Ibiden

Phase A — Understand the business

Lens 1 · Company Overview

Ibiden is a 113-year-old Japanese industrial-materials company (founded 1912 in Ogaki, Gifu Prefecture, originally as a hydroelectric power generator — hence "IBI-DEN", Ibi River electricity). It has quietly become the most important company almost nobody in AI names: the world's leading maker of the high-end ABF (Ajinomoto Build-up Film) flip-chip ball-grid-array (FC-BGA) package substrate — the multi-layer organic "motherboard-under-the-chip" that carries the silicon die, routes thousands of I/O connections, and delivers power to high-end CPUs and, now, AI GPUs.

Three reportable segments (H1 FY2025, Apr–Sep 2025, net-sales mix):

• Electronics — ¥113,547M, 58.1% of sales (op. profit ¥25,429M, 22.4% margin). The IC package substrate + printed-wiring-board business. This is the AI story and where the market cap lives.
• Ceramics — ¥39,689M, 20.3% of sales (op. profit ¥3,839M, 9.7% margin). Diesel particulate filters (silicon-carbide DPF) for autos + graphite/specialty ceramics + ceramic fibers.
• Others — ¥42,248M, 21.6% of sales (op. profit ¥3,250M). Construction materials, services, etc.

Business model. Electronics is a capital-intensive, technology-gated manufacturing oligopoly: build a multi-billion-dollar fab, qualify a 16–20-layer substrate at sub-10-micron tolerances over a ~36-month cycle, and sell into a handful of chip customers who cannot get supply anywhere else. Contract structure has flipped in Ibiden's favor: management's stated policy is "conclude contracts with customers and then receive an amount equivalent to the investment as an advance payment". The balance sheet carried ~¥92.1B of customer advance payments — Nvidia, Intel, AMD and hyperscalers effectively pre-funding the capacity. That is the single most important structural fact about this company: its customers compete to reserve its output and pay upfront.

Customer concentration, historical and current. For ~25 years Intel was 70–80% of Ibiden's substrate revenue. The AI wave has diversified that base toward Nvidia, AMD, and custom-ASIC buyers, shifting the power dynamic from "Intel dictates" to "everyone queues." Research-layer customers.csv is empty — this is ``-only and worth verifying against the next annual report.

Lens 2 · Supply Chain

Map, upstream → Ibiden → end customer, named at every link:

Upstream (inputs — and the constraint above Ibiden):

• Ajinomoto Fine-Techno (subsidiary of Ajinomoto, 2802.T) — makes the ABF dielectric film itself, with a ~95%+ global monopoly from two Japan-based plants. A high-performance-computing substrate uses >10× more ABF film than a PC substrate. This is a chokepoint above Ibiden — Ibiden's own management flagged it has "not yet secured materials to accommodate potential upside," i.e. ABF allocation can cap growth even when Ibiden's fabs have room. Single most underappreciated dependency in the chain.
• Copper foil, laminates, photoresist, plating chemistries, ultra-pure water — commodity-ish, multiple suppliers.
• Fab equipment (laser drilling, lithography, plating lines) — long lead-time capex.

The company: Ibiden fabricates the FC-BGA substrate across its Gifu-Prefecture plant cluster — Ono Plant (started AI-server substrate output Oct 2025, ~50% of building utilized), Kawama Plant (Cell 6 the first phase of the ¥500B expansion, production FY2027), and the flagship Ogaki operations.

Downstream (the chokepoint feeds the whole AI stack):

• OSAT / foundry packaging: TSMC — Ibiden's organic substrate is the carrier inside CoWoS-L (Chip-on-Wafer-on-Substrate, Large), the packaging TSMC uses for Nvidia Blackwell. After the logic die + HBM stacks are bonded, they sit on an organic substrate "made by Unimicron or Ibiden."
• Chip designers: Nvidia (GB200/Blackwell), Intel, AMD, plus custom-ASIC programs — Ibiden and Unimicron are prioritizing Nvidia for CoWoS substrate needs.
• End buyers: the hyperscalers (the four largest plan up to $725B capex in CY2026, +77% YoY ) → AI data centers.

