Japanese GAA Supply Chain — 7-Ticker Swarm Comparison
The seven names span four distinct supply-chain layers feeding the 2nm/A14 GAA ramp at TSMC, Samsung and Intel. The synthesis below merges the per-ticker deep-dives with a GAA supply-chain layer overlay, an incentive alignment screen, and an STF Research coverage map — the published-thesis dimension matters because four of seven names already have dedicated STF posts, which compresses the available information edge.
Top-line conclusion
The basket as a whole has already moved. Every name is up 100%+ over twelve months; four are up more than 300%. Cheap forward multiples are not on the table. The decision is not "Japan-semi yes or no" but which positions still have asymmetry, and where management quality + STF crowding actually favours the buyer.
Three names earn full BUY conviction; three earn WATCH; one earns conditional starter-size BUY pending Japanese-language disclosure work.
Ranked by current asymmetry:
- 6525 Kokusai Electric — the only major name without STF coverage; deposition POR for GAA bulk-dielectric; half-size starter at ¥6,300-6,400, scale on China headlines.
- 4966 C. Uyemura — ENEPIG monopoly; 33.9% family ownership; STF published 4 days ago so edge is closing — scale-in 35/35/30 over 6-8 weeks.
- 6327 Kitagawa Seiki — only press supplier qualified for M9/Q-Glass on Vera Rubin; FY26E P/E ~11x — but +329% rally compresses asymmetry and disclosure is thin. Starter only, 1.5-2.5% sizing.
- 7826 Furuya Metal — Ru-at-2nm liner decided, bulk-Ru contested; +500% from low; ~7% probability-weighted return at spot. WATCH, scale only on pullback to ¥7,500-8,000 or A14 material-of-record confirmation.
- 5344 MARUWA — AlN submount monopoly for InP lasers in 800G/1.6T optics; NOT actually a GAA play, it's the optical-interconnect plumbing one layer downstream. 52x trailing P/E, consensus PT at spot. WATCH, entry zone ¥55-65k.
- 4971 MEC — CZ-series copper surface treatment; STF thesis from 3/1 has fully played out, stock +100% in 75 days. PASS at ¥11,050, re-engage ¥7,000-9,000. Verdict reaffirmed from 5/12.
- 4062 Ibiden — ABF substrate leader at 74x P/E; glass-core ramp may be 12-24 months earlier than prior assumed with AMD/Absolics, not Intel/Ibiden, as launch customer. STF routes around Ibiden. WATCH, not BUY.
Trading comps (live, 2026-05-15)
| Ticker | Price (¥) | MktCap (¥B) | Fwd P/E | TTM P/E | EV/EBITDA | P/B | Div % | 12mo % |
|---|---|---|---|---|---|---|---|---|
| 4062 Ibiden | 15,845 | 4,425 | 51.3 | 73.6 | 37.3 | 8.0 | 0.24 | +569.7 |
| 4966 C. Uyemura | 24,660 | 396 | 31.4 | 28.5 | 16.3 | 3.5 | 2.26 | +177.9 |
| 4971 MEC | 10,780 | 197 | 55.6 | 39.6 | 75.7 | 14.2 | 0.95 | +375.7 |
| 5344 MARUWA | 75,500 | 932 | 39.2 | 51.2 | 30.0 | 6.3 | 0.14 | +135.7 |
| 6327 Kitagawa Seiki | 2,580 | 21 | 34.9 | 42.5 | 20.2 | 3.7 | 0.52 | +330.3 |
| 6525 Kokusai Electric | 6,387 | 1,492 | 33.4 | 49.7 | 25.0 | 7.0 | 0.78 | +110.0 |
| 7826 Furuya Metal | 9,440 | 232 | 23.6 | 30.7 | 159.0 | 3.5 | 2.92 | +306.5 |
Three things to read off this table. First, every name is up at least 110% in twelve months; four exceed 300%. Second, Kokusai and Uyemura are the cheapest on forward P/E and the only two under 35x. Third, the EV/EBITDA at MEC and Furuya is distorted: MEC by Q1 seasonality and an installed base that hasn't yet ramped, Furuya by precious-metal working-capital build flipping the balance sheet to net debt for the first time in years — read gross profit, not revenue, for Furuya.
Supply-chain layer overlay
The basket maps to four distinct layers, and each layer has different pricing-power dynamics into the GAA / advanced-AI-substrate ramp.
