Home/Reports/Evolve DD — 2026-05-03

Evolve DD — 2026-05-03

2026-05-03 19:16 · 17 KB

Stocks: Rohm (6963.T) · Renesas (6723.T) · DISCO (6146.T)

*Deep evolution round. Research date: 2026-05-03. All three stocks had data_date 2026-04-12 (21 days stale) with imminent earnings catalysts.*


Executive Summary

TickerNameThemePrior ConvictionUpdatedThesis Status
6963.TRohm Co.EV / Power SemiLOWLOW → HOLDThesis pivot: pure-play SiC → Japan power semi consolidation anchor
6723.TRenesas ElectronicsSemi / AIMEDIUMMEDIUM (upgrade signal)Auto recovery confirmed Q1 2026; GF partnership; HOLD pre-May 8
6146.TDISCO CorpSemi / AIMEDIUMMEDIUM (upgrade signal)FY2026 6th consecutive record; Q1 FY2027 re-acceleration guided

Rohm Co. (6963.T) — EV / SiC Power Semiconductor

What Changed Since April 12

Most important: Three-way Japan power semiconductor merger MOU (March 27, 2026)

Rohm signed an MOU with Toshiba Electronic Devices & Storage (TDSC) and Mitsubishi Electric's power device business to discuss business integration. Combined entity would hold ~10% global power semiconductor market share — No. 2 globally after Infineon (17%). Complementary tech stack: Rohm SiC + Toshiba Si MOSFET + IGBT + Mitsubishi IGBT/modules. METI has already approved ¥129.4B ($900M) in subsidies for Rohm + Toshiba power semiconductor manufacturing collaboration. This is a government-backed consolidation move — the thesis is no longer "pure-play SiC" but "anchor of Japan's power semiconductor national champion."

Denso acquisition bid collapsed (April 25–28, 2026)

Denso proposed ~¥1.3T ($8.3B) acquisition of Rohm in March 2026 (~4.98% existing stake). Rohm management rejected, citing the three-way merger as the preferred strategic path. Denso withdrew April 28. Stock fell 16% on April 27 — the M&A premium has been removed. The stock's 52-week range is ¥2,247–¥3,890; it is currently trading near ¥3,431.

Supply chain corrections

  • Tesla: REMOVED. Tesla Model 3 uses STMicro, not Rohm. This was an error.
  • Toyota bZ5: ADDED CONFIRMED. Rohm 4th Gen SiC MOSFET in traction inverter. Mass production from HAIMOSIC (Shanghai) JV, 360k units/year. (Source: Semiconductor Today, June 2025)
  • Schaeffler: UPGRADED to mass production. 800V inverter brick using Rohm SiC MOSFET in mass production since September 2025. Customer: unnamed major Chinese automaker.

5th Gen SiC MOSFET announced (April 21, 2026)

30% lower on-resistance at high temperature vs 4th Gen. Discrete/module samples available July 2026. Roadmap: Gen 6 (2027), Gen 7 (2029). Targets: xEV traction, OBC, AI server PSUs.

Wolfspeed tailwind

Wolfspeed filed Chapter 11 June 2025, emerged September 2025. Rohm is a primary beneficiary for customer requalification. Wolfspeed's market position weakening long-term.

FY2026 earnings date corrected: May 12 (not May 8)

Nikkei preview article suggests actual FY2026 net income ~¥7B — below company guidance of ¥10B (raised Feb 2026) and well below analyst consensus of ~¥14.1B. Cause: heavier-than-expected SiC quality assurance costs in Q4 (tied to automotive mass-production qualification). These are described as temporary. Watch at May 12: actual NI, FY2027 guidance, SiC quality cost resolution timeline, merger JV structure update.

Updated Metrics (May 2026)

MetricApr 12May 3Source
Market Cap¥1.41T¥1.35Tcompaniesmarketcap, Apr 28
P/B1.52x1.37xIRBank, Mar 2026
Forward PE55.75x~125xIRBank/Simply Wall St
Op Margin-4.1%+1.6% (Q3 FY2026)TDnet Q3 filing, Feb 2026
FCF-¥44BStill negativeCapex investment cycle

Anti-Pattern Check

  • Peak Earnings: CLEAR — FY2026E NI ¥7-10B vs historical peak ~¥50B+. Early recovery.
  • Capacity Hangover: FLAG — ¥180B/yr capex peak drove largest-ever operating losses. Pulled back to ¥50B/yr. SiC quality assurance costs in Q3-Q4 FY2026 are hangover symptoms. Capex reduction now confirmed.
  • Concept Stock: MONITOR — SiC is ~20-22% of group revenue (discrete 44% × SiC ~50% within). Three-way merger would further dilute SiC exposure.