Chokepoints: (1) ABF film at Ajinomoto (95%+); (2) high-layer-count FC-BGA at Ibiden (70–80% of the hardest tier); (3) CoWoS capacity at TSMC. All three are sold out. Ibiden sits at the second, hardest-to-replicate node. Names present — lens passes.

Lens 3 · Competitive Advantages (moats)

Ibiden's moat is process + scarcity + qualification lock-in, and it widens as the chip package gets harder:

1. Yield at the frontier. AI substrates need 16–20 routing layers (vs 4–8 for PCs), >60 discrete process steps, sub-10-micron tolerances. At a 120mm body an AI panel holds only ~16 packages, so a single defect is a 6.25% yield hit vs ~1% for PC substrates. Cumulative yield on 20-layer/5-micron parts is materially below the 75–80% seen on 14-layer. Ibiden runs this tier at commercial yield; rivals are described as "one generation behind."
2. Highest-complexity share: 70–80% in the 20+ layer, 100+mm segment — i.e. exactly where Nvidia's GPUs live. In the broader high-end FC-BGA market Ibiden is ~35% (Shinko ~18%, Unimicron ~14%, AT&S ~10%, Nan Ya ~5%). The premium narrows to a near-monopoly as you climb the complexity curve.
3. Capital + time barrier. A new entrant needs a billion-dollar fab and a ~36-month qualification cycle and ABF allocation from a monopolist that favors incumbent high-volume buyers. Three compounding gates.
4. Bargaining power has inverted. ¥92.1B in customer prepayments and customers co-funding ~50% of expansion capex is the clearest possible market signal that the buyers need Ibiden more than Ibiden needs any one buyer — a reversal from the 25-year Intel-dependence era.

Where the moat is softer: it does not extend to the legacy substrate (PC / standard server), which is commoditizing and price-pressured; and the ABF-film dependency means part of Ibiden's fate is decided one link upstream. Ground truth on positioning.md/bottlenecks.md was unavailable (hardware wikis not yet written), so this lens is ``-sourced.

Lens 4 · Segments

Primary-source segment detail, H1 FY2025 (Apr–Sep 2025) vs prior-year H1, plus full-year FY2025 company forecast:

The trend that matters: Electronics op. profit grew +41.4% YoY in H1 (¥17,990M → ¥25,429M) and the segment margin jumped from 18.3% to 22.4% — and the full-year forecast implies Electronics does ~79% of group operating profit on ~61% of sales. The blended 14.5% group margin understates the AI engine because Ceramics (structurally declining on Europe's EV shift away from diesel) and Others drag the average. A separate read puts the Electronics 9M FY2025 margin at 19.2%, up from 13.6% in FY2024, absorbing ¥44.5B of new depreciation — i.e. the mix shift to AI substrate is powering through a heavy D&A ramp. Geographic split not sourced at segment granularity — n/a.

---

Phase B — Measure performance

Lens 5 · Earnings Result

Latest reported: H1 FY2025 (six months to 30 Sep 2025):

• Net sales ¥195,485M, +7.7% YoY.
• Operating profit ¥32,573M, +14.2% (margin 16.7%, up from 15.7%).
• Ordinary profit ¥32,305M (+9.4%); profit attributable to owners ¥22,069M, +7.5%.
• Balance sheet (as of 30 Sep 2025): total assets ¥1,037,938M; net assets ¥532,153M; capital adequacy ratio 50.6% (up from 45.3% at Mar-2025) — a conservatively financed balance sheet.
• H1 capex ¥34,810M (down 58% YoY as the last cycle's build completed); D&A ¥25,429M and rising; FY2025 capex plan ¥100,000M, D&A ¥70,000M.
• FX assumption ¥145/USD (vs ¥153 prior).

The one caveat on the print. The standalone Q2 (Jul–Sep 2025) was softer than the H1 headline: net sales ~¥98.02B (+5%), operating income ~¥14.94B (−13%), net income ~¥9.34B (−20%). So the strong H1 was front-loaded into Q1; Q2 shows the legacy/PC drag and pricing caution biting even as AI ramps. Management was explicitly "not optimistic about product prices" in Oct 2025.

Market reaction / what's priced in: the stock is at/near all-time highs (see Lens 8) — the market is paying for the forward AI ramp and the mid-term-plan upgrade, not the trailing print.

