Register D | Date: 2026-05-12 |
Spot: JPY 16,550 | Mkt cap: JPY 4.62 tn (~USD 30.8 bn) | Companion:
[[4062-profile]],
[[packaging-glass-substrate-primer]]
Register D | Date: 2026-05-12 | Spot: JPY 16,550 | Mkt cap: JPY 4.62 tn (~USD 30.8 bn) | Companion:
[[4062-profile]],[[packaging-glass-substrate-primer]]
This deep-dive builds on [[4062-profile]] (corporate
overview, customer map, segment splits, basic financials, ownership).
The profile covered the “what”. This dive covers the “how, why now, and
what’s it worth” — first-principles substrate engineering, value-chain
position, sector inflection, incremental margin analysis, valuation
under several scenarios, and the decision frame. Industry primer
prerequisite is satisfied —
[[packaging-glass-substrate-primer]] (2026-05-03) and
[[ai-server-pcb-primer]] (2026-05-03) cover the technology
context; not duplicated here.
Three SA pieces in the local mirror directly cover Ibiden’s substrate position; a fourth — the GTC 2025 / Vera Rubin piece (May 2025) — describes the substrate-area extrapolation that underwrites the FY3/27 revenue ramp.
| Date | SA piece | Key claim about Ibiden / substrate | My finding |
|---|---|---|---|
| 2023 | ai-expansion-supply-chain-analysis | “Ibiden is Nvidia’s sole supplier and is expanding capacity”; separate caveat that AI cannibalisation = substrate weakness in dollar terms | Stale on both halves. Sole-supplier framing was Hopper-only; AI substrate dollar content has reversed — per-package area is now 5,625mm² (B200) → 8,100mm² (Vera Rubin). The “substrate weakness” call did not age well. |
| 2024 | gb200-hardware-architecture-and-component | “Unimicron becomes a second source of ABF for Blackwell GPU; main supplier of ABF for Grace CPU with Ibiden as second. Ibiden will still grow, but Unimicron is taking a lot of the slack.” | Confirmed by Ibiden’s own capacity additions and Unimicron’s pulled-forward expansion. Ibiden is no longer monopolist on the flagship socket — but volumes are still up materially because the AI pie outgrew the share loss. |
| 2024 | nvidias-blackwell-reworked-shipment | GB200A NVL36 has lower ABF $-content per GPU vs GB200 NVL72 (smaller B200A, fewer layers, 1:4 vs 1:2 CPU:GPU ratio) — “affects Unimicron and Ibiden.” | Real but second-order. The B200A-NVL36 mix is a minority of total Blackwell ship volume; Rubin (CoWoS-L, 8,100mm² area) reverses the per-unit content trajectory dramatically. |
| 2025 | nvidia-gtc-2025-built-for-reasoning-vera-rubin… | B300 NVL16 will require ABF substrates “beyond current JEDEC package size limits (120mm for both width and height)”; CoWoS-L substrate must carry inter-die communication via two interposers on one substrate. | This is the single most bullish underlying datapoint on Ibiden the SA corpus contains. JEDEC-exceeding substrates are exactly the SKUs where Ibiden’s process moat (low warpage at large panels, multi-build-up dielectric ladder, embedded passives) is hardest to replicate. |
Net SA take: SA’s framing of the substrate franchise has migrated from “monopoly” (2023) → “share loss to Unimicron” (2024) → “extreme package size pushing physical limits” (2025). Ibiden has been the de facto reference customer for every one of those waves. The 2024 share-loss narrative is what kept Ibiden under-owned through 2023-24; the 2025 size-inflection narrative is what re-rated it 6.7x in 12 months.
Thesis (bull) in three sentences. Ibiden is the global leader in high-end ABF / FCBGA IC substrate at the precise moment AI accelerator package areas crossed organic substrate’s prior physical comfort zone (90×90mm) and pushed up to 120×120mm with 20+ build-up layers — a regime in which Ibiden’s process moat is structurally widest. The Oono phase 2 ramp commissioning through 2026 unlocks the FY3/26 (+34% rev) → FY3/27 (+18% rev) step-up that’s now baked into 17-analyst consensus, with EPS doubling FY3/25 → FY3/27E. The same R&D ladder that built the multi-decade ABF moat is also Intel’s preferred development partner on glass core substrate (commercial 2027-2030), so the bridge from organic ABF to glass is structurally Ibiden’s franchise to lose.
Current price: JPY 16,550 Market cap: JPY 4.62 trillion (~USD 30.8 bn) Enterprise value: JPY 4.28 trillion Target price (base case, FY3/27 P/E 50x on JPY 287 EPS): JPY 14,350 — −13% from spot. Bull case (60x FY3/27 EPS): JPY 17,200. Bear case (35x FY3/27 EPS): JPY 10,050. Expected return (base case, 12-18 month horizon): −13% — the stock is consensus-ahead. Conviction: Medium — high quality of business + high quality of through-cycle trajectory, but stretched valuation at the precise moment dual-sourcing is structurally accelerating. This is a quality business at a tactical price.
Recommendation: Quality watchlist, not entry-priced. Wait for either (a) sell-side target re-rating to JPY 18-20k range (closes the consensus-ahead problem), (b) a 15-25% pullback that resets multiples to the high-30s forward P/E range, or (c) Vera Rubin / Rubin Ultra substrate sourcing confirmation that defends share against Unimicron.
Full corporate overview in [[4062-profile]] §1.
