Micronics Japan (TSE: 6871) — Deep Dive
Status: WATCH at current price ¥13,360. Earnings 2026-05-13 (+16d). Probe-card pure-play, Japanese MJC, niche but established. Stock has tripled in 12 months and trades ~35% above analyst PT mean (¥9,900). Quality is real (48% gross margin, net cash, +26% rev growth, ROIC well above WACC) but valuation has caught the AI-test re-rating. Re-engage post-print on a base build to ¥10,500-11,500.
1. Executive Summary
Thesis: Micronics Japan ("MJC") is a top-5 global probe card maker, specialized in vertical and MEMS probe cards for memory (DRAM/HBM) and advanced logic test. Probe cards are the consumable interface between automatic test equipment (Advantest/Teradyne) and the wafer — every die that ships is touched by a probe card, and each probe card has a finite life. As HBM testing complexity scales (longer probe contact times, more pins, KGD-grade test) and Intel's 18A / chiplet ramp drives a step-function in test intensity, probe-card consumption per chip is rising structurally — see semi-probe-card-primer and burn-in-test-primer.
Conviction: Medium on the thesis, Low at current price. Medium-High at <¥10,500 (-21%) on a post-print pullback or cycle pause.
| Metric | Value |
|---|---|
| Price | ¥13,360 (2026-04-27) |
| 52w range | ¥2,922 – ¥13,750 (2.8% below high) |
| Market cap | ¥518B (~$3.5B USD) |
| EV | ¥474B (net cash ¥12B) |
| Trailing P/E | 43× |
| Forward P/E | 61× |
| P/Sales (TTM) | 7.4× |
| P/Book | 7.8× |
| Gross margin (TTM) | 48.2% |
| Op margin (TTM) | 19.6% |
| Revenue growth (TTM) | +25.8% |
| Net cash | ~¥12B |
| Insider / Inst ownership | 18.5% / 42.8% |
| Beta | 1.47 |
| Analyst PT mean / high / low | ¥9,900 / ¥14,000 / ¥5,800 (mean is 26% BELOW current) |
| Coverage | Thin (3-4 analysts; recommendation key: "none") |
| Next earnings | 2026-05-13 (+16d) |
Targets: ¥13,000 base / ¥17,000 bull / ¥6,500 bear → risk/reward at current price is symmetric, not asymmetric to the upside.
2. Corporate Overview
Full name: Micronics Japan Co., Ltd. HQ: Musashino, Tokyo. Founded: 1970. Listed: TSE Prime (originally TSE 1st Section). Family/founder lineage but professionalized management.
What they do
Designs and manufactures probe cards — the precision consumable that connects automatic test equipment (ATE) to a semiconductor wafer during electrical test. Every chip is tested at wafer level (sort) and final test (FT); probe cards are the contact interface. Each card has a finite touchdown life (millions of contacts) before pads wear out. Pure-play probe-card maker — not diversified into other equipment categories.
Business lines (estimated mix from prior IR)
- Memory probe cards (DRAM, HBM, NAND) ~50-55% — historically dominant, Korean memory makers (Samsung, SK Hynix, Micron) major customers
- Logic probe cards (CPU/GPU/SoC) ~25-30% — TSMC, Intel, Samsung Foundry advanced node testing
- Specialty (image sensor, MEMS, automotive) ~10-15% — niche but high-mix
Business model
Capital-light specialty hardware. Probe cards are consumables with replenishment cycles tied to (a) wafer volume, (b) chip complexity (more pins / longer touchdowns), and (c) node transitions (each new node requires new card design + qualification). High switching costs once qualified — every die has a 2-4 month qualification cycle for a new probe card supplier. Margins reflect this: 48% GM, 19.6% op margin is best-in-class for semi consumables.
Geographic mix
- Korea: largest single market, driven by Samsung/SK Hynix HBM/DRAM
- Japan: domestic Kioxia, Renesas, Sony image sensors
- Taiwan: TSMC advanced logic, foundry probe
- US: Intel, Micron
- China: SMIC (where allowed), domestic memory (YMTC where allowed under export controls)
Latest IR materials
TSE Prime listed; quarterly earnings with English supplements at https://www.mjc.co.jp/eng/ir/. FY ends December (calendar year). Most recent: FY2025 full-year results March 2026; Q1 FY2026 expected 2026-05-13.