Lens 6 · Earnings Calls (sentiment trend)

No transcripts/ on the shelf; this is reconstructed from IR commentary and coverage.

• FY2023 (to Mar-24) tone: defensive — deep substrate downturn, inventory correction, PC/standard-server demand "slower than expected"; the story was "survive the trough".
• FY2024 (to Mar-25) tone: inflecting — AI-server revenue guided to ~3× FY2023; management flagged the demand-recovery pace as still uneven outside AI.
• FY2025 (in progress) tone: confidently expansionary on volume, cautious on price. The ¥500B capex approval (Feb 2026) and the mid-term-plan upgrade (OP target to ¥150B for FY2027, long-term ≥¥300B by FY2030) are the loudest statements. The recurring phrases: customer advance payments, sold-out capacity, ABF material as the constraint. The thing they stopped saying: the old Intel-dependence framing. The thing they added: hedged language on ASP ("not optimistic on prices") — a tell that even a monopolist expects some give-back as the whole industry expands. Net trajectory: defensive → inflecting → expansionary-but-price-cautious.

Lens 7 · Comps

Peer set = the ABF/FC-BGA substrate oligopoly. Multiples are `` with source/date, or n/a. Do not treat these as a like-for-like screen — fiscal calendars, split timing, and cycle position differ, and several names are mid-up-cycle so trailing P/Es are distorted.

Read: the whole peer group trades at nosebleed trailing multiples because earnings are cyclically depressed relative to the AI-driven forward ramp (the classic "high P/E at the bottom of the cycle" setup) — Morgan Stanley upgraded Unimicron and Nan Ya PCB to Overweight arguing the ABF up-cycle "will last through the end of the decade". Ibiden's dividend yield reads anywhere from ~0.13% to ~1.0% across aggregators; the spread is because the price roughly ~7×'d off its 52-week low (¥3,032 → ¥27,480 split-adjusted), collapsing the yield — treat ~1% on trailing DPS as the honest figure and ~0.13% as a stale-price artifact. Payout ratio ~16.6%.

Lens 8 · Stock-Price Catalysts

Price context: ¥23,345 as of 3 Jul 2026, 52-week range ¥3,032–¥27,480 (post-split); a substack cited ¥17,060 on 13-May-2026, so the name ran a further ~37% into early July — the price is moving fast; any target here is date-stamped 2026-07-06.

What moves this stock (the pattern):

• The AI-server substrate narrative — every data-point confirming Ibiden as an irreplaceable GB200/Blackwell supplier. The dominant driver of the ~7× move.
• Capex announcements as demand proxies — the ¥500B (Feb 2026) plan and the 2.5× capacity target were read bullishly (customers pre-funding = de-risked demand), not as balance-sheet strain.
• Mid-term-plan upgrades — hiking the FY2027 OP target from ¥90B to ¥150B, and setting ≥¥300B by FY2030, repriced the equity.
• Hyperscaler capex prints (the $725B CY2026 aggregate) — Ibiden trades as a leveraged proxy for AI-infra spend.
• Downside triggers historically: substrate inventory corrections and PC/standard-server weakness (the FY2023–24 downturn), plus any ASP-erosion signal. The Q2-FY2025 −13% OP wobble shows the legacy drag can still surprise.

Takeaway: the market reacts to forward AI capacity and demand signals, and largely forgives soft trailing prints — a sentiment regime that is powerful on the way up and brutal on the first genuine AI-demand disappointment.