Valuation Guardrails

  • P/B 1.37x: No peak concern — actually below typical floor for Japanese industrials.
  • Forward PE ~125x: Appropriate for cyclical trough — buying near-breakeven with recovery to >20% OP margin target by FY2028. Catalyst named: Toyota bZ5 ramp + Schaeffler ramp + Wolfspeed customer wins.

Thesis (Updated)

Anchor participant in Japan's government-backed power semiconductor consolidation (Rohm + Toshiba TDSC + Mitsubishi Electric MOU; METI ¥129.4B subsidy). Combined entity targeting No. 2 globally. SiC recovery intact: Toyota bZ5 in mass production, Schaeffler 800V inverter in mass production, Wolfspeed tailwind. Gen 5 SiC MOSFET sampling July 2026. FY2028 target: >20% OP margin, SiC breakeven.

Thesis breaks if: FY2026 NI <¥7B (SiC quality costs structural); three-way JV collapses; China SiC achieves automotive qualification at scale by 2027; FY2027 OP margin guidance <10%.

Risk Flags

  • AP02 triggered: Capacity expansion hangover — peak ¥180B/yr capex cycle drove operating losses. SiC quality assurance costs in Q4 are residual hangover effect.
  • Earnings miss risk: Nikkei preview ¥7B NI vs ¥10B guide. Q4 likely weaker than planned.
  • Denso premium removed: Stock -16% on Apr 27; fair value now purely on fundamentals.

Renesas Electronics (6723.T) — Automotive MCU / AI Data Center

What Changed Since April 12

Q1 2026 results already reported (April 24) — strong beat

  • Revenue: ¥380.3B IFRS (+23.2% YoY) — beat consensus ¥368.9B
  • Operating profit (Non-GAAP): ¥102.9B; OP margin 33.7% (above management's 25-30% LT target)
  • Automotive: +10.6% YoY; channel inventory fell vs plan in Q1 (demand-pull signal)
  • H1 2026 guidance: Revenue ~¥760B (+20% YoY), OP margin 31.3%
  • Q2 guided: OP margin ~29% (FX/mix headwind, still healthy)
  • Note: GAAP "EPS miss" headlines were a GAAP/Non-GAAP definitional artifact. Underlying business strong.

GlobalFoundries expanded US manufacturing partnership (February 2026)

Multi-billion USD partnership for FDX/BCD/CMOS tape-outs starting mid-2026. US fabs, Germany, Singapore. Explicit rationale: tariff resilience, US supply chain credibility. Proactive hedge vs Section 232 risk. Renesas becomes a "made in USA" option for MCU supply chains.

RH850/U2C 28nm new MCU (March 4, 2026)

32-bit automotive MCU targeting ASIL-D: chassis/safety, BMS, body control, zonal ECU architectures. Extends RH850 roadmap to 28nm from 40nm. Direct response to EV/BMS growth and Toyota zonal ECU adoption.

R-Car X5H (3nm) sampling with Bosch and ZF

World's first 3nm automotive multi-domain SoC. Silicon sampling shipping to select customers. Bosch (CES 2026 fusion demo) and ZF (joint ADAS showcase) engaged — pre-production design stage. Production vehicles 2029-2031.

Automotive cycle position: early recovery, not peak

FY2025 automotive revenue -9% YoY (trough). Q1 2026 +10.6% — first recovery quarter. Channel inventory lean and falling. S&P Global projects 16.5% YoY auto semiconductor growth through 2026.

May 8 is H1 guidance reaffirmation, not new quarterly results

The key event on May 8 is confirmation of H1 2026 guidance already disclosed. Watch for: Q2 OP margin vs 29% guided; any China MCU commentary.