Assets & operations
- Domestic manufacturing in Japan (Musashino HQ + production)
- No major fab capex — probe cards are precision assembly + MEMS sub-component, not wafer fab
- Asset-light: PP&E modest, working capital cyclical with NAND/HBM order book
Joint ventures / partnerships
None publicly material. Standalone operator. Customer relationships are deep but not equity-linked.
3. First Principles — Probe Cards
See semi-probe-card-primer for full sector. This section summarizes what's specifically relevant to MJC.
The problem
After a wafer is fabricated, every die must be tested before it ships (or before being assembled into a package). The test runs at wafer level (Wafer Sort) — the wafer rests on a chuck, and a probe card descends, makes electrical contact with every die simultaneously (or in groups), and the ATE runs functional + parametric tests. Probe cards are the only thing in the test flow that physically contacts every die. They are the single point of failure for test throughput.
Probe card categories
- Cantilever — older tech, low pin count, simple sub-100 pin chips. Phasing out for advanced nodes.
- Vertical — high pin count, supports modern logic + DRAM. Micronics core strength.
- MEMS — silicon-fabricated micro-needles, fine-pitch (sub-50µm), used for advanced logic + HBM. Highest-margin, hardest-to-make. Micronics + FormFactor + Technoprobe + JEM dominate.
- Vertical probe array (VPA) — for ultra-high-pin-count CPUs/GPUs.
Key technical metrics that matter for MJC
- Pitch (distance between probe needles) — finer pitch = more advanced node compatibility
- Pin count — modern AI accelerators have 5,000-10,000+ probe points
- Touchdown life — how many contacts before card needs refurb/replace (consumable cycle = revenue!)
- Parallelism — number of die tested simultaneously (matters for cost-per-die test economics)
- HBM-specific: through-stack vertical probe + Known-Good-Die (KGD) test capability — this is the AI growth driver
Why HBM matters for MJC
HBM stacks 8-16 DRAM dies vertically with TSV interconnect. Each die in the stack must test as KGD before stacking — bad die in a stack means whole stack discarded. KGD test = more probe touchdowns per die, more touchdowns per stack, higher probe card consumption rate. SK Hynix's HBM4 ramp (and Samsung/Micron catching up) is a structural tailwind for memory probe card demand. JEM (6855.T) and Micronics are the two primary Japanese beneficiaries; Korean and Taiwanese alternatives exist but Japanese probe cards are entrenched at Samsung/SKH.
4. Product / Segment Read
| Segment | What | Customers | % Rev (est.) | Trend |
|---|---|---|---|---|
| Memory (DRAM, HBM, NAND) | Vertical + MEMS probe cards | Samsung, SK Hynix, Micron, Kioxia | ~50-55% | HBM driving acceleration |
| Logic (CPU/GPU/SoC) | Vertical + MEMS for advanced nodes | TSMC, Intel, Samsung Foundry | ~25-30% | Advanced-node tailwind |
| Specialty (image sensor, MEMS, auto) | Application-specific | Sony, Renesas, automotive Tier-1 | ~10-15% | Stable |
| Service / refurb | Consumable replacement, refurb | All customers | ~5-10% | Recurring |
Swing variable: HBM probe card demand. SK Hynix HBM3E/HBM4 ramp is the single biggest driver of memory probe card growth in 2024-26. If HBM volumes hit consensus (~$30B by 2026), MJC + JEM see meaningful pull. If HBM digestion happens before 2027, probe replenishment cycles slow.
5. Value Chain Position
[Wafer fab] → [Wafer test (sort)] ← [Probe card] ← [MJC, JEM, FormFactor, Technoprobe, MPI] → [Final test] → [Package]
★ MJC supplies probe cards + service
Suppliers (upstream): MEMS substrate fab (specialty foundries), tungsten/rhenium probe needles (specialty alloys, Chinese supply chain risk), ceramic substrates (Kyocera, NGK), space transformer PCBs (Ibiden, Kyocera). No 1-2 small-cap upstream bottleneck specific to MJC.