Headline: Japan-listed (TSE Prime, 4062), founded 1912 as Ibigawa
Electric, renamed 1982. Three segments: Electronics ~70-75% (substrate
franchise — Intel/Nvidia/AMD/hyperscaler customers), Ceramics ~20-25%
(diesel particulate filters, specialty ceramics), Others ~3-5% (long
conglomerate tail). FY3/25 revenue JPY 369.4B, net income JPY 33.7B.
Latest IR: https://www.ibiden.com/ir/library/.
The asset to watch is the Oono plant in Fukui — JPY 250-300B total capex, phase 1 operating since early 2024, phase 2 commissioning through 2026-2027. This is what FY3/25 –JPY 79.6B free cash flow paid for.
The technology context is in
[[packaging-glass-substrate-primer]] Part I; this section
adds Ibiden-specific process depth that the primer does not
duplicate.
Every advanced silicon die ships sitting on an IC substrate. The die has thousands of microbumps (40-130µm pitch); the PCB has solder balls at 1.0mm pitch. The substrate is the multilayer fan-out that connects the two — engineered to handle, simultaneously:
There is no single component in a server class GPU package — including the GPU silicon itself — that has more engineering constraints concurrently active than the substrate. The substrate is what Intel paid Ibiden to refine for two decades and what Nvidia underestimated until Blackwell.
ABF is a B-stage thermoset epoxy resin film, ~25-40µm thick, that Ibiden laminates onto a core (typically BT-resin/glass-cloth composite). The wet etch, microvia laser drill, copper plate, and resin film cure sequence is repeated 15-25 times to build a 30-50 layer count substrate. Each cycle introduces yield risk; each cycle requires panel-level alignment to micron precision; each cycle is a chance for the resin to over-cure, under-cure, or warp.
Ibiden’s process moat has three distinct components:
Build-up-cycle yield per layer. A 20-layer substrate is 20 sequential laser-drill / plate / film / cure cycles. If yield per cycle is 99.5%, end-of-line yield is 99.5%^20 = 90.5%. If yield per cycle drops to 99.0%, end-of-line yield is 81.8%. The 1% per-cycle yield delta compounds to a near-10% margin delta. Ibiden’s flagship-line yield is reportedly the industry benchmark.
Warpage management at large panel sizes. Modern Ibiden substrate panels are 510×515mm (panel-level packaging) carrying 12-24 flagship substrate dies. Warpage during cure scales non-linearly with panel size; warpage above a critical limit makes downstream die-attach yields collapse. Ibiden has spent two decades refining the press / cure / cooling thermal profile and the panel handling tooling. This is the most under-discussed half of the moat — it is mechanical, not materials.
Embedded passives and complex via geometry. Newer substrates embed decoupling capacitors and signal-conditioning resistors directly inside the substrate, reducing parasitic inductance. Building these embedded components requires alignment that Ibiden’s competitors generally don’t yet replicate at scale.
[BT-glass-cloth core panel (510×515mm) loaded into line]
↓
[Drill core panel through-holes (laser + mechanical depending on size)]
↓
[Plate through-hole walls with copper electroless + electroplate]
↓
[Fill through-holes (resin or copper depending on design)]
↓
[Pattern core copper (image, etch, strip)]
↓ —————————— BUILD-UP LOOP (×N, N=15-25) —————————————————
[ABF film laminated onto core/build-up surface — vacuum laminate]
↓
[Thermal cure (precise temperature ramp profile; this is the critical step)]
↓
[Laser-drill microvias (CO2 + UV combinations) — typically 50-80µm via diameter]
↓
[Desmear / cure-residue cleanup (plasma + chemical)]
↓
[Electroless copper seed → electroplate copper fill]
↓
[Image (photoresist) → develop → etch → strip → repeat for next layer]
↓
[Optical inspection of each build-up layer]
↓ ——————————————————————————————————————————————————————
[Solder mask coat + image + cure]
↓
[Surface finish — typically ENEPIG (electroless Ni / electroless Pd / immersion gold)]
↓
[Electrical test (panel-level + per-substrate)]
↓
[Singulation — laser scoring + routing per substrate]
↓
[Final inspection / packaging]
↓
[Substrate ships to OSAT (Amkor, ASE) for die attach, or to fab for in-house packaging]Each of these steps is industrially mature; the difficulty is in running 15-25 cycles back-to-back with cumulative yield staying above 90%. The end-product price for a flagship AI accelerator substrate (Blackwell-class, 20+ build-up layers, 90×90mm) is in the USD 150-400 range per substrate depending on layer count. Substrate dollar content per AI accelerator now exceeds substrate ASPs of mid-2010s CPU server SKUs by 5-10x.
| Metric | Definition | Why it matters | State of the art |
|---|---|---|---|
| Build-up layers | Count of dielectric layers above (and below) the core | Drives I/O fan-out; higher = denser signal routing; harder to make | 18-26 layers flagship AI; rising |
| Panel size | Largest panel through the line | Drives wafer-throughput economics; warpage rises non-linearly | 510×515mm common; pushing to 600mm |
| Microvia diameter | Smallest laser-drilled via in build-up layers | Drives I/O density | 50-60µm leading edge |
| Cumulative yield | End-of-line good substrate / panels started | Direct margin driver | ~85-92% flagship; ~95%+ mature |
| Warpage | Out-of-flat tolerance over substrate area | Drives downstream die-attach yield | ±50µm tolerance over 90×90mm |
For an investor, the metrics to track over time are: layer count progression (each flagship AI generation adds 2-4 layers; this is direct substrate dollar content uplift), panel size step-ups (cost-down per substrate but capex-up for new lines), and cumulative yield disclosure (rarely disclosed publicly but inferable from gross margin movement).