Layer 1 — Deposition equipment (front-end WFE). 6525 Kokusai Electric is the anchor here. Deposition is ~30-35% of WFE at GAA (up from ~25% at FinFET), and batch is ~25-30% of deposition — so Kokusai-addressable spend is ~$1.6-2B per leading-edge fab build. POR switching cost is 12-24 months per node. Material moat. The competitive question is ASMI for the high-value work-function-metal POR — both grow, neither displaces the other near term. Pricing power: high but tier-priced; service tail at high margin. Service revenue ~¥40-50B/year and growing with install base of ~2,000 tools.
Layer 2 — Precious-metal sputtering targets. 7826 Furuya Metal. Ru-as-liner at 2nm is essentially decided across TSMC/Intel/Samsung per Applied Materials' Endura Volta Ruthenium CVD tool. Bulk-Ru at A14/1.4nm is the contested step, 2027-2028 decision window. Pricing power: structurally high (Ru supply curve is "nearly vertical" per STF — no new mines because Ru is a Pt/Ni byproduct) — but obscured because Furuya cannot hedge Ru and FY25 OI was -3% despite revenue +20.7%. Gross profit, not revenue, is the volume proxy. HDD platter Ru content is a larger near-term volume driver than initially weighted — HAMR adds platters, each platter has more Ru.
Layer 3 — Surface-treatment chemistry. 4966 Uyemura (ENEPIG ~70% Taiwan / 50-55% Japan share) and 4971 MEC (CZ-series copper surface treatment). Both physics-locked into the FCBGA substrate flow — every substrate closes with ENEPIG; every multi-layer build needs copper-resin adhesion treatment. Pricing power: very high on the chemistry side because qualification cycles are long and replacement requires reliability re-testing across the substrate value chain. Uyemura: 440bp gross-margin step in FY3/25, Q3 OP margin 24.7% vs. 13.9% two years prior. MEC: Q1 print just landed, dividend forecast revised up.
Layer 4 — Substrates and packages. 4062 Ibiden (ABF substrates for advanced packaging) and 5344 MARUWA (AlN ceramic submounts for InP lasers — note: this is NOT GAA-node packaging, it's the optical-interconnect plumbing inside 800G/1.6T transceivers and CPO external-laser-source modules, which sits one layer downstream of GAA-enabled compute). Different physics, different customer set, different cycle. Pricing power: Ibiden faces glass-core disruption risk (Absolics/AMD MI400 may ramp 12-24 months earlier than prior thesis assumed) — that compresses Ibiden's long-cycle moat. MARUWA's pricing power is physics-locked (CTE 4.6×10⁻⁶/K matches InP at 4.5; no copper alternative) but valuation has caught up at 52x trailing.
Layer 5 — Precision press equipment (small-cap niche). 6327 Kitagawa Seiki. World #1 in CCL/PCB multi-daylight vacuum-press equipment. Per STF Research, only press supplier qualified for M9/Q-Glass tolerances on Vera Rubin-class boards. Pricing power: bottleneck-tier if true — but the "sole qualified" claim is unverified in any primary source beyond STF. Q1 FY2026 order intake at 10-year high (¥5.4B); PCB share of backlog stepped 68% → 85% YoY.
Incentive alignment screen
| Ticker | Guidance Tendency | Follow-Through | Incentive Alignment | Hurdle-to-Model Gap | Overall Grade | Key Flag |
|---|---|---|---|---|---|---|
| 4062 Ibiden | Conservative-to-straight | High (no broken promises FY3/25-FY3/26) | Modest cash-heavy comp; Oono capex executing on stated timeline | N/A — Japan PSU disclosure thin | A− / Green-Yellow | Yellow is structural Japan disclosure granularity, not Ibiden-specific |
| 4966 C. Uyemura | Conservative-sandbagger | High (zero equity raises despite ¥84B base) | 33.9% family ownership held outright, not granted — strong | N/A | B+ / Green | Trapped ¥52B cash; family unlikely to special-div or buy back at scale |
| 4971 MEC | Conservative | High (4 consecutive in-line/beats; div forecast revised up at Q1) | CEO Maeda owns 3.98% (~$49M) — meaningful for a non-founder JP CEO | N/A | B+ / Green | Analyst coverage thinned 3→2 at 52w high — small negative signal |
| 5344 MARUWA | Cannot fully assess (Japan disclosure) | Capex efficiency 0.8× incremental revenue per ¥ capex — solid | Yellow (cannot verify pay-for-perf); Kanbe internal-promote | N/A | B+ / Yellow-Green | Opacity is disclosure-regime artifact, not pathology |
| 6327 Kitagawa Seiki | Cannot assess without Japanese filings translation | Provisional B; dividend track + zero insider selling at 5x'd rally | 27% insider stake — strong via ownership, opaque on incentives | N/A | B- pending yuho review | Disclosure void is the single biggest structural risk; FY24 equity raise use-of-proceeds unverified |
| 6525 Kokusai Electric | Conservative-sandbagger | High (4 consecutive earnings beats) | Career-Hitachi mgmt, CEO Kanai ~$5M holding — meaningful but not founder-level | Disclosure gap (Japan PSU norm) | Yellow-Green | KKR still holds ~10-15% residual and is a steady seller — overhang |
| 7826 Furuya Metal | Conservative-to-straight | ~80% (Sano commissioned on time; capex as guided; FY24 margin call missed) | Family net-worth-concentrated in 7826; no selling at 5x'd rally — cleanest signal | N/A | B+ / Green-Yellow | FY24 buyback opportunity at ¥2-3k was missed — not active capital allocators |
No name in the basket triggers the "flag" criteria (follow-through <50%, misaligned incentives, aggressive/erratic guidance, low-bar hurdles). The Yellow grades across the table are structural to Japanese small-cap disclosure norms — PSU hurdle-vs-model reconciliation isn't a runnable exercise on this disclosure regime — not pathology in any individual name.