Customers (downstream): memory makers (Samsung, SK Hynix, Micron, Kioxia), foundries (TSMC, Intel, Samsung Foundry), IDMs (Renesas, Sony, Infineon, Microchip). Top customer concentration likely Samsung/SK Hynix combined ~30-40% of revenue.
Pricing power: Moderate. Once qualified at a customer for a specific chip, MJC has ~2-4 month switching cost. But customers run RFP cycles for new chips and dual-source where possible. Pricing reflects technical complexity, not pure scarcity — gross margin 48% is the proof.
Upstream bottleneck check: No, MJC is itself the bottleneck for its customers, not bottlenecked by a smaller supplier.
5b. Key Customers
- Samsung Electronics — memory + foundry, likely top-1 or top-2 customer
- SK Hynix — DRAM/HBM, key HBM growth driver
- Micron — DRAM
- Kioxia — NAND (note: same parent ecosystem as Pink's other research)
- TSMC, Intel, Samsung Foundry — logic/foundry advanced-node testing
Concentration risk: undisclosed but industry-typical 30-40% top-2.
6. Why It Matters — TAM & End Markets
- Global probe card market: ~$3-4B 2025; projected $5-6B by 2028 driven by HBM, advanced logic node count, AI accelerator complexity (TechInsights, Yole estimates).
- MJC SAM share: ~5-8% global probe card revenue, top-5 globally, #2 in Japan after JEM (6855.T which Pink holds).
- Tailwinds: HBM stacking; chiplet test complexity (more pins per package); Intel 18A / advanced node probe re-qualifications; AI accelerator pin-count escalation.
- Headwinds: Probe card downturn in client memory cycles (NAND consumer); FormFactor / Technoprobe gaining advanced-logic share at TSMC.
6b. Sector Inflection — Why Now?
Demand: HBM ramp (SK Hynix HBM3E in 2025, HBM4 in 2026) + chiplet test complexity (Intel EMIB, AMD MI300 series, Apple SoIC, Nvidia CoWoS-L) = step-function in probe card demand. Each chiplet/stack requires multiple probe touchdowns; pin counts on a single AI accelerator probe card now exceed 10,000.
Supply: Probe card capacity is labor-constrained (precision assembly + MEMS fabrication) and engineering-constrained (each new chip = new card design). Lead times in 2025 reportedly extended to 6+ months for advanced cards (per Collyer Bridge / Discord intelligence on MPI Corp).
Coming shortage / glut: Industry consensus = continued tightness through 2026 driven by HBM, then potential normalization in 2027 if HBM digestion + advanced-logic push-out.
Structural change: Probe card per-chip consumption is structurally rising — chiplet test, KGD demands, advanced-node parametric test all add probe touchdowns per shipped die.
Why now (concise): AI accelerators + HBM are driving probe-card consumption per AI chip 2-3× higher than equivalent prior-gen logic. MJC + JEM are the two Japanese pure-plays best positioned. Stock has tripled in 12 months on this thesis. The "now" question is whether the 4-5× re-rating already prices the cycle or there is more room — leaning latter only if HBM4 ramp surprises to the upside in 2026-27.
7. People & Governance
Limited public disclosure under JGAAP compared to S&P companies. Key structural notes:
- Insider ownership 18.5% — material, suggests founder/family lineage still meaningful
- Institutional 42.8% — diffuse Vanguard/Fidelity/iShares small-cap Japan funds; no activist holder of size
- Top non-strategic holders: Vanguard International 1.14%, College Retirement 0.78%, Vanguard Developed Markets 0.74%, iShares EAFE 0.50% — passive flow only
- No insider transactions reported in yfinance Form 4-equivalent feed
Compensation alignment: TSE Prime PSU/RSU standard; specific hurdle structure not pulled. Flag for follow-up via /filings if Pink wants to size larger.
Capital allocation: Net cash ¥12B on ¥518B mcap = 2.3% of EV — modest, not Japanese-mega-cash style. FCF was -¥13.5B TTM (negative!) per yfinance — likely working capital build into HBM ramp. Operating CF +¥12.9B — so the negative FCF is capex/WC heavy in the ramp, not a profitability concern.