---

Phase C — Judge people & books

Lens 9 · Management

• CEO: Koji Kawashima, President & CEO / Representative Director, appointed June 2024. A professional manager, not a founder — direct ownership 0.028% ($3.19M). Prior CEO Koji Kawashima's predecessor era ran the trough; Kawashima inherits the AI up-cycle.
• Track record / capital allocation: the defining decision on his watch is the ¥500B, 3-year capex program funded ~50% by customer prepayments — a structurally attractive way to finance growth (someone else's balance sheet carries the risk). The mid-term plan explicitly commits to restore ROE to ≥10% (vs 6.8% FY2024) and hold an equity-ratio discipline while balancing growth capex and shareholder returns. This is the crux management bet: can they push ROIC/ROE back above cost of capital as the new fabs fill?
• Skin in the game: low direct insider ownership is normal for a large Japanese industrial and not a red flag by local standards, but it does mean alignment runs through governance and comp, not a founder's stake.
• Shareholder returns: conservative — payout ratio ~16.6%, yield ~1%, 5-yr dividend growth ~+2.7%. No material buyback program surfaced. The company is prioritizing reinvestment over distributions, appropriate for a demand-constrained monopolist but a watch-item if the AI ramp under-delivers.
• Red flags: none material found — no related-party controversy, no promotional behavior; if anything management is understated (openly cautioning on price into a sold-out market). Balance sheet is conservative (50.6% equity ratio).

Lens 10 · Forensic Red Flags

Ground truth: research-layer financials.csv is empty, so this is IR-PDF + web derived; label every figure.

• Earnings vs cash: OCF ¥118,895M in FY2024 comfortably exceeded net income ¥33,704M — earnings are cash-backed, not accrual-flattered. Investing outflow of ¥164,182M (capex-heavy) is the story, financed partly by that OCF and the customer advances.
• Customer advance payments (~¥92.1B): a positive signal for demand, but a genuine accounting watch-item — advances are a liability and, per the substack, "ROIC may understate true returns if customer prepayments fund assets off-balance-sheet". The honest read cuts both ways: prepayments de-risk demand but can flatter reported capital efficiency and must unwind as product ships. Track the advance-payment balance every quarter.
• Depreciation ramp: ¥44.5B of new depreciation is hitting Electronics as fabs come online; the segment expanded margin through it — but if AI volume slips, that fixed D&A becomes brutal operating deleverage. Non-GAAP flattering is not the risk here (Japanese GAAP, no big SBC add-back culture); the risk is capacity-cost timing vs demand.
• Inventory / receivables: no anomaly sourced, but the FY2023–24 downturn was itself an inventory-correction event — cyclicality is the real "red flag," not fraud.
• Segment reporting: clean and granular in the IR PDFs; no obfuscation — if anything the blended group margin hides Electronics strength.

Regulatory findings (required). Per regulatory/regulatory-findings.md (generated 2026-07-06): Ibiden has no SEC CIK — no EDGAR enforcement search possible; total_sec_findings: 0. Non-SEC web search ("Ibiden" (FTC OR DOJ OR antitrust OR consent decree OR settlement OR fine OR penalty OR lawsuit) enforcement, 2024–2026) returned no material actions naming Ibiden. No 10-K Item 3 exists (foreign filer). Verdict: No material regulatory or legal findings — verified via the regulatory-findings file (SEC EDGAR EFTS LR/AAER, 0 findings, CIK-less) and a non-SEC web search as of 2026-07-06. (Caveat: Japanese-language JFTC/regulatory sources were not deep-searched; flag for a native-language pass before any concentrated position.)

---

Phase D — Project & stress-test

Lens 11 · Forward Projection

Anchors (company's own, ``):

• FY2025 (to Mar-26) guidance: sales ¥420,000M, OP ¥61,000M, net income ¥37,000M, EPS ¥132.50 post-split (¥265 pre-split).
• Mid-term plan: FY2026 (to Mar-27) OP ¥90B on ~¥500B sales; FY2027 (to Mar-28) OP ¥150B; long-term ≥¥300B OP by FY2030 (to Mar-31).
• Consensus next-FY EPS ~¥224; aggregator "FY2026" read ¥416.2B sales / ¥63.7B earnings (+89%) closely matches the company forecast.