Updated Metrics (May 2026)

MetricApr 12May 3Source
Market Cap¥4.82T¥5.29TTradingView, May 2026
P/B1.97x2.29xMultiple sources
Op Margin17.9%33.7% Non-GAAP Q1Renesas Q1 2026 earnings call
FCF+¥364B+¥364B confirmedGuruFocus (OCF ¥453B - capex ¥89B)
Forward PE17.78x16-18xAnalyst consensus

Anti-Pattern Check

  • Peak Earnings: CLEAR — FY2025 automotive -9% trough. Q1 2026 is first recovery quarter.
  • Capacity Hangover: CLEAR — fabless/fab-lite; GF is outsourced capacity, no greenfield risk.
  • Concept Stock: CLEAR — 48.5% automotive + 50.9% industrial genuine exposure. GaN AI growing but still modest.
  • China MCU Erosion: EMERGING — CEO cited 28nm MCU headwind in China. GigaDevice/SemiDrive/AutoChips targeting domestic substitution. 2-4 year risk (China ~40% of Asia-Pacific revenue).

Valuation Guardrails

  • P/B 2.29x: Moderate for #1 global auto MCU with 33.7% Non-GAAP OP margin and FCF ¥364B. Not at cyclical peak.
  • FCF yield ~6.9%: Strong cash generation supports valuation.

Thesis (Updated)

#1 auto MCU globally (~18% share). Auto cycle early recovery confirmed Q1 2026 — channel inventory lean, demand-pull. R-Car X5H world's first 3nm automotive SoC sampling with Bosch/ZF. AI data center GaN growing 2x YoY via NVIDIA 800V partnership. GF partnership de-risks tariff exposure. FCF ¥364B at ~6.9% yield. China MCU share erosion is the key medium-term risk.

Thesis breaks if: China revenue declines >15% in H1 2026; R-Car X5H loses Bosch/ZF engagement; GaN AI data center stays <3% of Industrial/IoT by end 2027; Q2 OP margin comes in materially below 29% guided.


DISCO Corp (6146.T) — Semiconductor Dicing / Grinding Equipment

What Changed Since April 12

FY2026 full-year results released April 22 — 6th consecutive record

  • Revenue: ¥436.9B (+11.1% YoY)
  • Operating profit: ¥185.0B (+10.9% YoY); OP margin 42.3% (beat consensus ~3%)
  • Annual dividend raised 22%: ¥413 → ¥505/share (management confidence signal)
  • Capex nearly doubled: ¥33B → ¥70B (building for orders already in hand)

Q1 FY2027 guidance (Apr–Jun 2026) — RE-ACCELERATION

  • Revenue: ¥106.1B (+18.0% YoY)
  • Operating profit: ¥42.0B (+21.8% YoY)
  • Net income: ¥29.5B (+24.1% YoY)
  • Shipments: quarterly record high forecast
  • Re-accelerates after FY2026 growth deceleration (28% → 11%) — removes near-term downside risk

FCF now confirmed: +¥63.7B

Operating CF ¥133.5B minus capex ¥69.8B. (Total investing CF appears near-zero due to fixed deposit movements — capex-based FCF is the correct measure.) FCF yield ~0.85% at ¥7.5T market cap — thin but positive.

CoWoS 3.7x expansion through end 2026

TSMC CoWoS capacity: 35k wpm → 130k wpm target by end 2026. ~3-4 dicers per 10k wpm incremental capacity implies ~27-36 additional DISCO dicers for TSMC alone.

HBM supercycle intact

2026 HBM market: $54.6B (+58% YoY). Samsung +50% HBM capacity. SK Hynix M15X + Yongin mega-fab: >400k wpm capacity addition 2026-2027 (→ ~120-160 additional dicers). HBM4 transition adds dicing complexity. Cumulative HBM purchase commitments exceed 3 years. DISCO confirmed "HBM grinding is currently DISCO monopoly supply."

Intel glass substrate: new supply chain opportunity

DISCO's proprietary "pull-back method" for SeWaRe defect elimination is the key solution for glass substrate singulation. Intel advancing glass substrates + EMIB packaging. Samsung Yokohama packaging R&D center (opening March 2027) deepens Japan equipment relationships.

Laser saw milestone: 4,000 cumulative shipments in 2025 (vs 1,000 in 2014, 2,000 in 2019) — accelerating adoption.