The substrate franchise. Within Electronics:
FCBGA / ABF substrate (the AI / server CPU SKU) — flagship product. ASPs USD 50-400 per substrate. Customer mix: Intel server CPU (Sapphire/Emerald/Granite/Clearwater Forest cycle); Nvidia data-centre GPU (Hopper, Blackwell, Rubin); AMD EPYC + MI accelerator series; hyperscaler custom silicon (AWS Trainium, Google TPU, Microsoft Maia, Meta MTIA). Replacement cycle: each new silicon generation = new substrate qualification; substrates ship per silicon part shipped. This is the segment driving the rerating.
PCB and plastic packaging — non-flagship, mature. Stable revenue, lower margins. ~5-10% of segment.
Diesel particulate filter (DPF) substrates and ceramic substrate holding mats — for automotive emissions systems. Customers: Cummins, Daimler Truck, Stellantis, others. ASPs in the USD 80-200 range per ceramic substrate.
Construction, building materials, agriculture, food, gas — the Gifu-region conglomerate tail. Not investible thesis-relevant; provides modest stable cash flow.
Underappreciated bit: Ibiden makes graphite specialty for silicon and SiC crystal pulling (Czochralski susceptors, crucibles) — an indirect AI / SiC EV tailwind. Small revenue contributor (~JPY 10-15B) but high-quality. Worth flagging because it ties Ibiden to the SiC power-electronics build-out.
[Specialty resin: Ajinomoto ABF] ─────────────────────┐
[Glass-cloth core: Nittobo, Asahi] ──────────────────┤
[Copper foil: Mitsui, JX] │
[Equipment: Disco, TEL, Applied Materials, Lam, LPKF] │
▼
★ [Ibiden — substrate fabrication, 510×515mm panel]
│
▼
[Substrate ships to OSAT]
│
▼
[Amkor, ASE, SPIL] — die attach, package assembly
│
▼
[Nvidia, Intel, AMD, hyperscalers] — silicon design
│
▼
[OEM / hyperscaler servers] — HGX, MGX, custom racks
│
▼
[End customer: cloud workloads, AI training/inference]Where Ibiden sits: a two-step-removed Tier 2 supplier — the silicon designers are the brand, the OSATs are the assemblers, but the substrate is where the substrate-tier oligopoly margin pool sits. The substrate maker’s value capture is currently ~10-15% of an AI accelerator package’s total BOM value, rising as substrate dollar content scales.
| Supplier | Ticker | Layer | Bypass-ability | Supplier MC vs 4062.T | Pricing |
|---|---|---|---|---|---|
| Ajinomoto Group | 2802.T | ABF resin (single-source) | No — Ajinomoto is monopoly globally | JPY 3.8 tn / 0.82x | Priced-in (ABF is well-known) |
| Nittobo | 3110.T | Glass cloth (multi-source w/ Asahi) | Partial — 2-3 firms | JPY 200B / 0.043x | Partial; specialised glass-cloth Q-glass narrative emerging |
| LPKF Laser & Electronics | LPK.DE | TGV laser equipment (glass core future) | Partial — E&R Engineering 8027.TWO is dual source | EUR 250M / 0.008x | Under-priced (early stage) |
| E&R Engineering | 8027.TWO | TGV laser equipment | Partial | TWD ~10B / 0.0014x | Under-priced |
| Disco | 6146.T | Substrate dicing/singulation | Partial — multi-source | JPY 3.0 tn / 0.65x | Priced-in |
| Mitsui Kinzoku, JX Nippon | 5706.T / etc | Copper foil | Diversified | various | Priced-in |
Bottleneck verdict. The single hardest-to-replace upstream node is Ajinomoto — they are the de facto monopoly supplier of ABF resin film globally, and every advanced organic substrate ships with their material. Ajinomoto is the only upstream node where Ibiden has no substitutability optionality. Their food business mute most of the substrate-cycle valuation premium — Ajinomoto trades as a consumer staple, not as a semi material — so this is a quasi-mispricing but is well known by sector specialists. Not a Pink-actionable small-cap mispricing.
The actionable small-cap upstream is E&R Engineering
(8027.TWO) — TGV laser equipment for the next
substrate generation (glass core), where Ibiden is one of the lead
qualification customers per the glass-substrate primer. Pink already has
this in [[packaging-glass-substrate-primer]] and on her
watch as a 2027-2030 industry inflection. Flag here, do not run separate
deep-dive.
The customer table is in [[4062-profile]] §2.
Headline:
Concentration risk in plain English. If Nvidia walked tomorrow (it won’t — qualification cycle is 18-36 months), Ibiden loses 20%+ of revenue and ~ 30-40% of operating income (AI substrate is highest-margin SKU mix). The Nvidia loss is structurally implausible over a 12-24 month horizon because (a) the substrate qualification is already locked for Vera Rubin generation, (b) the alternative supplier (Unimicron) has its own capacity-allocation tension, (c) Ibiden’s panel-level yield is the industry reference. The risk is not customer loss; the risk is share rebalancing within Nvidia’s flagship sockets — and that’s already in train, per SA 2024.
