBs used in high-density server interconnects
• AI/HBM-specific revenue percentage not separately disclosed, but electronics segment is the backbone

Bull Case

• Attractive valuation: forward P/E of 11.4x with 12.9% ROE
• Diversified revenue base reduces cyclical risk
• Critical process materials for HBM manufacturing (wafer thinning tapes)
• Guidance upgrade signals confidence in demand trajectory

Bear Case

• AI/HBM exposure is indirect and hard to quantify
• Consumer electronics (smartphone, display) segments face headwinds
• Lacks the "pure play" AI narrative that drives premium valuations
• Moderate growth rate relative to pure semiconductor plays

Sources

• Yahoo Finance - 6988.T
• Nitto Denko Upgrades Guidance - Simply Wall St
• StockAnalysis - Nitto Denko

6. Nidec (6594.T) -- Datacenter Storage & Cooling

Company Profile

World's largest precision motor manufacturer. AI exposure through: (1) Nearline HDD spindle motors for datacenter storage, (2) cooling solutions for AI servers, (3) EV traction motors. Recent accounting concerns cloud the investment case.

Fundamentals (April 2026)

AI/HBM Revenue Exposure

• Nearline HDD demand for data centers is a key driver -- forecast revised upward to 67M units
• AI cooling solutions and precision motors for server infrastructure
• Record sales and profit in FY2025, driven by AI and automotive
• Expanding into EV markets alongside datacenter -- dual AI/EV thesis

Risk Factors -- CRITICAL

• Accounting investigation ongoing: suspected inappropriate accounting practices with potential material misstatements requiring amendments to prior financials
• Shareholder lawsuit filed
• These governance issues materially increase risk

Bull Case

• Cheap on forward P/E (10.8x) if accounting issues resolve cleanly
• Dual tailwinds: AI datacenter + EV motors
• Nearline HDD demand structural growth with AI data explosion
• AI server cooling is emerging high-growth segment

Bear Case

• Accounting investigation is a major red flag -- potential restatements
• Low ROE (6.3%) suggests operational inefficiency
• Complex conglomerate structure makes analysis difficult
• EV motor business faces intense competition (Chinese makers)
• Governance track record issues (founder control dynamics)

Sources

• Yahoo Finance - 6594.T
• Nidec Earnings - Quartr
• StockAnalysis - Nidec

7. Murata Manufacturing (6981.T) -- MLCCs for AI Servers

Company Profile

World's largest MLCC (multilayer ceramic capacitor) manufacturer with 30-35% global market share. AI servers use 8x more MLCCs than traditional servers. Also entering AI server power modules.

Fundamentals (April 2026)

AI/HBM Revenue Exposure

• MLCC demand for AI servers projected 3.3x by FY2030 vs FY2025
• Raised MLCC growth forecast for AI servers to 30% CAGR (2025-2030), up from prior 18% CAGR
• AI server power module (VPD) mass production starting 2026 -- targeting ~50B JPY revenue over FY2026-27
• Exploring MLCC price increases for AI server components
• Murata + Samsung control 80% of global MLCC supply

Capacity & Strategic Moves

• Mass production of VPD power modules for AI servers begins 2026
• Coordinating directly with hyperscalers (Meta, Google, Microsoft, Amazon)
• MLCC supply tightening as demand outpaces capacity additions

Bull Case

• Duopoly with Samsung (80% combined share) = extreme pricing power
• AI servers use 8x more MLCCs -- structural demand multiplier
• New power module business adds TAM (~50B JPY near-term)
• Price increase potential as demand outstrips supply
• High barriers to entry in high-spec MLCCs

Bear Case

• Smartphone segment (~50% of revenue) faces cyclical headwinds
• P/E of 22.6x already reflects some AI premium
• Moderate ROE (8.1%) for a market leader
• China's Samsung could gain share in lower-spec MLCCs
• EV adoption slowdown would hurt automotive MLCC demand

Sources

• Murata MLCC Demand to Double - TrendForce
• Murata Power Modules 2026 - TrendForce
• Murata Raises MLCC Forecast - Investing.com
• MLCC Shortages for AI - Digitimes

8. TDK Corporation (6762.T) -- Power Components & Sensors for AI

Company Profile

Electronic components conglomerate. Products include passive components (capacitors, inductors, ferrites), magnetic heads for HDDs, rechargeable batteries (via subsidiary TDK-Lambda), and sensors. AI exposure through datacenter power components and HDD suspension assemblies.

Fundamentals (April 2026)

AI/HBM Revenue Exposure

• AI-related revenues >10% of total sales in FY2025, with 25-30% CAGR projected
• Nearline HDD suspension assemblies for data centers -- forecast revised upward to 67M units
• Point-of-load (PoL) converters designed for AI servers (up to 25A)
• New stackable uPOL modules providing up to 200A for vertical power delivery in AI servers
• Record sales and all profit items at half-year; raised full-year guidance
• Apple partnership for US-made components (additional revenue stream)

Bull Case

• AI revenue growing 25-30% CAGR from current ~10% base
• Diversified across AI value chain: power, storage, sensors
• New uPOL power modules targeting AI server power delivery
• Reasonable valuation at 17.2x P/E for growth profile
• Apple US manufacturing partnership adds visibility

Bear Case

• AI still only ~10% of revenue -- not a pure play
• Battery segment faces EV competition from Chinese makers
• Conglomerate discount appropriate for complex structure
• HDD business faces long-term secular decline (SSD replacement)
• Moderate ROE (8.3%)

Sources

• TDK Q2 2026 Earnings - Alpha Spread
• TDK uPOL Launch
• TDK AI Ecosystem Strategy
• TDK Investor Day 2025

9. Fujikura (5803.T) -- Optical Fiber & Connectivity for AI Datacenters

Company Profile

Optical fiber cable manufacturer and flexible printed circuit (FPC) maker. Emerged as Japan's premier AI datacenter connectivity play. Among the top 3 global optical fiber cable producers. Stock surged ~1,400% over 2 years on AI datacenter demand.

Fundamentals (April 2026)

AI/HBM Revenue Exposure

• Optical fiber is the physical backbone connecting GPUs in AI clusters
• Plans to triple optical fiber output for US AI infrastructure
• $200M dividend from US subsidiary signals US datacenter revenue strength
• FPC (flexible printed circuits) used in server interconnects
• Direct beneficiary of hyperscaler capex ($300B+ annually)

Bull Case

• Physical infrastructure monopoly -- you can't train AI without fiber
• Tripling capacity signals multi-year demand visibility
• US infrastructure buildout creates geographic diversification
• 6-for-1 stock split improves retail accessibility
• EPS growth of 40% supports premium valuation

Bear Case

• P/E of 37.7x is expensive -- priced for perfection
• 1,400% 2-year run creates extreme downside risk on any demand miss
• Optical fiber is ultimately a commodity with competition (Corning, Prysmian)
• Capacity expansion requires heavy capex
• AI capex could face a cyclical pullback

Other Japanese Optical Plays (For Reference)

• Mitsubishi Electric -- mass-producing 200Gbps EML chips for datacenter optical transceivers
• Sumitomo Electric (5802.T) -- optical fiber, but less AI-specific
• Furukawa Electric (5801.T) -- optical fiber cable maker

Sources

• Fujikura Surges 160% in 2025 - FastBull
• AI Demand Strains Fujikura - Bloomberg
• Fujikura -- Japan's Silent Giant - DividendJapan
• Fujikura Market Cap $33B - Techmeme

10. Resonac Holdings (4004.T) -- Semiconductor Materials for HBM

Company Profile

Formerly Showa Denko, merged with Hitachi Chemical in Jan 2023. World-class semiconductor materials company: bonding films, encapsulants, die bonding materials, thermal interface materials (TIM), non-conductive films (NCF) -- all critical for HBM packaging.

