日本マイクロニクス / Micronics Japan (6871)

Signal panel

Investment thesis

1. Momentum is accelerating, not peaking yet. Standalone OP YoY went from +29.6% (Q3 FY25) to +33.4%, +97.7% and +110.3% (Q2 FY26). OPM went from 21.6% to 35.1% over the same four quarters (fact sheet quarters; Q2 deck p.7, printed 6). Reported fact.
2. The guidance is conservative. The Q3 cumulative guide (¥74.2bn sales, ¥22.4bn OP) implies Q3 standalone sales of ¥25.0bn and OP of ¥6.8bn, against ¥28.3bn and ¥9.9bn in Q2 (短信 Q2 p.1; Q2 deck p.14, printed 13). The full-year guide implies H2 OP of ¥15.8bn, flat on H1's ¥15.6bn, even though the deck says orders "rose sharply QoQ, centred on HBM" (deck p.6, printed 5). The forecast assumes ¥156/USD. Company guidance and inference.
3. Capacity is the constraint, and it is being added. Capex was ¥15.3bn in FY25 and is planned at ¥21.0bn for FY26, mostly for the new Aomori building (deck p.16, printed 15). Q2 sales benefited from that added capacity and from shipments brought forward ("売上げ前倒し", deck p.6). Reported fact plus company guidance.
4. The sell-off is about the 2028+ cycle, not this year's orders. On 2026-10-15, SK hynix was down about 26% and Samsung about 22% over five days, on CXMT's 4F2 DRAM news and a columnist's 2029 oversupply call (KuCoin flash, 2026-10-15). Goldman sees CXMT's DRAM capacity rising 77% by end-2028, but calls the market's read of the technology overdone (TradingKey, 2026-10-16). MJC's revenue depends on new device designs and wafer starts at Samsung (49.9% of FY25 sales) and Micron (21.4%) (有報 FY25 p.25), not on CXMT. External estimate and inference.
5. The valuation is below its own band and below peers despite higher growth. Forward PER (consensus) is 17.74 with z −1.03. On FY27 consensus EPS of about ¥794 the multiple is 16.9x. Advantest trades at 37.9x and Tokyo Seimitsu at 23.8x on company forecasts (Yahoo! Finance JP, 2026-10-16). Reported market data.

Business and earnings

MJC makes probe cards: consumable, design-specific interfaces that contact every die during wafer test. In Q2 FY26, memory probe cards were ¥26.5bn of the segment's ¥27.9bn sales (95%). Non-memory was ¥1.4bn and has been flat for four quarters (deck p.9, printed 8). The U-Probe MEMS cantilever card with its "crescent" layout for DRAM is, by the company's account, the industry standard (deck p.20). The company states its target as keeping "overwhelming No.1 share" in memory probe cards (有報 p.9). Treat that as a company claim; no third-party share figure was checked.

• Profit engine. Probe cards made a ¥18.3bn segment profit in H1 FY26 at a 37.7% margin. TE (testers, probers, sockets) lost ¥0.8bn on ¥0.7bn of sales (短信 Q2 p.11, printed 9). TE is a loss-making option on the Testalio tester and Excelyze prober, both still in customer evaluation (deck p.6).
• Drivers of the H1 jump. DRAM volume rose on new capacity, and the mix moved toward "DRAM high-value products" (短信 p.4, printed 2). The company does not quantify the HBM share. Unknown: HBM revenue, so the HBM theme is supported only at the level of "DRAM/memory = 95% of probe card sales".
• Customers and regions. Samsung 49.9% and Micron Memory Taiwan 21.4% in FY25 (有報 p.25). Korea is 53.2% of sales and Taiwan 30.9% (fact sheet). SK hynix is not a major customer, so MJC's HBM exposure runs mainly through Samsung's HBM ramp and Micron. That is inference from the customer list.
• One-offs. H1 includes a ¥1.0bn subsidy gain below OP (短信 p.8, printed 6) and a ¥0.69bn FX gain. They have no effect on OP. Without the subsidy, the FY NP guide of ¥23.0bn falls to about ¥22.3bn and EPS to about ¥575 from ¥593 (inference, 30% tax assumed).
• Cash conversion. OCF/NP is 1.07–1.71 over FY22–25. H1 FY26 OCF was ¥12.2bn against NP of ¥11.5bn (短信 p.10, printed 8).