The Intel relationship is the most strategically valuable single customer. Intel’s substrate spend doesn’t grow as fast as Nvidia’s, but Intel is the customer with whom Ibiden is co-developing glass core substrate process — and Intel is the customer most likely to be Ibiden’s first commercial glass core volume in 2027-2028. The Intel relationship is what underwrites the 2027-2030 thesis. Intel losing more server CPU share to AMD is the largest single risk to the Ibiden / Intel substrate dollar volume; Ibiden does qualify on AMD CPU substrate too, but the Intel SKUs are higher-end and higher-margin per unit. AMD’s gain at Ibiden does not fully offset Intel’s loss — that’s the structural tension in the Intel relationship.
Strategic partnerships: - Ajinomoto
Group — single-source ABF resin supplier. Long-term industrial
supply relationship. Not a JV; pure customer-supplier; Ibiden is one of
Ajinomoto’s largest substrate-material customers globally. -
Intel — multi-decade qualified-supplier relationship,
including glass core development under Intel’s “GlassCore” program (per
[[packaging-glass-substrate-primer]]). - METI /
Japanese gov. — informal industrial-policy alignment; Oono
plant has received published METI subsidies. - Specialty glass
suppliers (AGC 5201.T, NEG 5214.T, Corning, Schott, Hoya
7741.T) — emerging qualification relationships for glass core
substrate. Hoya was the subject of a vault deep-dive — see
[[7741]] for glass photomask context, but Hoya is not
currently Ibiden’s named glass core supplier.
Substrate TAM and market share specifics are in
[[4062-profile]] §3. Headline: - High-end FCBGA /
ABF served market: USD 6-8B in 2025 → USD 12-16B by 2030. -
Ibiden ABF share: ~25-30% overall, ~50-70% in flagship
AI accelerator SKUs. - Glass core substrate TAM: USD
0.5-2B by 2027, growing to USD 5-10B by 2030 (Yole / Prismark
consolidated).
Tailwinds (multi-year): - AI accelerator volume + substrate area scaling - Heterogeneous integration (chiplet, 3D stack) - Glass core transition (Ibiden positioned as leader) - METI / Japan reshoring of advanced packaging capacity
Headwinds (multi-year): - Substrate-tier oligopoly is fracturing — Unimicron + Samsung E-M actively gaining - Diesel particulate filter business in secular decline - Hyperscaler vertical integration risk (long-tail — they might bring substrate qualification in-house)
This is the most important section of the dive. Why is Ibiden a 2026 buy, not a 2024 buy or 2028 buy? The answer is the convergence of four independent inflections.
Demand inflection. AI accelerator substrate area has scaled non-linearly across three generations: Hopper 3,025mm² → Blackwell 5,625mm² (+86%) → Vera Rubin 8,100mm² (+168% vs H100). The per-unit substrate dollar content has risen even faster because layer count has also gone up. Per Ibiden’s own commentary on the Q1 FY3/26 print (May 11, 2026), substrate revenue per AI accelerator is up 2.5-3x in three generations. The unit count is growing. The per-unit dollar content is growing faster. Compounded, substrate revenue is now growing at +30-50% YoY through FY3/27.
Supply constraint. Industry capacity has been adding incrementally — Unimicron pulled forward 3-4 quarters; Samsung E-M expanding; AT&S ramping Kulim Malaysia. But the highest-end SKU (20+ build-up layers, panels exceeding 100×100mm) has structurally tighter capacity because qualification cycles run 18-36 months and panel sizing requires net-new line investment. The flagship-tier capacity is tight through 2026; phase 2 Oono is the marginal source of high-end supply globally.
Inventory cycle. Substrate distributor inventory in the AI-accelerator SKU is short — there’s no spot market for flagship substrates, all volume is contract / qualified-source allocation. The destocking that hurt mid-range ABF in 2023 (per SA’s 2023 piece) has fully washed out at the flagship-AI tier; mid-range and consumer ABF inventory is normalising late-2025 to early-2026.
Coming shortage / glut (12-36 months out). At the flagship AI accelerator tier, supply remains tight through 2027 — assumes Vera Rubin (2026-2027), Rubin Ultra (2027-2028), and the next two AMD MI generations all qualify on the same substrate-maker shortlist. Glut risk is concentrated in mid-range ABF (consumer / enterprise CPU substrates) where Unimicron + AT&S are adding capacity faster than demand. Ibiden is overweighted in the tight-supply tier and underweighted in the glut-risk tier — this is by mix, not by accident.
What changed in the last 6-24 months:
The AI substrate dollar pool is now an order of magnitude larger. As of H1 2025, total AI accelerator substrate revenue at industry level was ~USD 1.5-2B per year. As of run-rate H2 2026, it’s ~USD 4-5B. This is the fastest doubling in substrate industry history.
Substrate quality competition has bifurcated. The flagship AI tier (Ibiden, Unimicron, partially Shinko) is structurally different from the mid-range commercial tier (AT&S, Nan Ya PCB, Kinsus). The bifurcation didn’t exist five years ago; it’s now structural.
Glass core has moved from R&D to production
qualification. Intel’s January 2026 NEPCON Japan demo (per
[[ATS]]/ats-deep-dive.md) of thick-core glass substrate
with EMIB was the substrate industry’s “demo day.” Ibiden is — per the
AT&S deep-dive — the most likely first commercial glass core
supplier; AT&S is the option. The clock on glass core is now
public.
Japan industrial policy is actively de-risking Ibiden’s Oono capex. METI explicit substrate subsidies (alongside Rapidus, JASM-Kumamoto support) reduce the project IRR’s downside.