Mgmt DD verdict: Not Yellow, Not Green — neutral on disclosure. No red flags surfaced; insufficient depth to assign confidence rating.
8. Competitive Landscape — Probe Cards
| Company | Ticker | Country | MC ($B) | Strength | Notes |
|---|---|---|---|---|---|
| FormFactor | FORM | US | ~$3.5B | Logic probe, MEMS, broadest portfolio | Largest pure-play; Cascade Microtech acq |
| Technoprobe | TPRO.MI | Italy | ~$3.2B | MEMS, advanced logic at TSMC | #1 at TSMC advanced node |
| JEM (Japan Electronic Materials) | 6855.T | Japan | ~$2.5B | Memory probe (HBM/DRAM), Pink HOLDS | #1 Japan, HBM-leveraged |
| Micronics Japan | 6871.T | Japan | ~$3.5B | Memory + logic, vertical + MEMS | #2 Japan; valuation now extended |
| MPI Corp | 6223.TW | Taiwan | ~$1.5B | Probe + RF probe systems | Growing share, AI-driven |
| MJ Probe (Korean) | private | Korea | n/a | Korean domestic memory probe | Captive Samsung/SKH supplier |
Read:
- Probe card industry is more fragmented than NAND but consolidating
- FormFactor + Technoprobe + JEM + MJC + MPI = ~80% of global advanced probe card revenue
- MJC's ~$3.5B market cap now LARGER than FormFactor (the global #1) and Technoprobe — this is a flag: either MJC is catching up rapidly on share or it's overvalued vs peers
- JEM (Pink's hold) at $2.5B with similar HBM leverage looks structurally cheaper
Moat:
- Customer qualification cycles (2-4 months per new chip) = sticky
- MEMS fab IP + 50+ years probe technology = moderate technical barrier
- Not network-effect, not regulatory; classic "established specialist" moat
3-test:
- 5-yr lock-up: Maybe — would I be comfortable owning at 7.4× sales / 7.8× book through a probe-card cycle pause? Not at this entry; comfortable below ¥10,500.
- Unique economic engine: Yes, somewhat — 48% GM with consumable refresh tail is a healthy specialty model. But replicable in principle (FormFactor and Technoprobe already do similar).
- Blank-check disruptor: Korean / Chinese domestic probe makers under state subsidy could pressure Korean memory customers to second-source. Real but not imminent.
Quality verdict: durable specialty — better quality than Kioxia (less cyclical, higher GM, net cash), but smaller market and fewer optionality.
9. Industry Cycle Position
Where are we? Mid-late stage of HBM-led probe card up-cycle. Lead times still extended; orders firm; consensus ~2026 strong, 2027 uncertain.
Cycle history: Probe cards are mid-cycle cyclical — not as violent as NAND, but real swings (-30 to +50% revenue swings tied to memory capex cycle). MJC's 2022→2023 revenue declined ~14% from ¥44B to ¥38B in the prior cycle pause; recovered to ¥56B in 2024.
10. Emerging Threats
- Korean domestic probe makers under state push to indigenize Samsung/SK Hynix supply
- Chinese probe industry under MIIT 2030 plan
- FormFactor MEMS expansion at TSMC advanced logic
- Technoprobe HBM push if it gains qualification at SK Hynix (currently weaker memory exposure than logic)
11. Financial Analysis
| Metric | FY22 | FY23 | FY24 | FY25 | LTM |
|---|---|---|---|---|---|
| Revenue (¥B) | 44.3 | 38.3 | 55.6 | 70.2 | 70.2 |
| Rev growth | — | -14% | +45% | +26% | +26% |
| Gross profit (¥B) | 20.2 | 17.1 | 27.1 | 33.8 | 33.8 |
| GM% | 46% | 45% | 49% | 48% | 48% |
| Op income (¥B) | 9.2 | 5.3 | 12.6 | 16.5 | 16.5 |
| Op margin | 21% | 14% | 23% | 24% | 20%* |
| EBITDA (¥B) | 12.5 | 7.7 | 14.9 | 22.3 | 22.3 |
| Net income (¥B) | 7.5 | 4.1 | 8.8 | 12.1 | 12.1 |
| FCF (¥B) | n/a | n/a | n/a | n/a | -13.5 (WC build) |
| Net cash (¥B) | n/a | n/a | n/a | ~12 | 12 |
*Discrepancy: FY25 reported op margin 23.5% (¥16.5B / ¥70.2B); yfinance "operatingMargins" (TTM) 19.6% — likely the TTM cut differs from FY result.