Three-year EPS path (fiscal years ending Mar-2026 / Mar-2027 / Mar-2028; all post-split share count ~281M implied from ¥37,000M ÷ ¥132.50; every line `` unless tagged):

• FY2025e (to Mar-26): EPS ≈ ¥132–133 — take the company guide (¥37.0B NI) ``.
• Base — FY2026e (to Mar-27): EPS ≈ ¥180–200. OP ¥90B ; assume ordinary→net conversion ~62% (in line with FY2024's ¥47.6B OP → ¥33.7B NI) → NI ≈ ¥55–58B → EPS ≈ ¥196–206 .
• Base — FY2027e (to Mar-28): EPS ≈ ¥300–330. OP ¥150B × ~0.62 → NI ≈ ¥93B → EPS ≈ ¥330. This is the mid-term-plan target year and the number the bulls underwrite.
• Bull: Electronics holds >22% margin and ABF allocation clears; FY2027 OP overshoots to ¥170–180B → EPS ¥370–400 . Aligns loosely with the substack's ¥340–470 FY2029 EPS bull/base range .
• Bear: ASP erosion + a PC/standard-server air-pocket + ABF supply cap; FY2027 OP lands ¥100–110B → EPS ¥210–240 — roughly the substack's bear (¥200–210).

Honest caveat: these lean on the OP→NI conversion ratio and a post-split share count derived from guidance, not disclosed line items — treat as directional. The valuation question (Lens 12) matters more than the third-decimal of EPS.

Forecast log: skipped per --watchlist rules (no forecast.ts create in the unattended loop). If promoted, the natural Brier line is: "4062.T FY2027 (to Mar-2028) operating profit ≥ ¥150B", p≈0.55, resolves 2028-03-31.

Lens 12 · Bull vs Bear

Bull case. Ibiden is the pick-and-shovel monopolist of the AI build-out at the one layer nobody can substitute on a 2–3 year horizon. 70–80% of the hardest substrate tier, customers prepaying to reserve capacity, a 2.5×-capacity ¥500B program de-risked by that prepayment, Electronics margins inflecting to 22%+, and a mid-term plan tripling operating profit (¥47.6B FY2024 → ¥150B FY2027 → ≥¥300B FY2030). If the market re-rates it from "cyclical Japanese substrate supplier" to "scarce, customer-reserved AI-infra toll-taker," the multiple and the earnings both climb — the classic double-driver. Secular tailwind: hyperscaler capex +77% to $725B, GPU packages getting harder (Blackwell → Rubin → beyond), which is exactly the direction that widens Ibiden's edge.

Bear case (2–3 permanent-impairment risks).

1. The valuation already is the thesis. ~67× trailing, ~34× EV/EBITDA, ~7× off the low, ~1% yield — the monopoly is known and owned. You are not buying a hidden gem; you are paying a premium multiple for a cyclical whose "up-cycle through the decade" is consensus (MS Overweight on the peers). A single quarter of AI-demand disappointment de-rates a 34× EV/EBITDA name violently.
2. ASP give-back in a capacity race. Everyone is expanding — AT&S (EU-funded), Samsung Electro-Mechanics, Unimicron (record capex), Nan Ya. Management itself is "not optimistic on prices." If the whole industry adds capacity into a demand plateau, the sold-out premium erodes and the fixed-cost D&A ramp (¥44.5B new) turns into operating deleverage.
3. The ABF ceiling and the glass horizon. Growth can be capped upstream by Ajinomoto's film allocation regardless of Ibiden's fabs; and downstream, glass-core substrates (TSMC CoPoS, Intel panel-level) are the multi-year replacement for organic ABF as packages outgrow warpage limits.

Pre-mortem (18 months out, thesis broken): Blackwell/Rubin ramp timing slips a couple of quarters, hyperscaler capex growth decelerates from +77% to +20%, and simultaneously the industry's new substrate capacity lands — Ibiden reports a sold-out-but-lower-ASP quarter, Electronics margin ticks back toward 18%, and a 34× EV/EBITDA multiple compresses to ~22×. The stock halves from a high base while the business is still fine. That is the realistic bear — a de-rating, not a bankruptcy.

Contrarian view (what the market refuses to see): the bulls treat Ibiden as the bottleneck, but the true chokepoint is one link up (Ajinomoto's ABF, 95%+) and one link down (TSMC CoWoS) — Ibiden is the most investable node but not the scarcest one. And the glass-substrate transition, which bears frame as an existential threat, is one Ibiden is a validation partner in (TSMC CoPoS, with Innolux) — so the disruption is more likely to migrate Ibiden's franchise than to kill it. The market is arguably mis-pricing which risk matters: not "Ibiden gets disrupted," but "Ibiden's multiple is a leveraged bet on hyperscaler capex timing."