Fundamentals (April 2026)

AI/HBM Revenue Exposure

• NCF (non-conductive film) and TIM (thermal interface materials) are core HBM packaging materials
• Expanding NCF/TIM production capacity from 3.5x to 5x baseline
• Semiconductor & Electronic Materials segment target: >50% of revenue, 20% EBITDA margin
• FY2025 Semi/Electronic Materials segment projected at 499.0B JPY
• Back-end materials subsegment has highest assumed growth rate in company forecast
• Developed temporary bonding film + photonic debonding process for advanced 2.5D/3D packaging (HBM, chiplets)
• Partnership with PulseForge for xenon flash debonding technology

Bull Case

• Cheapest stock in this universe (P/E ~7-17x) -- deep value play
• Direct HBM materials exposure: NCF and TIM are consumed with every HBM stack
• 5x capacity expansion for AI materials shows management conviction
• Post-merger integration creating a semiconductor materials powerhouse
• Strategic positioning in China's AI self-reliance push (additional demand)

Bear Case

• Low ROE (5.7%) -- still integrating Showa Denko + Hitachi Chemical
• Chemicals segment drags on margins and growth narrative
• Complex conglomerate with legacy businesses
• Integration execution risk continues
• Commodity chemicals exposure creates earnings volatility

Sources

• Resonac Q3 2025 Results
• Resonac HBM Bonding Film
• Resonac + PulseForge Partnership
• Resonac AI Self-Reliance Thesis - AInvest

Comparative Summary Table

Investment Thesis Tiers

Tier 1: Highest Conviction (Direct HBM/AI Substrate Plays)

1. Ibiden (4062.T) -- Dominant NVIDIA substrate supplier, massive capacity expansion. Expensive but justified by position.

2. TOK (4186.T) -- Photoresist monopoly. Every HBM chip needs TOK materials. Best ROE in the group (15.6%).

3. Murata (6981.T) -- MLCC duopoly with pricing power. New power module business. Largest market cap = most liquid.

Tier 2: Strong Exposure, Reasonable Valuation

4. Resonac (4004.T) -- Cheapest stock, direct HBM materials (NCF/TIM). Deep value if integration succeeds.

5. TDK (6762.T) -- Diversified AI exposure at reasonable 17x P/E. Power delivery products gaining traction.

6. Nitto Denko (6988.T) -- Best valuation metrics (forward P/E 11.4x, ROE 12.9%). Indirect but real HBM exposure.

Tier 3: High Risk / Thematic

7. Fujikura (5803.T) -- Best pure-play AI connectivity. Extreme valuation risk after 1,400% run.

8. Nidec (6594.T) -- Avoid until accounting investigation resolves. Governance red flag.

Not Investable

• JSR (4185.T) -- Delisted June 2024 (JIC acquisition)
• Shinko Electric (6967.T) -- Delisted June 2025 (JIC/DNP/Mitsui consortium)

Key HBM Supply Chain Facts (2026 Context)

• HBM revenue to surpass 30% of all DRAM revenue in 2026
• HBM to consume 23% of DRAM wafer capacity in 2026
• AI to consume 20% of total DRAM production in 2026
• CoWoS capacity target: 120K wafers/month by end 2026 (still undersupplied)
• Data center capex surged 57% in 2025; expected to exceed $300B annually by 2026
• Each AI server uses 8x more MLCCs than a traditional server
• SK Hynix, Samsung, and Micron are the big 3 HBM producers; all sourcing materials from these Japanese suppliers

*Research compiled from public sources. Not investment advice. All data as of April 13, 2026. Verify current prices before trading.*

REPORT: japan_ai_supply_chain_13layers_2026-04-21.md

Japan's AI Supply Chain: Complete 13-Layer Deep DD

Time: 02:00

Date: 2026-04-21

Type: Comprehensive DD — all 13 layers of AI semiconductor supply chain

Scope: Japan vs Taiwan vs Korea vs US vs China competitive analysis

Companies: 30+ Japanese stocks mapped across full value chain

Executive Summary

Japan controls the "invisible infrastructure" of AI. While Taiwan (TSMC) manufactures chips, the US (Nvidia) designs them, and the Netherlands (ASML) provides lithography — Japan controls what goes INTO and AROUND every single step. Without Japan, the global semiconductor supply chain stops.

This report maps all 13 layers of the AI supply chain, identifies Japan's TRUE monopolies vs competitive positions, and recommends specific stocks based on moat durability.

Key finding: Japan holds 7 genuine chokepoints where no alternative supplier exists at scale.

The 13-Layer AI Supply Chain — Japan's Position

LAYER 1: Wafer Materials — Japan's Silicon Duopoly

The Moat: 50+ Years of Crystal-Growing Expertise

Japan combined: 51% of global 300mm silicon wafers.

Every advanced logic chip from TSMC, Samsung, Intel starts on a Japanese wafer. The edge is crystal-growing expertise — determining wafer flatness, defect density, and purity at the atomic level. This took 50+ years to develop.

vs Taiwan: GlobalWafers is #3 but building first major US plant ($5B Texas fab). Still years behind on ultra-flat 2nm-grade material.

**Source:** IntelMarketResearch, WaferPro Top 5

LAYER 2: Fab Equipment — Japan's 5 Equipment Monopolies

The Moat: Tools That ONLY Japan Makes

The key insight: ASML gets all the headlines, but ASML cannot sell a single EUV scanner without Tokyo Electron's coater/developer track. TEL is to ASML what tires are to a Ferrari. Similarly, without Lasertec's inspection tools, you cannot verify EUV masks work — no chip below 7nm ships.

vs Taiwan: Taiwan has ZERO significant semiconductor equipment companies. TSMC is a customer, not a competitor.

**Sources:** Nomad Semi — TEL Deep Dive, Oreate AI — Lasertec Monopoly, SemiAnalysis — DISCO

LAYER 3: Fab Process Materials — Japan Controls 95% of EUV Chemistry

The Moat: Molecular Trade Secrets, Not Patents

Japan's 95% EUV photoresist share is the most underappreciated chokepoint. These are molecular trade secrets — you cannot reverse-engineer a photoresist because performance depends on synthesis conditions, molecular weight distribution, and proprietary additives undetectable in the final product.

**Source:** Fountyl — Japan EUV Photoresist Monopoly, SemiAnalysis

LAYER 4: Chip Design — Japan's Weakest Layer

Japan lost chip design leadership due to keiretsu vertical integration (no fabless ecosystem), missed PC/mobile era, and no EDA/software ecosystem. Renesas is solid but not irreplaceable. Rapidus (2nm foundry) is a wildcard — mass production target H2 2027, but 2-3 generations behind TSMC.