What sell-side is currently missing. Mean target JPY 10,441 vs. spot JPY 16,550 = 37% below current. The mean target was set before the May 11 Q1 FY3/26 print (+152% EPS surprise). Even the high target (JPY 18,400) is barely above spot. Sell-side has not yet revised after the print. Forward EPS estimates for FY3/27 still range JPY 229 - 343 (mean JPY 287). At the high end (JPY 343), 50x P/E implies JPY 17,150 — i.e., even the high consensus EPS doesn’t justify the current spot at the 50x multiple the stock currently trades at on FY3/27 P/E. The market is paying for an upside FY3/27 EPS print of JPY 330-380, which requires substrate mix + capacity + share continuation to all break favourably.
Narrative vs reality gap. The 2024 narrative (Unimicron taking share, Ibiden derated) is the one consensus targets reflect. The 2025-2026 narrative (substrate area scaling, dollar content rising, Vera Rubin re-anchoring Ibiden’s flagship position) is the one the stock has run on. The gap closes one of two ways: sell-side revisions catch up (re-rating consolidates), or sell-side revisions don’t come through fast enough and the stock corrects (re-rating cools). The Q1 FY3/26 print is the inflection point — the next 60 days of revisions will tell which way.
Near-term (0-12 months): - Late July / early August 2026 — Q2 FY3/26 earnings. Continuation print on Q1’s momentum. If EPS comes in JPY 50-70 (vs JPY 32.85 consensus), full-year FY3/26 revisions move toward JPY 250-280 range, and the FY3/27 P/E narrows to 40-45x. The single most consequential 12-month catalyst. - Sell-side target revision wave (next 30-60 days). Mean target JPY 10,441 → likely JPY 15,000-18,000 range over the next 60 days. Watch for revisions concentrating in the JPY 18-22k range — that signals consensus catch-up to spot. - Vera Rubin substrate supplier disclosure (likely H2 2026). If Ibiden retains primary share, multiples can expand further; if Unimicron is named primary, the share-loss narrative re-emerges.
Medium-term (1-3 years): - Glass core first commercial volume (Intel SKU, est. 2027-2028). If Ibiden is named, premium pricing kicks in; if not, ABF-tier reversion happens. - Oono phase 3 capex decision (timing uncertain, likely 2026-2027). Signals demand visibility into 2028+; further capex commitment is bullish for thesis but bearish for near-term FCF. - Diesel DPF segment transition — needs visibility on Hungary plant repositioning. Lagging timeline.
Leading indicators to watch. - Substrate ASP commentary in quarterly result decks (look for “high mix-up” language). - Capex run-rate at Oono (FY3/26 likely JPY 150-200B again; declines if peak reached). - Sell-side revisions clustering above JPY 15,000. - Unimicron’s quarterly result deck — substrate revenue YoY % is the read-through on Ibiden’s share position. - Intel’s GlassCore process milestones (quarterly tech-day reads).
Demand inflection is real (AI substrate area scaling), supply is structurally constrained at the flagship tier (Ibiden’s tier), sell-side is materially behind the print (mean target 37% below spot), and the catalyst path (Q2 print + sell-side revisions + Vera Rubin disclosure) is concentrated in the next 6-12 months. The “why now” is unambiguous. What is also true is the stock has already moved 6.7x in 12 months — the why-now thesis is partially priced. The cleanest formulation: this is a quality business with a real thesis but a tactical entry-point problem.
Full management deep-dive in companion file
[[4062-mgmt-dd]] (output by /mgmt-dd skill).
Headline items relevant to deep-dive:
The detailed forensic pass (insider trade history via EDINET,
related-party transactions in the Yuho, AGM convocation notice
review, golden parachute quantification) is in
[[4062-mgmt-dd]].
The full competitive table is in [[4062-profile]] §5.
Compressed for decision-making:
| Company | Ticker | Substrate share | Pure-play? | Position vs Ibiden |
|---|---|---|---|---|
| Unimicron | 3037.TW | ~20-25% global ABF | Mostly substrate | Aggressive share-gainer; Blackwell secondary, Grace primary, expanding Vera Rubin capacity |
| Shinko Electric | 6967.T | ~15-20% | Pure-play | Subject of JIC tender offer (going-private); strategic intent unclear |
| AT&S | ATS.VI | ~5-10% | Mostly substrate | European; AMD anchor; Kulim Malaysia ramp; Intel TGV collaboration |
| Nan Ya PCB | 8046.TW | ~8-12% | Mixed (PCB + substrate) | Lower-end consumer SKUs; Apple adjacency |
| Samsung Electro-Mechanics | 009150.KS | ~10-15% | Mixed | Aggressive glass core; hyperscaler ambitions |
| Kinsus | 3189.TW | ~3-5% | Substrate | ASIC-focused; smaller scale |
5-year lock-up test. Would I happily own Ibiden if the market closed for 5 years? Mostly yes. The substrate franchise is structural. Risk: if I’m forced to hold across a substrate cyclical down-leg, I take the volatility on the chin; if I’m forced to hold across a glass-core transition where Ibiden mis-executes, the bull case decays. Verdict: yes for a 5% position, qualified for larger.
Unique economic engine. Ibiden’s unique engine is the combination of (a) panel-level yield in the 20+ build-up layer flagship SKUs (highest in industry), (b) Intel co-development access (long-term substrate roadmap), and (c) Japan industrial policy capex backing. Durability: 5-10 years easily on the substrate-tier moat; the glass core question is the 10+ year horizon.