Forecast test

• Guidance practice changed. In Feb 2026 MJC stopped giving a full-year sales number. It gave only an H1 guide and the qualitative statement that FY26 growth would exceed 26% (FY deck p.17, printed 16). The first full-year figure came on 2026-08-12, so the fact sheet's "initial" FY26 guide is August's. The H1 guide moved from OP ¥12.3bn (Feb) to ¥12.9bn (May), and H1 delivered ¥15.6bn (+26% vs Feb).
• Track record. Actual vs initial OP: FY21 +3.0%, FY22 −11.3%, FY23 +56.2%, FY24 +8.4%, FY25 +19.9% (fact sheet). The two misses (FY20, FY22) both came as a memory cycle turned. The beat habit is partly priced in by the consensus being above guidance.
• Progress. It is in line with the seasonal norm, not ahead. The upside case rests on the implied QoQ drop in Q3 rather than on progress.
• Orders. Probe card orders were ¥28.3bn in Q4 FY25, a record, against sales of ¥19.4bn, a book-to-bill of 1.46, with backlog at ¥34.5bn (FY deck p.9, printed 8). The Q2 FY26 deck gives no figures. Unknown, and it matters most.
• Revision probability: Medium. A Q3 OP of ¥8bn or more (vs the ¥6.8bn implied guide) on 2026-11-12, with a raised FY guide, would confirm the thesis.

Financial strength and capital allocation

• Balance sheet. At 2026-06-30: cash ¥20.0bn, debt ¥6.4bn, net cash ¥13.6bn, equity ratio 65.8%, current ratio 1.78 (短信 p.6–7, printed 4–5). No liquidity risk.
• Hermes stake. Investment securities rose ¥25.7bn to ¥36.4bn in H1 with no purchases. The cause is a mark-to-market gain on the 1.5m-share stake in Hermes Testing Solutions (Taiwan), bought in FY25 for business ties (有報 p.71). It is carried at roughly ¥34bn (inference: total less the other holdings' ¥1.9bn), about ¥890 per share pre-tax, or 6% of market cap. It inflates equity (OCI +¥17.7bn) and so depresses reported ROE and lifts PBR. It is volatile and not operating value.
• Cash flow. FY25 FCF was −¥8.8bn on ¥15.3bn of capex. H1 FY26 FCF was +¥4.7bn. The FY26 capex plan of ¥21.0bn leaves about ¥12.1bn for H2 (deck p.16). It is self-funded at current OCF.
• Working capital. Inventory grew 39% in FY25 against sales growth of 26% (amber flag), with inventory days at 132. In H1 FY26 inventory rose ¥2.3bn (+17%) while H1 sales rose 49%, so the flag is easing.
• Capital efficiency. ROE was 20.9% in FY25: net margin 17.2% × asset turnover 0.71 × leverage 1.50. IFIS ROIC is 16.4%. The forecast ROE is 25.2% (IFIS), above the FV26 target of 23% (有報 p.9). This is well above an assumed 10% cost of equity for a high-beta semiconductor stock (analyst assumption).
• Returns. Payout has been about 30% every year from FY21 to FY25 (30.2–30.9%). DPS of ¥178 is guided for FY26 (+87%). There are no buybacks since 2018. Capital goes to capacity, which is the right call while orders exceed shipments.
• Gates. All pass. OCF has been positive every year. There is no going-concern note, and 修正再表示 is 無 (短信 Q2 p.2). EDINET shows corrected 有報 for FY20–FY23, filed 2025-02-27 with a 有価証券届出書. No restatement of the financials is shown in the IFIS data, and the 短信 shows none, but the stated reason for the correction was not verified (see limitations). The HBM theme is backed by memory sales figures, not keywords.

IR and management credibility

• FV26 (2024-11 upgraded targets for FY26). Sales ¥80bn, OP ¥20bn, OPM 25%, ROE 23% (有報 p.9). The FY26 guide of ¥103.8bn / ¥31.4bn / 30.3% beats every target by a wide margin. Management has delivered more than it promised.
• Disclosure regressions. The full-year sales guide was replaced by a qualitative outlook (2026-02-13), and the quarterly order/backlog table is absent from the Q2 FY26 deck. Both reduce visibility just as the cycle debate sharpens. The new medium-term plan that replaces FV26 (expected with FY26 results, inference) should be judged partly on whether it restores order disclosure.
• Explanation quality. It is adequate: the deck explains mix, the pull-forward and segment drivers in plain terms.

Valuation and scenarios

Basis: FY27 EPS. Shares outstanding 38.77m (短信 p.2).

At a 25/50/25 weighting the probability-weighted value is about ¥15,100 (+12%), before the dividend (1.3%). That is a modest edge with a wide range, which is why confidence is Medium.

• Bands. PER (1y fwd consensus) is 17.74 against a mean of 21.2 (z −1.03). PBR is 5.71 against a mean of 6.73 (z −0.83). PBR is overstated by the Hermes OCI. EV/EBITDA is 14.3x on forecast.
• Peers (2026-10-16, Yahoo! Finance JP, company-forecast PER / actual PBR). Advantest 37.9x / 24.0x. Tokyo Seimitsu 23.8x / 3.85x. FormFactor's forward PER is about 27–30x (GuruFocus 2026-09-09, Yahoo US 2026-09-04, non-GAAP basis, not like-for-like). MJC is on 22.7x company forecast. Peer ROE and OPM were not gathered.
• What the price implies. At a band-mean 21.2x, ¥13,450 needs only about ¥634 of EPS. That is about 7% above the FY26 guide and 20% below FY27 consensus. The market is pricing roughly flat earnings after FY26, so it assumes the cycle peaks next year.
• Why the market might be right. OPM of 35% is a record and includes a pull-forward. Two customers are 71% of sales. Memory capex and design-start cycles have hit MJC hard before (FY23 OP −42%). The target-price signal is unreliable: the IFIS consensus target of ¥27,850 comes from two analysts, while the one dated change is a US broker cutting to ¥16,000 on 2026-09-24 (report_topix).