Read:
- FY22-23 was the cycle pause (revenue -14%); FY24-25 the HBM-led recovery (+45% then +26%)
- Margin durability through the cycle is the standout — gross margin held 45-49% even in the FY23 dip; op margin only compressed to 14% (still profitable)
- Net cash + dividend (0.76% yield) — best-in-class capital structure for a specialty cap-light name
- Negative FCF TTM is CAPEX-driven, not WC drag (reconciled 2026-04-28). FY25 capex jumped to ¥19.4B from ¥7.0B in FY24 (+175%). D&A doubled ¥2.85B → ¥5.49B confirming new capacity coming online. Working capital change only -¥4.65B (smaller than capex). This is a deliberate capacity expansion ahead of expected HBM demand, funded out of net cash — not a profitability concern, but it does raise execution risk if HBM digests in 2026-27. FY26 FCF likely still negative if capex stays elevated.
Estimate revisions
- 0Q / +1Q: NaN (no quarterly estimates published — too small / thin coverage)
- 0Y: 0 up / 0 down last 30d — flat
- +1Y: 1 up / 0 down last 30d — modestly positive
Thin coverage (3-4 analysts) means revisions data is less reliable than for Kioxia. The screen attribution of "+35% rising EPS" likely refers to fiscal-year average EPS revision over the past 90 days, not the 30d feed shown above.
12. Valuation
| Metric | MJC | Probe peer median (FORM, TPRO, JEM, MPI) |
|---|---|---|
| P/S TTM | 7.4× | ~6-9× |
| P/B | 7.8× | ~3-6× |
| Trailing P/E | 43× | ~30-50× |
| Fwd P/E | 61× | ~25-40× |
| Op margin | 19.6% (TTM) / 24% (FY25) | 15-25% |
| Rev growth | +26% | +20-50% |
At 7.4× P/S and 61× fwd P/E, MJC is in the upper half of probe-card peers but not the most extended. JEM (6855.T) at $2.5B mcap with similar HBM exposure is structurally cheaper and Pink already holds. FormFactor (FORM) at $3.5B is similar size with broader portfolio and US listing.
The MJC entry case is "more leveraged to memory probe than FormFactor (which has more logic mix), and more under-the-radar than Technoprobe (which is the TSMC name)." Reasonable but not a free lunch at this multiple.
14. Catalysts
0-12 months:
- 2026-05-13 — Q1 FY26 earnings. Test of HBM-driven order book; +1Y revisions positive into print but thin coverage means surprise risk in either direction.
- HBM4 ramp data points (SK Hynix mass production timing, Samsung HBM3E qualification at NVIDIA, Micron HBM4)
- Intel 18A test plan disclosures (MJC might benefit if it's qualified for 18A advanced packaging probe)
- FormFactor/Technoprobe earnings — read-through to MJC
1-3 years:
- HBM4 → HBM5 transition (probe card density step-up)
- Chiplet test standardization (UCIe-related test requirements)
- Korean domestic probe-card competition intensification
15. Risks
| # | Risk | Likelihood | Mitigant | Closeable? |
|---|---|---|---|---|
| 1 | HBM digestion 2026-27 (memory cycle pause) | Medium-High | Customer qualification stickiness; advanced-logic offset | No — structural cyclicality |
| 2 | Korean probe-card indigenization | Medium | Long qualification cycles; Japanese ecosystem entrenchment | No — slow but real |
| 3 | Negative FCF persists / WC drag | Medium | Net cash buffer; OpCF still positive | Yes via FY26 normalization |
| 4 | Thin coverage = surprise risk on print | Medium | Q1 will reveal | One-time event risk |
| 5 | FormFactor/Technoprobe taking advanced-logic share | Medium | MJC core is memory, less exposed | No — competitive |
| 6 | JPY appreciation | Low | High export mix | Macro |
Bear case: HBM4 ramp delay + Samsung qualification dual-sources to a Korean domestic + 2026 logic capex pause → revenue decelerates to flat in FY26, op margin compresses to 17%, EPS stalls at ¥440-470. Multiple compresses to 25-30× fwd P/E → ¥6,500-9,000 (-32 to -51%).