Lens 13 · Devil's Advocate (short-seller)

You are short 4062.T. The case:

• You're paying ~34× EV/EBITDA and ~67× earnings for a company whose own CEO won't defend pricing. The entire thesis is a forward ramp already in consensus (MS Overweight the peers) — there is no variant view left to be right about, only execution to disappoint.
• Revenue concentration is brutal and shifting. After 25 years at 70–80% Intel, the base is now Nvidia-centric. Nvidia's roadmap dictates Ibiden's volume, packaging format (a shift in CoWoS-L intensity, or a move toward glass sooner), and timing. A Blackwell/Rubin air-pocket — of which the industry has already seen packaging-driven Blackwell delays — flows straight to Ibiden's P&L with a ~¥44.5B fixed-D&A anchor attached.
• The moat is real but the incremental moat is thinner than share implies. Ibiden dominates the hardest tier, but the ¥500B capex is buying more of the same into a market where four other players are also adding capacity with government subsidies. Sold-out today ≠ sold-out at price in 2028.
• The prepayment structure cuts both ways. ¥92.1B of advances flatters ROIC optics and locks Ibiden into building capacity on customers' schedules; if those customers over-ordered (double-booking across Ibiden and Unimicron, as buyers hedge supply), the unwind is ugly.
• The glass transition is a real terminal-value question. Even as a CoPoS partner, a shift from organic ABF to glass panels changes the competitive set and the capital base — a decade of ABF-fab investment could face accelerated obsolescence risk if glass ramps faster than the 2028 base case.
• What breaks it permanently? Not one thing — a combination: AI-capex growth normalizes + industry capacity floods + ASP falls 15–20% + glass pulls forward. Each alone is survivable; together they take a 34× multiple to ~18× and the earnings down 20–30%, a >50% drawdown from a euphoric base. Plausibility: moderate on a 2–3 year view, low on a 12-month view.

Lens 14 · Management Questions (ordered by information value)

1. Of the ¥500B expansion capex, exactly what share is covered by binding customer advance payments vs. take-or-pay volume commitments — and what is the contractual recourse if a customer's AI ramp slips?
2. What is your committed ABF-film allocation from Ajinomoto through FY2028, and is film supply — not fab capacity — the true ceiling on your AI-substrate output?
3. Give the unblended Electronics economics: AI-server substrate ASP, gross margin, and volume trajectory separately from PC/standard-server — how much of the 22% segment margin is AI vs. legacy?
4. You said you are "not optimistic on prices." Quantify: what ASP trajectory is baked into the ¥150B FY2027 and ≥¥300B FY2030 operating-profit targets?
5. How do you get ROE from 6.8% back above the ≥10% cost-of-capital target while carrying the heaviest capex program in your history — what's the ROIC hurdle on the new Kawama/Ono cells?
6. On glass substrates (TSMC CoPoS partnership): is glass a threat that strands your ABF fab base, or a migration you lead — and what's your capex split between organic and glass through 2030?
7. What is customer concentration today by revenue (Nvidia / Intel / AMD / others), and how do you manage the risk that buyers are double-booking capacity across you and Unimicron?
8. What utilization are the Ono and Kawama AI cells running at now, and what utilization do the FY2027 targets assume?
9. How much of the ¥44.5B new depreciation is already in the run-rate, and what group operating margin do you reach at full utilization of the funded capacity?
10. What is your contingency if hyperscaler capex growth decelerates from +77% to, say, +20% in CY2027 — how fast can you flex capex and protect margin?
11. Ceramics (DPF) is structurally declining on the EV shift — will you divest, harvest, or repurpose those assets, and what's the capital tied up there?
12. What differentiates your 20+-layer yield from Unimicron and Shinko such that customers pay to reserve your capacity specifically — and how durable is that gap as they invest?
13. Post the 2:1 split, what is the shareholder-return framework — will the ~16% payout ratio rise, or is all excess cash committed to growth capex through 2030?
14. What are the two or three technical risks (warpage, layer count, power delivery beyond 1kW) that could force a packaging-architecture change and reset the competitive field?
15. Where do you expect to be underestimating demand or overestimating your competitive position three years out — what keeps you up at night?

---