LAYER 5: Photomask + EUV Mask Blanks

EUV Mask Blanks: Japan's Hidden 93% Monopoly

Hoya is the ONLY vendor qualified for High-NA EUV (0.55 NA, sub-2nm). 40+ alternating Mo/Si layers deposited with sub-angstrom precision on ultra-pure LTEM glass. No other country has this capability.

Merchant Photomask

Tekscend (spun from Toppan, IPO'd Oct 2025): Revenue ¥118B, OP margin 24%. Global merchant leader but not a monopoly (Photronics competes).

**Source:** Karim Almansour — Mask Blanks, SemiEngineering

LAYER 6: CMP Slurry — Shared Moat

Japan combined ~35%+ of CMP slurry. Fujimi dominates poly-Si segment. Real but shared moat — US players are strong competitors.

**Source:** JEES — Top CMP Slurry Manufacturers

LAYER 7: IC Substrates — Duopoly with Taiwan

For the most advanced AI server substrates, ONLY Ibiden and Unimicron have the capabilities — a quiet duopoly. Ibiden AI substrate revenue 3x YoY. MGC's BT resin has 45% share with enormous switching costs (12-18 month requalification).

**Source:** DigiTimes — IC Substrate 2026

LAYER 8: Glass Cloth / CCL — Nittobo's Absolute Monopoly

The single tightest chokepoint in AI outside ASML. Every AI chip substrate requires T-glass. CCL lead times: 6 months. Prices: +15-30%. No alternative. 3x capacity expansion won't normalize supply until 2027+.

**Source:** Tom's Hardware, DigiTimes

LAYER 9: PCB Manufacturing — Taiwan Dominates

Japan is NOT irreplaceable here. Taiwan dominates PCB manufacturing volume. Japan's edge is upstream materials (glass cloth, copper foil, resins) feeding into Taiwanese PCB makers.

**Source:** DigiTimes — Zhen Ding 2026

LAYER 10: Connectors / High-Speed Interconnects

Japan is NOT dominant in connectors — US companies (Amphenol, TE) lead. Hirose is a quality niche player. Fujikura is strong in optical fiber for AI data center interconnects (already in our portfolio, +160% in 2025).

LAYER 11: Passive Components — Japan's Other Monopoly

MLCC: Murata's 40% Dominance

Japan combined: ~60% of global MLCC. 10-year technology lead vs China (1,000 vs 300 ceramic layers). Every AI server needs 20,000-440,000 MLCCs. Price increases 15-35% underway.

Inductors: Japan's Hidden Dominance

Japan controls ~70% of high-end inductors. Less discussed than MLCCs but equally critical for GPU power delivery.

Power Semiconductors

Japan is competitive but NOT dominant in power semiconductors globally. Infineon leads. Japan's edge is automotive IGBT modules and emerging SiC/GaN.

LAYER 12: Cooling / Thermal Management

Taiwan dominates thermal management (~70% of thermal modules). Japan has niche positions — Nidec's CDU for AI servers, Furukawa's heat pipes. NOT a Japan moat.

LAYER 13: AI Platform / Infrastructure

US hyperscalers dominate 10-100x. Japan's AI platform layer is the weakest. NTT IOWN (photonic computing) is promising but years from commercialization.

JAPAN'S 7 TRUE CHOKEPOINTS — The Irreplaceable Moats

STOCK RECOMMENDATIONS — Final Rankings

TIER 1: TRUE MONOPOLY — Must Own

These companies have positions that CANNOT be replicated within 10+ years.

TIER 2: STRONG MOAT — Core Positions

These have defensible positions but face some competition.

TIER 3: GOOD POSITION — Tactical

Solid companies but replaceable or in competitive markets.

TIER 4: WATCH — Not Recommended Yet

JAPAN vs TAIWAN: Where Each Wins

The key insight: Japan and Taiwan are COMPLEMENTARY, not competitive. Japan makes the materials that Taiwan needs to manufacture. If Japan restricts materials exports, TSMC stops. If TSMC stops, Nvidia has no chips. The entire AI supply chain is a Japan-Taiwan symbiosis.

Sources (Complete List)

Layer 1-2 Sources

• IntelMarketResearch — Silicon Wafer Market
• Nomad Semi — Tokyo Electron Deep Dive
• Oreate AI — Lasertec EUV Monopoly
• SemiAnalysis — DISCO Corporation
• Brookings — Renaissance of Japan Semiconductor

Layer 3-5 Sources

• Fountyl — Japan EUV Photoresist Monopoly
• SemiAnalysis — EUV Photoresist Battle
• Karim Almansour — Mask Blanks
• SemiEngineering — EUV Mask Blank

Layer 6-9 Sources

• DigiTimes — IC Substrate 2026
• Tom's Hardware — T-Glass AI Shortage
• DigiTimes — CCL Price Hikes 2026
• Global Tech Research — Nittobo

Layer 10-13 Sources

• TrendForce — Murata MLCC AI Demand
• Bloomberg — Murata 35% Price Increase
• Signal Integrity Journal — Capacitors in AI
• DigiTimes — MLCC Orders 2x Capacity

Cross-cutting Sources

• AMRO — Japan's Strategic Comeback in Chip Race
• East Asia Forum — Why Japan Matters for TSMC
• Rapidus 2nm Progress

REPORT: murata_deep_dd_2026-04-21.md

Murata Manufacturing (6981.T) — Deep DD: The Salt of AI

Time: 09:30

Date: 2026-04-21

Type: Deep DD — MLCC global dominance and AI criticality

Why "The Salt of AI"

MLCCs are less than 2% of an AI server's bill of materials cost — but 100% necessary. Without them, the GPU can't function. Murata controls 40% of global supply. This is the "salt of electronics" — cheap per unit, absolutely essential, and controlled by one company.

The Numbers That Matter

Why AI Needs So Many Capacitors

A GPU drawing 1,400W (GB300 TDP) creates violent current transients during inference/training. MLCCs serve three critical functions:

1. Bulk decoupling — energy storage near the GPU for instant current delivery

2. Mid-frequency decoupling — responding to sudden load changes in nanoseconds

3. High-frequency bypass — suppressing switching noise below 10mV tolerance

Poor power quality causes ~40% of unplanned server outages. Every additional GPU watt requires MORE decoupling capacitors. As GPU power doubles each generation, MLCC content grows proportionally.

MLCC Content Growth

Financial Snapshot

Pricing Power — April 2026 Price Hikes

Murata SETS the price. Everyone follows. This is oligopoly pricing power.

Drivers: silver price surge (electrode material) + AI demand (orders 2x capacity) + capacity utilization >80% industry-wide.

Competitive Moat — Why China Can't Catch Up

Key barrier: it's not just equipment — it's materials science (ceramic powder formulation), manufacturing precision (nano-layer stacking), and quality control (yield at extreme specs). Chinese firms are poaching Japanese engineers but institutional knowledge is hard to transfer.