The cleanest alignment signals in the basket are Uyemura (33.9% family, held outright, not granted), Furuya (family net-worth-concentrated, no selling at 5x'd low), and MEC (CEO owns 4% of outstanding — high for a non-founder Japanese CEO). Kokusai has the weakest alignment signal of the seven — career-PE-installed management with modest equity — but offset by 4 consecutive earnings beats which is the better behavioural proxy.
STF Research coverage map
This dimension matters because STF Research is the marginal information force driving this entire basket — four of seven names have dedicated STF posts in the last six weeks, which compresses the available information edge.
| Ticker | STF Coverage | Date | Status |
|---|---|---|---|
| 4062 Ibiden | Indirect (treated as consensus centre; STF routes picks AROUND Ibiden) | Multiple | Edge: low; Ibiden is the assumed default |
| 4966 C. Uyemura | "Uyemura: Another Invisible Beneficiary" | 2026-05-11 (4 days ago) | Edge: closing fast |
| 4971 MEC | "Finding the Last Cheap AI Stock in the Japanese Rally" | 2026-03-01 | Edge: gone (stock has doubled since) |
| 5344 MARUWA | "Maruwa: Quiet Winner Behind the Optical Boom" | 2026-04-05 | Edge: closing; stock has 3x'd |
| 6327 Kitagawa Seiki | Flagship Kitagawa post | 2026-03-16 | Edge: closing; STF was NOT trimming Kitagawa in their April "Updates" post |
| 6525 Kokusai Electric | None found | — | Edge: open — only major name in the basket uncovered by STF |
| 7826 Furuya Metal | "Furuya Metal: Selling the World's Scarcest Metals" | 2026-05-04 | Edge: closing; published 11 days ago |
This is the single most important overlay in the synthesis. Kokusai is the only name where you are not trading against an actively-distributed published thesis — and it's also the largest GAA-content-per-wafer name in the basket. The other six all have STF subscribers' price action overlaid on their fundamental setup, which is why most are 200%+ higher than they were when STF began publishing them.
The hierarchy of "uncrowded thesis" matches surprisingly well to the asymmetry ranking: Kokusai (no coverage), Kitagawa (coverage but STF holding through the rally, mixed signal), Uyemura/Furuya (coverage published this month, edge closing), Maruwa (coverage 6 weeks old, edge mostly priced), MEC (coverage 75 days old, edge fully played out), Ibiden (consensus default, no STF edge ever).
Final action ranking
A position-by-position view that combines the supply-chain layer, the multiple, the incentive grade, and the STF coverage status.
6525 Kokusai Electric — half-size starter at ¥6,300-6,400 with scale-in plan. Forward P/E 33x is the cheapest in the basket. POR moat at GAA is real and 12-24 month switching cost protects it. Service tail is ~¥40-50B/year at high margin and grows mechanically with install base. No STF coverage = no published-thesis crowding. Two real risks: China at 45-50% of revenue (binary on METI rules), and ASMI grabbing more of the high-value WFM POR. -25% max loss at ¥4,800. Scale on China headlines.
4966 C. Uyemura — scale-in BUY 35/35/30 over 6-8 weeks, 2.5-3% target portfolio size. ENEPIG monopoly at 70% Taiwan / 50-55% Japan share; 33.9% family ownership held outright; 440bp gross margin step in FY3/25 with consensus modeling FY3/26 EPS below FY3/25 actual. The STF post 4 days ago is the timing risk — edge is closing fast. Hard re-evaluate trigger at ¥20k.