Bull case

• Q3 beats the ¥6.8bn implied OP guide by a wide margin, and the FY guide rises toward ¥34–35bn (consensus high ¥34.5bn).
• HBM4/HBM4E design starts at Samsung and Micron keep probe-card demand ahead of capacity into 2027, and FY27 OP reaches ¥45bn or more.
• The sector fear fades, and the multiple returns to its 21x mean, giving about ¥16,800 on FY27 consensus.

Bear case

• The Q2 pull-forward means Q3 really is down, and the absent order table hides a lower book-to-bill.
• Memory makers cut 2027 capex, or Samsung loses HBM share, and FY27 OP stalls near ¥31bn. The stock then derates to about 15x, near ¥8,900.
• A customer insources or dual-sources (Samsung is half of sales). Yen strength below ¥145/USD cuts margin (the guide assumes ¥156).

Catalysts

What would change the rating

• Upgrade (higher conviction) if Q3 OP is ¥8bn or more, the FY guide is raised, and order figures return with a book-to-bill above 1.
• Downgrade to Hold if Q3 OP comes in at or below the ¥6.8bn implied guide with no FY raise, or if FY27 consensus OP falls below ¥38bn.
• Thesis invalidated if MJC cuts FY26 guidance, or if Samsung or Micron announce probe-card dual-sourcing or 2027 DRAM capex cuts.

Monitoring

Scorecard

Fact-sheet flags

Computed by factsheet.mjs. Each flag is a prompt to investigate, not a conclusion.

• ▲amberinventory-outpacing-sales inventory grew 39.0% in 2025/12 (9,489 to 13,189) against sales growth of 26.1%
• iinfonet-cash Net cash of 12,755 in 2025/12 (BS_NET_DEBT -12,755)
• ▲amberfcf-negative Free cash flow -8,789 in 2025/12 (OCF 12,903, ICF -21,692, capex 15,252)
• iinfocapex-heavy Capex 15,252 is 2.78x depreciation 5,494 in 2025/12
• iinfoconsensus-rising 2026/12 consensus OP changed +8.6% over 3 months to 32,400
• iinfoper-cheap Forward PER 17.74 vs 1-year mean 21.2 (+1σ 24.55), z -1.03: cheap vs own history
• iinfofar-below-52w-high Price 13,450 is -26.0% from the 52-week high 18,180

Sources and limitations

• filings/2025-12_FY_短信_20260213.pdf, filings/2025-12_FY_説明会_20260213.pdf (p.9 orders, p.15 H1 guide, p.17 guidance policy), 2026-02-13.
• filings/2025-12_FY_有報_20260325.pdf (p.9 FV26 targets, p.25 major customers, p.70–71 stockholdings), 2026-03-25.
• filings/2026-12_Q2_短信_20260812.pdf (p.1–2 results and guidance, p.4–5 commentary, p.6–11 statements and segments), 2026-08-12.
• filings/2026-12_Q2_説明会_20260812.pdf (p.4–9 results and mix, p.14–16 guidance and capex), 2026-08-12. filings/2026-12_Q2_半期_20260813.pdf, 2026-08-13.
• IFIS snapshot data/6871/2026-10-17/ and fact sheet data/6871/analysis/2026-10-17/factsheet.md. pages/report_est.html and pages/report_topix.html (broker TP change 2026-09-24).
• Memory sell-off: KuCoin flash, 2026-10-15; TradingKey, 2026-10-16. Secondary news, used as leads only.
• Peers: Yahoo! Finance JP 6857, 7729 (2026-10-16); GuruFocus FORM forward PE (2026-09-09).
• Q3 results date: 2026-11-12 per Kabutan via search summary. Not confirmed on TDnet.
• Limitations.
• Q1 FY26 短信 and deck were not in the snapshot, so it is unknown whether Q1 still disclosed orders.
• The reason for the 2025-02-27 corrected 有報 (FY20–FY23) was not found. It was filed with a 有価証券届出書, and no financial restatement is shown, but it is unverified.
• The HBM revenue share is not disclosed. Peer ROE and OPM were not gathered, and FormFactor's PER basis differs.
• No disclosure after 2026-08-12 was found except a land purchase (2026-07-31) and a Nomura holding-change report (2026-10-05). TDnet was not checked directly.
• This is research, not individualised investment advice.