The Real Competitor: Samsung EMCO

Samsung Electro-Mechanics is closing the gap in AI server MLCCs:

• AI share: ~40% vs Murata's ~45%
• Advantage: makes own barium titanate powder (vertical integration)
• Challenge: still behind in overall technology breadth and yield

Murata's response: MF Material JV with Ishihara Sangyo and Fuji Titanium for captive BaTiO3 supply by 2027.

New Growth: AI Server Power Modules

Murata is moving UP the value chain — from components to modules:

• AI server power supply units (5.5kW with GaN technology from Rohm)
• Target: ¥50B revenue by FY2027 from power modules
• Mass production starting 2026
• This is a much higher-margin business than individual MLCCs

Risks

Bull vs Bear

Bull (60% probability):

• 30% CAGR for AI MLCC through 2030
• Price increases stick and expand margins back toward 20%+
• Power module business adds ¥50B revenue
• Forward PE 18.2x re-rates to 22-25x on earnings growth
• Target: ¥6,000-7,000 (25-45% upside)

Bear (20% probability):

• China geopolitical escalation disrupts 48% of revenue
• Samsung gains AI share parity, pricing power erodes
• AI capex cycle peaks mid-2027, MLCC correction follows
• Target: ¥3,200-3,500 (25-30% downside)

Base (20% probability):

• Steady growth, margins stable at 17-18%
• AI growth offsets smartphone decline
• Target: ¥5,000-5,500 (5-15% upside)

Conviction: HIGH — Maintained

Murata is the toll-gate on AI infrastructure. Every GPU needs 3,000 MLCCs. Every AI rack needs 440,000. No substitute. 40% global share. 10-year tech lead vs China. Orders at 2x capacity. Price increases being absorbed.

The stock is NOT cheap at 22.6x trailing, but 18.2x forward with 30% CAGR for the key segment is reasonable. The risk is China (48% revenue) — but that's a known risk being actively managed.

Sources

• TrendForce: Murata AI MLCC demand, price hikes, Samsung competition
• Bloomberg: Murata 35% price increase
• Digitimes: AI server MLCC orders 2x capacity
• Signal Integrity Journal: Capacitors in AI power delivery
• Murata corporate: earnings, factory completions, MF Material JV
• StockAnalysis: financial data

REPORT: physical_ai_robotics_dd_2026-04-16.md

Physical AI & Robotics — Deep DD (2026-04-16)

Time: 22:18

Theme: Physical AI & Robotics

Catalyst: SoftBank/NEC/Sony/Honda founded "日本AI基盤モデル開発" (Apr 13, 2026) — 1 trillion-parameter Physical AI model for robots and factory equipment

Market Size: Japan warehouse automation USD 1.47B (2025) → USD 5.04B (2034), 14.7% CAGR. Japan FA market USD 18.4B (2026) → USD 28.6B (2031), 9.2% CAGR

Government Spending: ¥1T over 5 years for domestic AI foundation model development (NEDO). METI targeting 30% global Physical AI market share by 2040

Macro Context

Japan is ground zero for Physical AI adoption, driven by converging structural forces:

1. Labor Crisis — 29% of population over 65. 570K care worker shortage by 2040. Unemployment at 2.5%. Government reframing robotics from "industrial automation" to "Physical AI" in 2026.

2. NVIDIA Physical AI Ecosystem — At GTC 2026 (Mar 16), NVIDIA named ABB Robotics, FANUC, KUKA, and Yaskawa as the "Big Four" Physical AI partners (combined 2M+ installed robots). Released GR00T N1.7 (early access), Cosmos 3, Isaac Lab 3.0. Jensen Huang: "every industrial company will become a robotics company."

3. 日本AI基盤モデル開発 (Japan AI Foundation Model Development) — Founded Apr 13, 2026 by SoftBank, NEC, Honda, Sony + 3 mega-banks + Nippon Steel + Kobe Steel. Goal: 1T-parameter Physical AI model trained on Japanese industrial data. SoftBank/NEC lead AI dev, Honda deploys in autonomous vehicles, Sony covers robotics/gaming hardware.

4. SoftBank ABB Robotics Acquisition — $5.375B deal (Oct 2025, closing mid-late 2026). 7,000 employees, $2.3B revenue (2024), 12.1% EBITA margin. Integrating with SoftBank Robotics, Berkshire Grey, AutoStore, Agile Robots, Skild AI.

5. Humanoid Robot Production — 50K-100K global shipments expected 2026 (Tesla 50-100K, Figure 12K, Chinese OEMs 10K+ each). Japan's component-excellence strategy: Harmonic Drive + Nabtesco supply ALL humanoid OEMs globally. Actuators = ~50% of humanoid production cost.

6. Cobot Market Explosion — Japan cobot CAGR 45.6% (2022-2028). Global cobot market USD 3.06B (2025) → USD 3.74B (2026). APAC leads with 49% share.

Sources:

• 日本AI基盤モデル開発: Nikkei, TechWire Asia
• NVIDIA GTC 2026: NVIDIA Newsroom
• SoftBank ABB: SoftBank Press, Interact Analysis
• Humanoid Market: IDTechEx
• Cobot Market: Markets & Markets
• Japan FA Market: Mordor Intelligence

Portfolio Summary

NEW Stock Cards

4. Daifuku (6383.T) — 12% | HIGH [NEW]

WHY: World's #1 material handling / warehouse automation company. Record FY2025: ¥660B revenue (+8%), 15.3% OP margin (above 10-year avg of 10%). FY2026 guidance: ¥700B rev (+5.9%), ¥105B OP (+4.2%). ROE 18.4% (best in theme). Semiconductor cleanroom AMHS is pure-play on global fab buildout (TSMC, Samsung). PE 31.7x reasonable for 18% ROE. +85% 52W. Japan logistics automation megatrend: warehouse robotics market growing 17% CAGR to $1.26B by FY2033.

RISK: FY2026 guidance modest (+5.9% rev). Backlog conversion timing. PB 5.23x rich. Semiconductor capex cycle dependency. China fab restrictions could limit cleanroom business.

TRIGGER: Q1 FY2026 order intake (Jul). Semiconductor capex cycle confirmation. New AGV factory output ramping. OP margin <13% sustained → TRIM.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 31.71 | Fwd PE: 27.92 | PB: 5.23
• ROE: 18.38% | OP Margin: 15.26%
• D/E: 14% | Div Yield: 1.25%
• FCF: +¥53.9B | Cap: ¥2.36T

Supply Chain:

• TSMC / Samsung / Intel → Cleanroom AMHS (confirmed: global fab installations)
• Amazon Japan / Rakuten → E-commerce fulfillment systems (probable: largest logistics automation provider in Japan)
• Toyota / auto OEMs → Factory material handling (confirmed)
• Airport operators (Haneda, Narita) → Baggage handling systems (confirmed)

Key Dates:

• 2026-05-09: FY2025 earnings + FY2026 guidance confirmation
• 2026-07: Q1 FY2026 order intake — semiconductor cleanroom backlog conversion

Sources:

• StockAnalysis
• Smartkarma Earnings
• Daifuku Cleanroom

5. Denso (6902.T) — 10% | MEDIUM [NEW]

WHY: World's largest user of small assembly robots. COBOTTA PRO 900 (Hannover Messe 2026) — fastest cobot in industry (0.28s cycle time) with edge computing AI. PE 13.9x / Fwd PE 9.6x is cheapest in theme. FCF +¥201B (massive). PB 0.94x — trading below book. Toyota ecosystem anchor (35% owned by Toyota). Denso Wave subsidiary is Japan's cobot/SCARA leader. $7.37T revenue scale dwarfs pure robotics peers. COBOTTA deployments +12% YoY in semiconductor/electronics. Physical AI play: edge AI integration in cobots for autonomous decision-making.