Dilution risk: Low — mature small-cap, no shelf, stable share count.
Key-person risk: Possible founder/family overhang in succession, but no specific signal flagged.
16. Ownership
| Holder | Type | Stake |
|---|---|---|
| Insiders (founder/family + management) | Strategic | 18.5% (aggregate per yfinance) |
| Institutional | Public | 42.8% |
| Free float / retail | Public | ~38% |
Top non-strategic institutions are diffuse passive flow (Vanguard, College Retirement, iShares, DFA Japan small-cap). No 13D-equivalent activist or strategic holder detected.
Analyst sentiment: thin coverage (3-4 analysts), recommendation key "none." Mean PT ¥9,900 = -26% from current ¥13,360. High ¥14,000 (+4.8%), low ¥5,800 (-57%). The market has run well above sell-side mean — either the Street is missing the HBM thesis, or the stock is over its skis.
17. Position Sizing & Entry
Conviction: Medium on thesis, Low at current price.
At ¥13,360: WATCH. Stock 26% above analyst PT mean, near 52w high, 17% above 50dMA, 72% above 200dMA. Coverage is thin so the "premium to analyst PT" is less meaningful than for Kioxia, but the technical extension is real.
Entry plan if accumulating:
- Initial 1% position on pullback to ¥11,500 (-14%, near 50dMA)
- Add 1-1.5% at ¥10,000 (-25%, post-cycle-pause evidence)
- Stop on initial entry: ¥9,500 (-29%; just below 200dMA-implied support)
- Full position: 2-3%
Better question first: Pink already holds JEM (6855.T) which is structurally cheaper at $2.5B mcap with similar HBM exposure. Does adding MJC genuinely diversify the probe-card thesis or just double up? Answer in the compare write-up.
Recommendation
Verdict: WATCH at current price (¥13,360).
Probe-card thesis is real and durable. MJC is a high-quality specialty operator with 48% GM, net cash, and a credible HBM growth tailwind. But the stock has tripled in 12 months and trades 26% above analyst PT mean. Better to wait for either (a) post-earnings pullback to ¥10,500-11,500 with stop just below ¥9,500, or (b) a cycle pause shake-out that re-tests ¥9,000-10,000.
Compare with Pink's existing JEM holding before sizing — likely overlap not diversification. See compare next.
Open items (updated Apr 28)
- [x] FCF reconciliation 2026-04-28: confirmed negative FCF is capex-driven (¥19.4B FY25 capex, +175% YoY), not WC drag. See Section 11. FY26 FCF likely negative if capex stays elevated.
- [x] IR scan 2026-04-28: MJC English IR landing page surfaces no April 2026 disclosures with material content. Continue monitoring through May 13 print.
- [ ] Run /filings 6871.T (TSE filings — manual review of EDF for any post-May-13 changes)
- [ ] Pull Q1 FY26 print (2026-05-13) and update FY26 outlook
- [ ] Confirm top customer concentration (likely from Securities Report annual filing)
- [ ] PSU/RSU hurdle structure
- [ ] Critical: verify FCF TTM negative figure — confirm WC build, not actual cash burn
Sources
- semi-probe-card-primer — probe card industry primer
- burn-in-test-primer — adjacent test consumable thesis
- 6855 — JEM (Pink's existing holding) — direct peer comparison
- themes/memory-sector-brief — HBM-driven memory cycle context
- yfinance (financial / price / estimates) 2026-04-27
- Micronics IR (https://www.mjc.co.jp/eng/ir/)
- TechInsights, Yole probe card market 2025 estimates
Briefings
- 2026-04-27 · Stock Showdown: Kioxia (285A.T) vs Micronics Japan (6871.T) · vault