Blank-check disruptor. Could a competitor with infinite capital disrupt Ibiden? Partially. Capex barrier is real but not absolute (Samsung E-M is funded to be disruptive). What capital alone cannot buy: the 35-year process knowledge curve and the qualified-customer trust. Disruptor scenario requires either (a) capital + a 10-year process learning curve, or (b) a step-change technology that bypasses Ibiden’s accumulated process knowledge — glass core is the candidate. Glass core where Ibiden is also positioned as a leader is defensive, not disruptive, to Ibiden.
Quality verdict: high-quality, durable franchise with a glass-core transition that is more likely to be a tailwind than a headwind for Ibiden specifically. The investment risk is in price, not in business quality.
Earnings surprise progression (8 quarters):
| Quarter | EPS Estimate | Reported EPS | Surprise % | Direction |
|---|---|---|---|---|
| Apr 2024 | 28.87 | 14.66 | –49% | down |
| Aug 2024 | 26.54 | 31.55 | +19% | up |
| Oct 2024 | 33.25 | 41.91 | +26% | up |
| Feb 2025 | 29.63 | 15.32 | –48% | down (mix-down qtr) |
| May 2025 | 22.28 | 31.88 | +43% | up |
| Aug 2025 | 32.85 | 45.59 | +39% | up |
| Oct 2025 | 34.09 | 33.45 | –2% | meet |
| Feb 2026 | 36.98 | 31.99 | –14% | down |
| May 2026 | 46.40 | 117.14 | +152% | massive up — the print that drove the rerating |
Second-derivative read. The trend through Feb 2026 was lumpy (a function of cyclical destock + mix). The May 2026 print is a step-function up — EPS run-rate is now JPY 117 quarterly, or JPY 350+ if sustained four quarters. If even half of that ($175 EPS) sustains as run-rate, FY3/27 EPS pushes to JPY 350-400 vs current consensus JPY 287. The second derivative is materially positive; consensus has not caught up.
Implied exit rate FY3/26: ~JPY 250-280 (vs consensus JPY 217). Implied FY3/27 if Q2 momentum continues: JPY 350-400 (vs consensus JPY 287). At spot JPY 16,550, this implies FY3/27 P/E of 41-47x on the upside print — still rich, but not absurd for a flagship substrate maker in mid-cycle expansion.
| Metric | Value |
|---|---|
| Market cap | JPY 4.62 tn |
| Enterprise value | JPY 4.28 tn |
| P/E (TTM, EPS ~214) | 77.2x |
| P/E (FY3/26E, EPS 217) | 76.3x |
| P/E (FY3/27E mean, EPS 287) | 57.6x |
| P/E (FY3/27E high, EPS 343) | 48.3x |
| P/E (FY3/27E bull, EPS 380, my est.) | 43.6x |
| EV/EBITDA (TTM) | 34.4x |
| EV/Revenue (TTM) | 10.3x |
| P/B | 8.5x |
| FCF yield | negative (FY3/25 capex peak) |
| Dividend yield | 0.26% |
| 52-week range | JPY 2,425 – 18,365 |
| Metric | FY3/22 | FY3/23 | FY3/24 | FY3/25 | LTM | FY3/26E | FY3/27E |
|---|---|---|---|---|---|---|---|
| Revenue | 401.1 | 417.5 | 370.5 | 369.4 | ~416 | ~495.6 | ~584.5 |
| Revenue growth | — | +4.1% | –11.3% | –0.3% | +13% | +34.2% | +17.9% |
| Gross margin | 29.9% | 30.6% | 27.7% | 30.7% | ~31.6% | ~33% | ~34% |
| Operating margin | 17.7% | 17.3% | 12.8% | 12.9% | ~14.1% | ~20% | ~24% |
| Net margin | 10.3% | 12.5% | 8.5% | 9.1% | ~15.3% | ~12.5% | ~14.5% |
| EPS (JPY) | 147.7 | 186.9 | 112.4 | 120.7 | ~214 | ~217 | ~287 |
| Metric | FY3/22 | FY3/23 | FY3/24 | FY3/25 |
|---|---|---|---|---|
| Operating cash flow | 108.4 | 125.7 | 145.2 | 118.9 |
| Capex | (67.2) | (104.0) | (86.4) | (198.5) |
| Free cash flow | 41.2 | 21.7 | 58.8 | (79.6) |
| Cash & equivalents | 185.6 | 302.4 | 443.6 | 390.7 |
| Total debt | 170.2 | 270.2 | 343.6 | 343.1 |
| Net debt (cash) | (15.4) | (32.2) | (100.0) | (47.6) |
| Total equity | 370.7 | 425.6 | 501.8 | 497.3 |
Ibiden discloses quarterly results in Kessan Tanshin format; the granular quarterly P&L decomposition is in the JP filings and not fully exposed by yfinance. Using the available annual data + quarter EPS print to back into operating leverage:
Estimated FY3/26 vs FY3/25 (current run-rate):
| FY3/25 | FY3/26E (mid) | Δ | |
|---|---|---|---|
| Revenue | JPY 369B | JPY 495B | +126B |
| Gross profit (est) | JPY 113B | JPY 163B | +50B |
| Operating income (est) | JPY 48B | JPY 99B | +51B |
| Incremental Gross Margin | — | — | ~40% |
| Incremental Operating Margin | — | — | ~40% |
Estimated FY3/26 → FY3/27:
| FY3/26E | FY3/27E | Δ | |
|---|---|---|---|
| Revenue | JPY 496B | JPY 585B | +89B |
| Gross profit (est) | JPY 163B | JPY 199B | +36B |
| Operating income (est) | JPY 99B | JPY 140B | +41B |
| Incremental Gross Margin | — | — | ~40% |
| Incremental Operating Margin | — | — | ~46% |
Read. Incremental gross margin is running ~40%, consistent with substrate mix (flagship AI SKU running materially higher gross margin than the FY3/25 base mix). Incremental operating margin is rising into the mid-40s, signalling that R&D + SG&A is staying disciplined while the gross profit increment flows through. This is textbook mid-cycle operating leverage — incremental margin > base margin, confirming the cycle has not topped on margins.