RISK: Toyota dependency (largest customer). Automotive cyclicality dominates (robotics <5% of revenue). EV transition headwinds for core auto components. Operating margin 6.7% thin. ROE 7.9% mediocre for the scale. Robotics is an option, not the core business.

TRIGGER: Robotics segment breakout in earnings (currently buried in "Others"). COBOTTA PRO 900 commercial orders. Toyota Physical AI partnership announcements. OP margin <5% → EXIT.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 13.93 | Fwd PE: 9.62 | PB: 0.94
• ROE: 7.89% | OP Margin: 6.69%
• D/E: 18% | Div Yield: 3.34%
• FCF: +¥201.2B | Cap: ¥5.16T

Supply Chain:

• Toyota → Primary customer and 35% owner (confirmed)
• Semiconductor fabs → COBOTTA cobots for wafer handling (confirmed: Denso Robotics)
• Electronics assembly (Apple suppliers) → SCARA robots (probable)
• NVIDIA → Potential Physical AI integration via edge computing cobots (inferred)

Key Dates:

• 2026-04-20 to 04-24: Hannover Messe — COBOTTA PRO 900 launch event
• 2026-05-09: FY2026 earnings
• 2026-06-30: Semiannual rebalance

Sources:

• StockAnalysis
• Denso Hannover Messe 2026
• COBOTTA PRO

9. THK (6481.T) — 5% | MEDIUM [NEW]

WHY: Invented the linear motion guide (1972). LM Guides + ball screws are in virtually every industrial robot and CNC machine globally. Critical humanoid component: actuators and linear motion systems = ~50% of humanoid production cost. Forward PE 26x (vs trailing 62.6x) signals earnings inflection. FCF +¥23.9B. Dividend yield 3.2% (highest in component peers). Morgan Stanley "Humanoid 100" value chain inclusion. Pure picks-and-shovels play on ALL robot types.

RISK: Trailing PE 62.6x (net loss reported). ROE 3.2% weak. Revenue -4.8% YoY declining. China/semiconductor cycle dependent. No confirmed humanoid OEM contracts (generic component supplier). Altman Z-Score 2.82 (borderline).

TRIGGER: Revenue growth turning positive (next 2 quarters). Named humanoid OEM supply agreements. China FA order recovery. NI turning positive consistently → upgrade conviction.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 62.60 | Fwd PE: 25.99 | PB: 2.37
• ROE: 3.22% | OP Margin: 6.75%
• D/E: 49% | Div Yield: 3.23%
• FCF: +¥23.9B | Cap: ¥630B

Supply Chain:

• Fanuc / Yaskawa / ABB / KUKA → LM Guides for robot axes (confirmed)
• TSMC / Samsung → Semiconductor equipment linear motion (confirmed)
• Tesla Optimus / Figure / Unitree → Humanoid actuator components (inferred: Morgan Stanley Humanoid 100 list)
• Machine tool OEMs (DMG Mori, Okuma) → Ball screws (confirmed)

Key Dates:

• 2026-05-12: FY2025 earnings — critical for NI recovery confirmation
• 2026-06-30: Semiannual rebalance

Sources:

• StockAnalysis
• Morgan Stanley Humanoid 100
• Humanoid Production Economics

10. Hamamatsu Photonics (6965.T) — 5% | MEDIUM [NEW]

WHY: Japan's leading optical sensor manufacturer. Photomultiplier tubes, image sensors, photodiodes used in robot vision, semiconductor inspection, medical imaging, and LiDAR. FY2026 guidance: ¥222B rev (+4.7%), ¥17.2B OP (+6.4%). Active ¥13B share buyback underway. Semiconductor inspection demand growing as node shrinks require better detection. 3D Machine Vision market USD 8.1B (2024) → USD 22.4B (2032), 14.8% CAGR. Monopoly-like positions in niche photonics. R&D-heavy culture (Hamamatsu Photonics R&D campus is legendary).

RISK: PE 45.7x / Fwd PE 38.5x expensive. ROE 4.1% low. Revenue growth modest (+4.7%). Niche products — limited addressable market expansion. Semicap cycle downturn risk. No direct Physical AI partnership announced.

TRIGGER: Semiconductor inspection order acceleration. Named robot vision OEM partnerships. ROE recovery >6% → upgrade. PE >50x sustained → TRIM.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 45.68 | Fwd PE: 38.54 | PB: 1.78
• ROE: 4.12% | OP Margin: 6.70%
• D/E: 30% | Div Yield: 2.00%
• FCF: N/A | Cap: ¥571B

Supply Chain:

• ASML / Applied Materials / Tokyo Electron → Semiconductor inspection sensors (confirmed)
• Medical device OEMs → Imaging detectors (confirmed)
• Autonomous driving / LiDAR OEMs → Photodetectors (probable)
• Robot vision integrators (Keyence, Cognex) → Image sensors (probable: component supplier)

Key Dates:

• 2026-05-07: Next earnings report
• 2026-06-30: Semiannual rebalance

Sources:

• StockAnalysis
• Hamamatsu IR
• 3D Machine Vision Market

13. Cyberdyne (7779.T) — 2% | LOW [NEW]

WHY: Only listed Japanese humanoid/wearable robot pure-play. HAL (Hybrid Assistive Limb) exoskeleton for medical rehab and industrial labor assist — directly addresses Japan's aging workforce crisis. Cleaning robots + transport robots deployed in real facilities. Revenue ¥4.1B (small). Cash ¥9.2B > market cap of ¥62.9B's enterprise value. Near-zero debt (D/E 0.01). Potential acquisition target for SoftBank/Denso given Physical AI push. 38.65% insider ownership (founder controls company).

RISK: UNPROFITABLE: negative operating margin -12%, negative FCF. Forward PE 125.5x absurd. Revenue barely growing (+0.7% YoY). Altman Z-Score 2.91 (bankruptcy risk). No path to profitability visible. HAL certification progress slow (only EU/Japan medical). Speculative position only.

TRIGGER: Revenue >¥5B → shows traction. Named SoftBank/大手 partnership → ADD. Operating profit positive → UPGRADE. Continued losses >2 years → EXIT entirely.