Earnings date note: The April 12 data listed "May 12 earnings" — this was incorrect. FY2026 full-year results were released April 22. The next earnings event is Q1 FY2027 results (around July 2026).

Updated Metrics (May 2026)

MetricApr 12May 3Source
Op Margin41.7%42.3%TDnet FY2026 filing, Apr 22
FCFN/A+¥63.7Birbank.net CF history
Market Cap¥7.26T¥7.5TYahoo Finance JP
Forward PE44.98x~43xQ1 FY2027 guidance
Div Yield0.7%0.66% (¥505)TDnet, Apr 22
P/B13.37x~13.4xYahoo Finance JP

Anti-Pattern Check

  • Peak Earnings: MONITOR — 6th consecutive record with growth decelerating (28%→11%). Q1 FY2027 re-acceleration (+18%) is meaningful but one quarter. Not a clear peak trap signal — orders/shipments still at records.
  • Capacity Hangover: CLEAR — dicing market $0.88B total is small, disciplined oligopoly. DISCO capex doubled but building for orders in hand.
  • Concept Stock: CLEAR — DISCO IS the AI packaging infrastructure. 70%+ dicing market share, HBM grinding monopoly, structurally embedded in every CoWoS wafer and HBM module.

Valuation Guardrails

  • P/B ~13.4x approaching historical peak 15.47x: TRIGGERED. Current P/B within ~15% of all-time high (set in FY2025). Priced for perfection. FCF yield 0.85% provides no cushion. If growth decelerates, multiple compression risk is significant.
  • Morningstar DCF fair value ¥17,891 vs current ¥72-76k — represents a severe outlier (DCF methodology is notorious for undervaluing moat businesses). Not actionable but noted.

Thesis (Updated)

~70%+ global dicing/grinding share. FY2026 6th consecutive record (¥436.9B, 42.3% OP margin). Q1 FY2027 re-acceleration guided (+18%/+21.8%). CoWoS 3.7x expansion implies ~27-36+ additional DISCO dicers for TSMC alone. HBM supercycle: $54.6B 2026 (+58% YoY), grinding monopoly confirmed. New: Intel glass substrate singulation opportunity (proprietary SeWaRe method). FCF +¥63.7B confirmed.

Thesis breaks if: CoWoS capacity ramp stalls (TSMC capex cuts); HBM demand collapses post-AI capex peak; HANMI achieves >10% share at Samsung HBM dicing; P/B exceeds historical peak 15.47x — reduce position.

Risk Flags

6963.T Rohm

  • AP02 Capacity Hangover: ¥180B/yr peak capex cycle → largest-ever operating losses. SiC quality assurance costs in Q3-Q4 FY2026 are residual hangover symptoms. Now pulled back to ¥50B/yr.
  • Earnings miss risk: Nikkei preview ¥7B net income vs ¥10B guidance and ¥14.1B analyst consensus.
  • Merger execution risk: Three-way JV still in MOU/discussion — structure, valuation, and timeline TBD.
  • Guardrail: FCF still negative. FCF growth projection requires named catalyst (Toyota bZ5 ramp + Wolfspeed requalification wins = named).

6723.T Renesas

  • China MCU erosion: CEO-flagged 28nm MCU headwind. GigaDevice/SemiDrive pursuing domestic substitution. 2-4 year structural risk.
  • Q2 margin step-down: OP margin guided 33.7% (Q1) → 29% (Q2) due to FX and OpEx ramp.
  • No guardrails triggered: P/B 2.29x, FCF yield 6.9%, early-cycle positioning all within normal range.

6146.T DISCO

  • PB guardrail triggered: P/B ~13.4x vs historical all-time peak 15.47x. Above 80th percentile of historical range. Override: monopoly moat + HBM supercycle + Q1 FY2027 re-acceleration justify premium but ceiling is visible.
  • AP01 watch: 6th consecutive record with decelerating growth. Monitor Q1 FY2027 actuals (July 2026) for trend confirmation.

Sources


*Disclaimer: This is internal research documentation, not investment advice. All figures sourced from Tier 1-2 sources as defined in research_philosophy.md. Supply chain links marked confirmed have source_url citations.*