If the May 11 print run-rate sustains (JPY 117 EPS × 4 ≈ JPY 350+ annual), incremental operating margin in H2 FY3/26 will run 55-65% — exceptional and likely unsustainable beyond a few quarters but consistent with a once-a-decade up-cycle.
One-time distortions to flag: FX is a tailwind (weak JPY benefits Ibiden’s export-priced substrates). If JPY rallies to JPY 130/USD, ~5% of operating margin compresses purely on FX. The cycle peak EBIT margin is also flattered by Oono phase 2 ramp efficiency — when capex normalises and depreciation step-up arrives in FY3/28-29, reported EBIT margin compresses from peak even with revenue growth.
Pre-2024 Ibiden traded in the JPY 4,000-7,000 range with FY3 P/E 15-25x. Through 2024-2025, the substrate up-cycle started pricing in: P/E expanded to 30-50x range. The May 2026 print triggered the parabolic move: P/E now 57-77x on consensus FY3/27 / FY3/26.
Peer multiples (rough; cross-listing comparison):
| Company | P/E (FY3/27E or FY+1E) | EV/EBITDA |
|---|---|---|
| Ibiden | 58x | 34x |
| Shinko Electric | ~25-30x (going-private context) | mid-teens |
| Unimicron | ~25-30x | mid-teens |
| AT&S | NM (loss-making; turning) | 10-14x |
| Nan Ya PCB | ~12-15x | low-teens |
| Samsung Electro-Mechanics | ~12-15x | ~8x |
Ibiden trades at a 2x premium to its substrate peers on forward P/E. Reasonable for the flagship AI mix exposure and Intel relationship, but stretched against the high end of historical valuation. The peer premium of 2x suggests Ibiden’s flagship-tier moat is at minimum partially priced in.
| Scenario | FY3/27E EPS | Target P/E | Implied price | vs spot JPY 16,550 |
|---|---|---|---|---|
| Bear — destock returns 2H 2027, Vera Rubin secondary slot to Unimicron, JPY rallies to 130 | JPY 230 (low-end consensus) | 30x | JPY 6,900 | –58% |
| Base — consensus FY3/27 EPS lands at JPY 287, AI substrate mix sustains, no major share shift | JPY 287 | 50x | JPY 14,350 | –13% |
| Bull — Q1 momentum sustains; FY3/27 EPS JPY 350-380; Vera Rubin reaffirmed Ibiden primary | JPY 350 | 50x | JPY 17,500 | +6% |
| Super-bull — glass core volume confirmed for 2027 Intel server; substrate-area inflation continues into Rubin Ultra | JPY 400 | 55x | JPY 22,000 | +33% |
Probability-weighted average: bear 15% × JPY 6,900 + base 40% × JPY 14,350 + bull 35% × JPY 17,500 + super-bull 10% × JPY 22,000 = JPY 15,103. Roughly 9% below spot. The risk/reward is symmetric at best; modestly bear at current price.
A back-of-envelope DCF: FY3/26 revenue JPY 495B, growing 18-20% near-term, decelerating to 10% mid-term, terminal 4%. Operating margin expanding to 25% mid-cycle, declining to 18% terminal. Capex normalising to 18% of revenue. WACC 6.5%. Terminal value at 15x EBIT at year 10. DCF range: JPY 11,000-15,000 fair value depending on assumed terminal margin. Consistent with the multiple-based base case.
At JPY 16,550, the market is implicitly assuming FY3/27 EPS lands in the JPY 330-380 range and trades at 45-50x. That is closer to my bull case than my base. The stock is priced to deliver, not priced for the consensus print.
The premium relative to substrate peers (~2x) is partially earned (flagship AI mix, Intel relationship, glass core option), but is also reflecting the fact that sell-side hasn’t yet updated targets post the May print. If consensus FY3/27 EPS revises from JPY 287 to JPY 340, the 50x P/E gives spot JPY 17,000 — i.e., spot validates with revision. If FY3/27 consensus stays at JPY 287, spot looks 15-20% rich. Watch the next 30-60 days of sell-side revisions.
No publicly disclosed individual contract values (standard Japanese practice). Implied contracts: multi-year flagship supply for Intel, Nvidia, AMD; METI subsidies tied to Oono (specific amounts disclosed in Japanese government industrial policy filings but not consolidated here). No single contract makes or breaks the thesis — the thesis is aggregate substrate volume + mix, not a single deal.