Fundamentals (StockAnalysis 2026-04-16):

• PE: N/A | Fwd PE: 125.47 | PB: 1.58
• ROE: -0.13% | OP Margin: -11.95%
• D/E: 1% | Div Yield: 0%
• FCF: -¥222M | Cap: ¥62.9B

Supply Chain:

• Hospitals (Japan, EU) → HAL medical exoskeleton (confirmed)
• Logistics facilities → Cleaning/transport robots (confirmed)
• Construction firms → HAL Lumbar for worker assist (confirmed)

Key Dates:

• 2026-05-14: Earnings announcement
• 2026-06-30: Semiannual rebalance

Sources:

• StockAnalysis
• Cyberdyne Products

UPDATED Stock Cards

1. Fanuc (6954.T) — 15% | HIGH [UPDATED]

WHY: Picks-and-shovels monopolist of global manufacturing: ~60% CNC global share, ~25% industrial robot share. Zero debt fortress (¥697B net cash). NVIDIA Physical AI partnership live (Jetson + Isaac Sim + Omniverse): voice command → Python codegen for robot programming. 1,000+ CRX cobot orders post-iREX with multi-thousand-unit pipeline. Robot segment now 44.8% of revenue (+21.9% YoY). FY2025 guidance revised upward to ¥840.7B. $90M US factory investment as tariff hedge. Named "Big Four" at NVIDIA GTC 2026.

RISK: ~27% China revenue. 10% US tariff active since Apr 5. Section 232 investigation on robots/CNC (report by May 30) — potential 25%+ tariff. 24% reciprocal tariff suspended until ~Jul 9. Forward PE 30.7x.

TRIGGER: Apr 25 earnings: FY2026 guidance with tariff assumptions. China PMI > 52 = bullish. May 30: Section 232 report. Jul 9: tariff suspension expiry.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 35.61 | Fwd PE: 30.74 | PB: 3.18
• ROE: 9.37% | OP Margin: 21.07%
• D/E: 0% | Div Yield: 1.66%
• FCF: N/A | Cap: ¥5.76T

Changes from 2026-04-14:

• PE: 35.69 → 35.61 | Fwd PE: 30.81 → 30.74 | Market Cap: ¥5.77T → ¥5.76T

2. Yaskawa Electric (6506.T) — 12% | HIGH [UPDATED]

WHY: World's largest servo motor manufacturer (~30% global share) — the "muscles" of every robot. Named among NVIDIA's "Big Four" Physical AI partners at GTC 2026 (Jetson modules in controllers, Omniverse/Isaac integration). SoftBank Physical AI collaboration MOU for logistics/offices (MOTOMAN NEXT with VLM+VLA). FY2027 guidance: ¥580B rev (+7%), ¥60B OP (+26.8%) driven by AI/semiconductor demand. $180M Wisconsin campus hedges Section 232 tariff risk. Humanoid ambition now in official filings: "deepen engagement in humanoid robot domain."

RISK: Operating margin 8.7% thin — Robotics segment margin compressed to 8.3%. Section 232 investigation (May 2026 decision). NI -38% YoY optically ugly (one-time FY2025 gain distortion). Trailing PE ~39x high for 8.7% margin. No named humanoid OEM contracts yet.

TRIGGER: May 2026: "Dash 35" mid-term plan with humanoid/AI targets. May 30: Section 232 report. FY2027 Q1: Motion Control order confirmation. Named humanoid OEM contracts.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 38.88 | Fwd PE: 28.57 | PB: 2.77
• ROE: 7.84% | OP Margin: 8.73%
• D/E: 25% | Div Yield: 1.36%
• FCF: +¥5.9B | Cap: ¥1.37T

Changes from 2026-04-15:

• PE: 37.96 → 38.88 | Fwd PE: 29.1 → 28.57 | PB: 2.71 → 2.77 | Cap: ¥1.34T → ¥1.37T

3. Keyence (6861.T) — 12% | HIGH [UPDATED]

WHY: Most profitable industrial company globally — 50.9% operating margin. Q3 standalone revenue +11.4% YoY (Americas +15%, Asia +18.9%). Five major product launches in Q1 2026 including VS-G (AI vision + full-image storage) and GX-1000 (3D printer). ¥1.33T net cash, zero debt. ROIC 28.75%. Dividend raised +57% YoY. Fabless model insulates from direct tariff impact. AI vision systems (IV4 with 99.8% accuracy) are the "eyes" enabling Physical AI in factories.

RISK: Forward PE 32.5x expensive. ~15% China revenue at risk. Chinese vision players (Hikrobot, IVSEMI) gaining share. FY2025 full-year results Apr 24-29.

TRIGGER: Apr 24-29: FY2025 full-year + FY2026 guidance. China revenue breakdown. Cognex quarterly results for competitive read.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 36.40 | Fwd PE: 32.47 | PB: 4.56
• ROE: N/A (ROIC 28.75%) | OP Margin: 50.88%
• D/E: 0% | Div Yield: 0.88%
• FCF: N/A | Cap: ¥15.22T

Changes from 2026-04-15:

• PE: 36.68 → 36.40 | Fwd PE: 32.52 → 32.47 | OP Margin: 49.9% → 50.88% (improved)

6. Harmonic Drive Systems (6324.T) — 8% | HIGH [UPDATED]

WHY: Makes precision strain-wave gears — THE critical joint component for ALL humanoid robots. ¥2.5B humanoid orders in FY3/26, doubling to ¥5B in FY3/27. Every humanoid OEM globally (Figure, Unitree, Tesla Optimus, Kawasaki Kaleido) needs Harmonic Drive gears. Market cap ¥444B (up from ~¥300B). Humanoid production scaling from lab to factory: 50-100K units expected globally in 2026.

RISK: PE 96.8x / Fwd PE 103.3x extreme — prices in significant humanoid ramp. Operating margin only 2.7% (compressed). ROE 5.9% below cost of capital. If humanoid adoption delays, multiple contracts. Dividend yield 0.5% negligible.

TRIGGER: Quarterly humanoid order intake acceleration. Named humanoid OEM supply agreements. OP margin recovery >5%. PE >150x → EXIT.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 96.77 | Fwd PE: 103.34 | PB: 5.73
• ROE: 5.91% | OP Margin: 2.67%
• D/E: 24% | Div Yield: 0.46%
• FCF: N/A | Cap: ¥444B

Changes from 2026-04-14:

• Market cap: ~¥300B → ¥444B (+48%). Fundamentals now populated from StockAnalysis (were N/A).

7. Kawasaki Heavy Industries (7012.T) — 7% | MEDIUM [UPDATED]

WHY: Defense rerate + robotics option. Japan defense budget doubling to ¥10T+ by FY2027. Kaleido humanoid (fire/rescue/heavy labor) is most advanced Japanese humanoid. +162% 52W.

RISK: D/E 1.19 is high. Fwd PE 31.7x. Robot segment only 11% of revenue. Execution risk on government programs.

TRIGGER: Japan defense supplemental budget. Kaleido commercial deployment contracts. Boeing production rate increases.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 25.35 | Fwd PE: 31.65 | PB: 3.18
• ROE: N/A | OP Margin: 5.28%
• D/E: 119% | Div Yield: 0.90%
• FCF: +¥46.9B | Cap: ¥2.78T

Changes from 2026-04-14:

• PE: 25.77 → 25.35 | Fwd PE: 32.17 → 31.65 | Cap: ¥2.83T → ¥2.78T

8. Nabtesco (6268.T) — 5% | MEDIUM [UPDATED]

WHY: RV-series reduction gears in 60%+ of global industrial robot joints. Like Harmonic Drive, a critical component for humanoid robots. Also makes aircraft flight control systems (defense angle). FY2026 guidance: ¥327B rev (+6.2%), 8.5% OP margin. FCF +¥19.6B positive.