4971 MEC — PASS at ¥11,050; re-engage ¥7,000-9,000. Next decision: Aug 2026 H1 print. Verdict reaffirmed from 5/12. STF thesis from 3/1 has fully played out (stock +100% in 75 days). Mean PT ¥8,900 unchanged; spot ¥11,050. The catalyst window for sell-side re-rating is the next 2-4 weeks — if no upgrades flow, the stock drifts toward PT.
6327 Kitagawa Seiki — starter only at 1/3 target sizing, 1.5-2.5% portfolio cap until Japanese yuho review. FY26E P/E ~11x is genuinely cheap if order book converts. ¥5.4B Q1 order intake is a 10-year high and PCB backlog share 68%→85% YoY is the operational signature. But +329% YoY rally has compressed asymmetry — what was a 5-bagger at ¥600 in mid-2025 is now a 30-50% trade at ¥2,580. The disclosure quality is the most important caveat: no English IR, no sell-side consensus, single-source on key claims via STF. Meaningful position size requires Japanese yuho review first.
7826 Furuya Metal — WATCH; build starter only on pullback to ¥7,500-8,000 or A14 material-of-record confirmation. Ru-as-liner decided, bulk-Ru is the contested next-leg event in the 2027-2028 window. Stock has 5x'd from low; forward P/E 24x is the high end of own history; probability-weighted return ~7%. Family net-worth-concentrated alignment is the strongest signal in the basket but the entry needs to be better.
5344 MARUWA — WATCH; entry zone ¥55-65k. Critical re-framing: this is NOT a GAA-node packaging play, it is the AI-optical-interconnect plumbing one layer downstream of GAA-enabled compute. AlN submounts are physics-locked under every InP-based EML/CW laser chip in 800G/1.6T optics and CPO ELS modules. But at 52x trailing P/E and mean PT at-spot (¥76,800 consensus vs ¥75,500 spot), there is no consensus upside left to capture. Confirmed Q1 FY3/27 beat in Jul 2026 is the upside trigger.
4062 Ibiden — WATCH, not BUY at ¥15,845. TTM P/E 74x. Base-case target gap narrowed only to -11% on the recent 12.5% pullback. The 5/14 STF post on glass-core flags a real risk: Absolics shipping volume samples to AMD for the MI400 series as of January 2026, commercial 2026-2027 — possibly 12-24 months earlier than the prior deep-dive assumed Intel/Ibiden launch. If AMD MI400 substrate sourcing 2H 2026 confirms Absolics, Ibiden's bridge-to-glass-core optionality weakens.
Open items / follow-ups
Three things to chase after this swarm:
First, the SA mirror file at KB/wiki/semianalysis/2023/going-vertical-gate-all-around-3d.md is misrouted — contains Fabricated Knowledge content rather than the actual SA "Going Vertical — GAA 3D" piece. Re-fetch from https://semianalysis.com/2023/10/15/going-vertical-gate-all-around-3d/. This is the highest-value SA piece for the Kokusai thesis specifically.
Second, two STF posts could not be reliably loaded during the swarm runs due to opencli tab-management issues — "Hidden Japanese AI Beneficiary That Saved Nvidia: Monopoly Moat, Cash-Rich and 15x P/E" and "The Substrate Wars" (Ibiden). Title heuristics on the first (15x P/E + monopoly) do NOT match Kokusai (33x P/E, not monopoly) — best guess is one of Nittobo, Seikoh, or Union Tool. Worth re-running substack-reader on these when the bridge is stable, especially the Substrate Wars piece for the 4062/4966/4971/5706 substrate-chain refresh.
Third, Japanese-language yuho reviews are the gate to scaling Kitagawa Seiki above 1% of book — top-10 shareholders, related-party transactions, executive comp structure, FY24 ¥667M equity-raise use of proceeds are all unverified in English. The "bottleneck-tier" thesis holds up on first-principles physics (M9/Q-Glass thermal-uniformity envelope is real, sequential lamination cycle-count multiplier is real, 70-year process-IP moat is real) but the "sole qualified press supplier" claim is unverified outside STF.
Generated by /swarm-research orchestrator on 2026-05-15. Source artifacts: ~/claude/output/{profile,deep-dive,mgmt-dd,checklist}/{ticker}-*.md for each of 4062, 4966, 4971, 5344, 6327, 6525, 7826. STF Research Substack coverage cross-referenced via substack-reader skill in each sub-agent.
Topics
- optical-components
- japan-semi