| Risk | Likelihood | Existing mitigants | Mgmt de-risk plan | Closeable? |
|---|---|---|---|---|
| Vera Rubin substrate share rebalanced toward Unimicron | Medium | Multi-decade Nvidia trust; flagship-tier yield benchmark | Oono phase 2 capacity announcement; capacity LTAs | Partly — share contained, not fully recovered |
| Sell-side revisions don’t catch up to spot price | Medium-High | Q1 print is recent and large | Stronger IR / forward guidance | Not closeable in 30-60 days — depends on next print |
| Q2 FY3/26 EPS disappointment (vs run-rate) | Medium | Q1 was Yes/No moment; Q2 needs confirmation | Conservative guidance; one-time benefits flagged | Self-resolves at Q2 |
| JPY appreciation | Medium | Natural hedge from Japan cost base | Selective overseas pricing | Not closeable structurally |
| Substrate cyclical destock 2027-2028 | Medium-High | Ceramics + Others non-cyclical buffer; net-cash balance sheet | Inventory discipline; capex flexibility | Cyclical — closes itself over 12-18 months |
| Glass core mis-execution | Low-Medium | Intel co-development; specialty glass supplier relationships | Active R&D; capacity for 2027 | Closeable as glass core volume materialises |
| Diesel DPF secular decline | High (multi-year) | Slow decline; existing fleet replacement | Hungary plant repositioning | 5-10 year transition; manageable |
| Hyperscaler vertical integration | Low (5-year) | Capex + scale + qualification cycle barrier | Continued flagship-tier specialisation | Not actively closeable; tail risk |
Standard CEO-led Japanese corporate. CEO Kawashima career Ibiden; succession is internal. Low.
What would make you wrong: 1. Q2 FY3/26 print disappoints — EPS comes in JPY 40-60 vs current expectations of JPY 50-70. Stock drops 20-30% on multiple compression. 2. Vera Rubin substrate supplier disclosure names Unimicron as primary — repeat of Blackwell share loss. Stock derates back to pre-rerating multiples. 3. JPY rallies from 150 to 130 — 5% operating margin headwind; spot returns 15-20%.
Downside target price: Bear case scenario above gives JPY 6,900-10,000 (multiple compression to 30-35x FY3/27E EPS). That’s –40% to –58%.
What would invalidate the thesis: Q2 FY3/26 print disappointing combined with Vera Rubin share loss. The substrate dollar content thesis depends on Ibiden’s flagship-tier customer relationships holding; if both are challenged simultaneously, the bull case collapses.
Full ownership detail in [[4062-profile]] §10 and
[[4062-mgmt-dd]]. Key items:
The single most important sell-side datapoint: Ibiden trades above the mean target. This is the cleanest “consensus-ahead” signal in the substrate tier.
Conviction: Medium. Quality is high; valuation is the constraint.
Sizing recommendation: If buying today, start at 1.5-2% portfolio weight maximum, with scale-in triggers below.
Entry strategy: Scale-in across three tranches rather than single entry. - Tranche 1 (1% of portfolio): wait for either a 10-15% pullback OR a sell-side target revision putting mean target above JPY 14,500. - Tranche 2 (additional 1%): on confirmed Q2 FY3/26 beat AND Vera Rubin primary slot retention. - Tranche 3 (additional 1%): on glass core volume confirmation for 2027.
Stop-loss / re-evaluation triggers: - Hard stop: spot below JPY 12,500 (–25% from current). Triggers re-evaluation, not blind exit — usually means Q2 print disappointed. - Soft re-evaluation: Vera Rubin secondary status confirmed; trim by half.
Add triggers: - Vera Rubin primary substrate slot reaffirmed → add to full position. - FY3/27 EPS revision into the JPY 320-380 range → add to full position. - Spot in the JPY 13,000-14,000 range → reset entry tranches.
Trim triggers: - Spot in the JPY 18-22k range → start trimming on overhead resistance. - FY3/27 EPS revisions fade below JPY 280 → trim by half.
This deep-dive drew on yfinance + Ibiden FY3/25 disclosures + SA
mirror coverage. For filings-grade DD, see [[4062-filings]]
(output of /filings skill, currently pending for this
swarm) and [[4062-mgmt-dd]]. Key Japan-specific filings to
reference:
Insider transaction data for Japanese-listed Ibiden flows through
EDINET’s Yakuin Tokutei Yu-uka-Shouken Tousho-shozai filing
system rather than the US Form 4 feed. This is the
/mgmt-dd skill’s responsibility — not pulled
here.
| Source | Use |
|---|---|
| Yfinance financials, estimates, recommendations, calendar | Quantitative data |
| Ibiden FY3/25 Yuho / Kessan Tanshin (via JP IR) | Segment splits, FY trajectory |
[[packaging-glass-substrate-primer]] (May 3, 2026) |
Glass core context, TGV equipment ecosystem |
[[ai-server-pcb-primer]] (May 3, 2026) |
High-layer PCB substrate adjacencies |
| SA mirror: ai-expansion-supply-chain-analysis (2023) | Sole-supplier framing, ABF cycle |
| SA mirror: gb200-hardware-architecture-and-component (2024) | Blackwell dual-source detail |
| SA mirror: nvidias-blackwell-reworked-shipment (2024) | B200A vs B200 dollar content |
| SA mirror: nvidia-gtc-2025-built-for-reasoning-vera-rubin (2025) | JEDEC-exceeding substrate size, Vera Rubin context |
[[ATS/ats-deep-dive]] |
AT&S Intel collaboration on TGV; peer pricing benchmarks |
[[4062-profile]] |
Corporate overview, customer map, financial baseline |
~/claude/output/deep-dive/4062-deep-dive.md (this file)
— will sync to KB/wiki/4062/4062-deep-dive.md.KB/index.md to be updated.KB/log.md to be appended.KB/inv-q.md “Own Research to Review” to be
appended.Deep-dive drafted 2026-05-12 against yfinance (verified live), SA mirror (4 pieces), vault primers (glass substrate, AI server PCB), and existing 4062-profile. Forward EPS scenarios use 16-analyst consensus range; valuation framework adds bull/super-bull beyond consensus. Glass core context cross-references primer without duplication.