RISK: Revenue declined -4.8% YoY in FY2025. PE 41.2x elevated for 5.5% ROE. China ~30% of robot orders. No confirmed humanoid contracts yet.

TRIGGER: Quarterly robot gear orders. Humanoid OEM supply announcements. Revenue growth turning positive.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 41.23 | Fwd PE: 29.04 | PB: 1.93
• ROE: 5.46% | OP Margin: 6.78%
• D/E: 21% | Div Yield: 1.75%
• FCF: +¥19.6B | Cap: ¥559B

Changes from 2026-04-14:

• All fundamentals now populated from StockAnalysis (were N/A). Cap: ~¥500B → ¥559B.

11. Omron (6645.T) — 5% | MEDIUM [UPDATED]

WHY: Quality Japanese FA franchise trading at PB 1.01 — nearly book value. IAB segment (45% rev) has best-in-class sensors, PLCs, safety, cobots, AMRs. China FA recovery from -22.6% trough would drive margin recovery from 6% toward historical 12-14%. Healthcare (blood pressure) levered to aging demographics. Forward PE 22.6x (cheapest mid-cap in theme).

RISK: Payout ratio 87% on thin FCF. ROE 2.3% awful. China demand recovery timing uncertain. Trailing PE 41.8x on depressed earnings.

TRIGGER: IAB quarterly order intake turning positive YoY (May 2026 earnings). China PMI > 51. OP margin >8% → upgrade.

Fundamentals (StockAnalysis 2026-04-16):

• PE: 41.84 | Fwd PE: 22.57 | PB: 1.01
• ROE: 2.28% | OP Margin: 6.08%
• D/E: 24% | Div Yield: 2.17%
• FCF: +¥5.9B | Cap: ¥979B

Changes from 2026-04-14:

• PE: 39.25 → 41.84 | Fwd PE: 21.18 → 22.57 | PB: 0.94 → 1.01 | Cap: ¥919B → ¥979B

12. Nidec (6594.T) — 2% | LOW [UNCHANGED — AVOID]

WHY: E-Axle leader + industrial motors. Forward PE 13.7x with PEG 0.9 prices no growth.

RISK: CRITICAL: Securities fraud investigation active (Mar 2026). Dividend cancelled. ROIC 4% below WACC 7.7%. Highest risk name. Do NOT increase until resolved.

TRIGGER: Corrective action plan. Restatement risk. Oasis governance proxy.

Watchlist (Not in Portfolio)

SoftBank Group (9984.T) — WATCHLIST

WHY: Orchestrator of Japan's Physical AI ecosystem. ABB Robotics acquisition ($5.375B, closing mid-late 2026). Founded 日本AI基盤モデル開発 (Apr 13). Portfolio: SoftBank Robotics, Berkshire Grey, AutoStore, Agile Robots, Skild AI ($1.4B round). Masayoshi Son: "Physical AI is our next frontier."

RISK: Not a pure robotics play — conglomerate holding company. $3.6B junk bond issuance (Apr 2026). Massive debt load. OpenAI $40B loan. Valuation difficult to assess. Execution risk on ABB integration.

TRIGGER: ABB Robotics close (mid-late 2026). 日本AI基盤モデル開発 first model release. Humanoid robot product announcements from portfolio companies.

MUJIN — WATCHLIST (Private)

WHY: Japan's most promising warehouse robotics AI company. MujinOS platform enables autonomous picking/palletizing. Series D: $233M raised (Dec 2025, NTT lead). Deploying at scale in Japanese logistics. IPO candidate.

TRIGGER: IPO filing or TSE listing announcement.

Theme-Level Risk Assessment

Errata / Cross-Validation Notes

*Report generated 2026-04-16. Not investment advice. All data from StockAnalysis.com, company IR filings, and cited web sources.*

REPORT: q_glass_competitive_analysis_2026-04-21.md

Q Glass Competitive Analysis — Asahi Kasei vs Taiwan vs China

Time: 08:30

Date: 2026-04-21

Type: Explore — Q glass (quartz cloth) competitive landscape for M9 CCL

Key Finding: Asahi Kasei is NOT the Only Q Glass Player

Our initial thesis that "Asahi Kasei Q Glass wins M9" was incomplete. The Q glass market has multiple players across Japan, Taiwan, and China. Asahi Kasei is a NEW ENTRANT (mass production starting 2026), not an incumbent.

Q Glass Supply Chain Map

`

QUARTZ YARN (upstream)

├── Glotech (5475.TWO) — Taiwan, 99.995% SiO2 ← ONLY Taiwan quartz yarn maker

├── Feilihua/中益新材 (300395.SZ) — China, full vertical chain

├── Shin-Etsu (4063.T) — Japan, SQX series (small capacity)

└── Unknown suppliers to Asahi Kasei

QUARTZ CLOTH WEAVING (midstream)

├── Asahi Kasei (3407.T) — Japan, Q Glass brand, entering 2026

├── Glotech (5475.TWO) — Taiwan, vertically integrated

├── Feilihua (300395.SZ) — China, 50,000 m/month

├── Honghe/宏和 (603256.SH) — China, ramping

└── Shin-Etsu (4063.T) — Japan, niche only

M9 CCL LAMINATION (downstream)

├── EMC/台光電 (2383.TW) — FIRST Nvidia M9 cert, ~70% target share

├── Lianmao/聯茂 (6213.TW) — SECOND M9 cert

├── TUC/台燿 (6274.TW) — alternative Z-cloth approach (avoids Q glass)

├── Panasonic (6752.T) — Japan

├── Shengyi (600183.SH) — China

└── Doosan — Korea

`

Competitive Comparison

Critical Insight: Glotech May Be Asahi Kasei's Upstream Supplier

Glotech's quartz yarn (99.995% SiO2) is HIGHER purity than Asahi Kasei's Q Glass cloth (99.9%). This suggests Glotech may supply the raw quartz yarn that Asahi Kasei weaves into finished Q cloth — though unconfirmed.

The Q glass supply chain is fundamentally different from T-glass:

• T-glass: Nittobo monopolizes both yarn AND weaving (hence 90% share)
• Q glass: Quartz yarn is more accessible (purifying SiO2 to 99.9%+ is technically easier than optimizing complex glass compositions)
• This means Q glass could break the monopoly model — multiple suppliers can coexist

Revised Asahi Kasei Thesis

Previous: "Asahi Kasei Q Glass monopoly wins M9"

Updated: "Asahi Kasei is ONE OF SEVERAL Q glass players, but with Japanese quality premium"

Bull case (still valid):

• Q Glass does meet M9 spec (Df 0.0005 vs required 0.0007)
• Japanese quality reputation matters for Nvidia/Apple supply chain
• PE 13.4x with 3.5% yield — even if Q glass is just one of several, it's priced for zero premium
• Target: 3x glass cloth sales by 2030

Bear case (new):

• NOT a monopoly like Nittobo's T-glass — multiple Q glass competitors
• Feilihua (China) already has Nvidia certification and volume production
• Glotech (Taiwan) may be upstream supplier — Asahi Kasei could be a reseller/weaver, not a monopolist
• Conglomerate dilution remains

Conviction: MEDIUM (downgraded from HIGH)

Still worth owning at PE 13.4x but it's NOT a Nittobo-like monopoly. It's a competitive market with Japanese, Taiwanese, and Chinese players.

M9 CCL Maker Rankings (Who Actually Makes the Final Product)

Note: The real M9 value capture may be at the CCL maker level (EMC, Lianmao), not at the Q glass cloth level. EMC targets ~70% M9 share and has first-mover advantage with Nvidia.

Stocks to Watch (Non-Japan)

Sources

• Digitimes: Glotech quartz cloth demand, Asahi Kasei market entry
• Nikkei Asia: Japan glassmakers retooling for AI chip demand
• globaltechresearch: NEZ vs Q Glass analysis
• Sina Finance: Q glass supplier overview (Chinese)
• TrendForce: Nittobo capacity expansion, TSMC CoPoS

REPORT: restocking_cycle_dd_2026-04-20.md

Restocking Cycle DD — Glass Fiber / Power Semi / Passive Components

Time: 06:00

Date: 2026-04-20

Type: New narrative — destocking to restocking to price increase cycle

Sectors: Glass fiber cloth, Power semiconductors, Passive components

Executive Summary

Three key electronics material sectors are entering a destocking to restocking to price increase cycle. The thesis: hold tight to products going through destocking into shortage and price increases.

Glass fiber cloth is the most acute — Nitto Boseki holds 90% monopoly in T-glass for AI chip packaging. CCL lead times hit 6 months. Prices up 15-30%.

Passive components (MLCC) — Taiyo Yuden initiated May 2026 price increases. Murata expected to follow. AI servers use 30,000 MLCCs each (6x normal).

Power semiconductors — Lead times 20-45 weeks. SiC critically tight. Fuji Electric investing JPY 200B in capacity.

Top Picks

Sector 1: Glass Fiber Cloth

Supply Crisis

• CCL lead times: 6 months with quota systems
• CCL prices up 15-30% (Digitimes Apr 2026)
• Nvidia, AMD, Apple, Google executives visiting Nittobo HQ
• New capacity not until Q4 FY2026

3110.T Nitto Boseki — HIGH

90% T-glass monopoly. Zero substitutes. 3x capacity expansion underway. PE 29x / Fwd 22x.

3407.T Asahi Kasei — MEDIUM

#2 glass cloth. PE 13.4x. Earnings +208% YoY. Conglomerate discount.

5208.T Arisawa Manufacturing — MEDIUM

Small-cap pure-play. PE 14.85x. 5.93% dividend yield.

Sector 2: Power Semiconductors

Structural Tightness

• Lead times: 20-45 weeks
• SiC critically tight
• Japanese firms investing aggressively

6504.T Fuji Electric — HIGH

JPY 200B investment. SiC + IGBT. Operating profit JPY 128.5B guided.

6963.T Rohm — MEDIUM

#1 Japan SiC. Earnings inflection from JPY -57B loss. High risk/reward.

Sector 3: Passive Components

Price Cycle Turning

• Taiyo Yuden price increase May 1, 2026
• Murata expected to follow (Goldman Sachs)
• AI servers use 30,000 MLCCs each
• Murata book-to-bill: 1.12x

6981.T Murata — HIGH

40% global MLCC. New capacity online. Operating leverage on price increases.

6762.T TDK — MEDIUM-HIGH

Forward PE 12.3x. Raised guidance to JPY 180B net profit.

6976.T Taiyo Yuden — MEDIUM-HIGH

Initiated the price increase cycle. Mizuho upgrade. Morningstar +25% fair value.

Sources

• Nittobo T-glass: tomshardware.com, globaltechresearch.substack.com
• CCL prices: digitimes.com (Apr 2026)
• MLCC prices: digitimes.com (Apr 2026)
• Murata capacity: evertiq.com
• Component shortage: j2sourcing.com (Mar 2026)
• Rohm recovery: spglobal.com
• All financials: stockanalysis.com

REPORT: restocking_evolve_m9_2026-04-21.md

Restocking Theme Evolution — M9 CCL Grade Analysis

Time: 06:30

Date: 2026-04-21

Type: Evolve — M9 material specification deep dive

Stocks updated: 3110.T (Nitto Boseki), 3407.T (Asahi Kasei)

Key Finding: M9 is a CCL Material Grade, Not a TSMC Node

M9 is the next-generation CCL (Copper-Clad Laminate) material specification requiring ultra-low dielectric loss tangent (Df <= 0.0007). Nvidia has confirmed M9 adoption for Rubin-era products including 1.6T switches, NVL288 orthogonal backplanes, mass production H2 2026.

Critical: Nitto Boseki NEZ Glass CANNOT Meet M9

Nitto Boseki's NEZ glass achieves Df 0.001 — it falls short of the M9 requirement of Df <= 0.0007. Asahi Kasei's Q Glass (99.9% silica quartz fabric) achieves Df 0.0005, comfortably meeting M9.

Industry consensus emerging: "NEZ from Nittobo has lost, M9 will adopt Q glass."

Impact by Segment — Nitto Boseki Has THREE Glass Businesses

The T-glass monopoly for AI chip packaging (Nvidia GB200, Apple, AMD) remains intact. This is where the 15-30% CCL price increases are happening. The M9 issue affects the NE/NER PCB motherboard cloth segment.

Thesis Updates

3110.T Nitto Boseki — Conviction: HIGH (maintained, risk noted)

T-glass monopoly remains the core thesis. 90% share, 3x capacity expansion, pricing power. BUT:

• NE/NER segment faces M9 disruption from Asahi Kasei Q Glass
• Long-term structural risk from glass core substrates (Intel, Samsung, TSMC CoPoS pilot June 2026)
• These are separate revenue streams — T-glass is the bull case, NE/NER is the bear case

3407.T Asahi Kasei — UPGRADED to HIGH

Q Glass wins M9 specification for Nvidia Rubin-era products. PE 13.4x with 3.5% yield is dirt cheap for a company that just won the next-gen AI networking material spec. The conglomerate discount now works in your favor — you're getting M9 exposure at a massive discount to what a pure-play would trade at.

Supply Chain Map

`

Raw Materials

Nitto Boseki (T-glass yarn + cloth) -----> ABF/BT Substrates -----> AI Chips (Nvidia, AMD)

Nitto Boseki (NE/NER glass cloth) ---X---> PCB Motherboards (LOSING to Q Glass for M9)

Asahi Kasei (Q Glass quartz fabric) -----> M9 CCL -----> 1.6T Switches, NVL288 (Nvidia Rubin)

Long-term disruption:

Glass Core Substrates (Intel/Samsung/TSMC) -----> Could bypass glass cloth entirely (2028-2030)

`

Risk Watch: Glass Core Substrates

Glass core substrates use glass PANELS not glass CLOTH. If widely adopted, they would structurally eliminate demand for glass cloth in substrates. Timeline: 2028-2030 for meaningful volume.

Sources

• NEZ vs Q Glass analysis: globaltechresearch.substack.com
• Nittobo next-gen T-glass 2028: TrendForce
• TSMC CoPoS pilot: TrendForce Apr 2026
• M9 Nvidia confirmation: Longbridge, Digitimes
• Glass core substrates: NEG, 36Kr, Intel