Home/Reports/Light Theme Evolve — 3 Stocks Updated for TSMC COUPE/MRM 2026 Announcement

Light Theme Evolve — 3 Stocks Updated for TSMC COUPE/MRM 2026 Announcement

2026-05-15 03:50 · 8.1 KB

Date: 2026-05-15 | Time: 03:50 JST | Type: Evolve (fired from research_evolve.sh workflow inline) | Theme: Light/Photonics

Trigger

TSMC's COUPE 2026 mass-production announcement (200Gbps MRM, External Laser Source architecture, 3D hybrid bonding) directly affects 3 stocks already in the Light theme. Their thesis, conviction, and supply chain entries need updates.

TickerOld convictionNew convictionDirection
5802.T Sumitomo ElectricMEDIUMHIGHUPGRADED
6503.T Mitsubishi ElectricHIGHMEDIUMDOWNGRADED
6920.T LasertecHIGHHIGHHOLD (thesis strengthened)

Stock 1: Sumitomo Electric (5802.T) — UPGRADED MEDIUM → HIGH

Why upgraded

CPO architecture uses External Laser Source (ELS) + silicon-native MRM modulation. The ELS is an InP laser sitting outside the photonic IC package. From viksnewsletter explanation of the CPO laser architecture:

> "Each laser must provide 24.5 dBm of optical power to compensate for losses and ensure reliable transmission. Key materials and laser technologies are provided by global players such as Coherent and Sumitomo Electric."

And per 36Kr InP industry analysis: "Sumitomo Electric of Japan leads with a 60% market share in indium phosphide production" — every CPO ELS laser needs InP.

Before TSMC COUPE, Sumitomo Electric was a "general InP wafer leader" with somewhat speculative CPO exposure. Post-TSMC, the named explicit supplier path is confirmed. The InP TAM expands because CPO uses ELS at higher power (24.5 dBm vs typical EML output ~13 dBm).

Updated thesis

Sumitomo Electric is now positioned as a directly named supplier in the CPO supply chain — not just an inferred upstream player. 60% global InP wafer share + named ELS laser supplier role + IOWN multicore fiber JV with NTT/NEC. The pre-cycle valuation (Fwd PE 23.9×) is anomalous given the named-supplier status; expect re-rating once CPO unit shipments confirm in 2026 H2.

Updated key dates

  • Add: 2026-Q3 — first hyperscale CPO module shipments from TSMC COUPE customers (NVIDIA + Broadcom); Sumitomo Electric ELS InP order signal
  • Add: 2027 H1 — second-gen 400G MRM with multi-wavelength → more ELS lasers per module, multiplicative on InP demand

Source URLs added

Stock 2: Mitsubishi Electric (6503.T) — DOWNGRADED HIGH → MEDIUM

Why downgraded

CPO replaces EML with silicon MRM. Mitsubishi's photonic-segment moat was its 200Gbps EML chip — direct competitor to Lumentum's EML — but the X post-cited dominance ("Lumentum's monopoly bottleneck") evaporates inside the photonic die: TSMC's COUPE uses silicon-native ring modulators, not InP EMLs.

Key clarification: EMLs are NOT extinct. Pluggable 800G/1.6T transceivers will keep using EMLs through 2028. But:

1. Lumentum is named as "the only supplier shipping 200G-per-lane EMLs at volume" — Mitsubishi has samples but is behind in volume ramp.

2. CPO/COUPE eliminates EMLs from the photonic die path entirely.

3. Mitsubishi's POET 3.2T partnership uses 400G EMLs in optical engine chipsets — still relevant for non-CPO pluggable evolution, but a smaller and shorter window than originally assumed.

The PE 32× / Fwd PE 27× valuation that looked cheap relative to Lumentum's PE 60× was partly justified by exactly this: Mitsubishi is one tier below in EML volume + the EML window itself is narrowing.

Updated thesis (truncated)

Mitsubishi Electric remains the only non-US-aligned EML supplier at meaningful scale, but the addressable market for EMLs narrows as CPO ramps from 2026. Position Mitsubishi as a pluggable-cycle EML beneficiary with a 2-3 year window, NOT a CPO-cycle name. Downgrade conviction to MEDIUM. If POET 3.2T optical engine wins a named hyperscale design-in (in the 2027 pluggable refresh), re-upgrade. Otherwise size accordingly.

Updated monitoring trigger

  • If Lumentum's Nvidia EML exclusivity extends to 1.6T (200G/lane × 8 lanes), Mitsubishi's EML window narrows further → trim
  • If Mitsubishi announces volume shipment of 200G EML (catching up to Lumentum), upgrade back to HIGH
  • If POET 3.2T optical engine wins a Tier-1 design (NVIDIA, Broadcom, Coherent), upgrade

Source URLs added

Stock 3: Lasertec (6920.T) — HOLD HIGH (thesis strengthened, no conviction change)

Why thesis strengthened

TSMC COUPE 2026 mass production = more photonic IC production = more EUV/DUV masks required = more inspection passes on Lasertec's ACTIS A300 systems. The COUPE EIC (electronic die) is produced on TSMC's N4P or N3 advanced node, both EUV-printed — so Lasertec mask inspection is mandatory.

Additionally, the SoIC-X 3D hybrid bonding process requires bond-pad alignment and overlay accuracy that depend on precision mask quality. Lasertec's monopoly on actinic EUV mask inspection means every COUPE photonic-IC mask ships from a Lasertec-inspected blank.

Why no conviction change

Lasertec was already HIGH at PE 44× / +257% 52w. The valuation already prices in TSMC's EUV ramp. Adding "COUPE will use TSMC N3" doesn't change the conviction — it strengthens it but doesn't reset the price target.

Updated supply chain — add COUPE mask demand

New supply_chain entry: "TSMC COUPE photonic IC masks — 200Gbps MRM photonic die produced on N3, requires Lasertec ACTIS inspection for EUV-printed mask blank quality, confirmed indirect via TSMC COUPE 2026 ramp and Lasertec exclusive actinic mask inspection position."

Source URLs added

Other stocks NOT evolved this round (with reasoning)

TickerWhy not evolved this round
9432 NTTAlready at HIGH conviction with explicit Broadcom 2nd-gen PEC 2026 H2 narrative; TSMC COUPE doesn't change the IOWN thesis (different layer of optical stack)
6965 HamamatsuPhoton detection (PMT) thesis is particle physics + medical, not AI data center; tangential to COUPE
5801 Furukawa Electric / 5803 FujikuraOptical fiber thesis is intact — CPO doesn't remove fiber, just changes what's inside chips
6754 AnritsuTest instrument for pluggable transceivers; CPO test will eventually need different instruments — evolve at next round if Anritsu announces CPO test product
6702 Fujitsu / 6701 NECOTN backbone — unchanged by chip-package CPO
6856 HORIBAMass flow + Raman — slightly strengthened by more silicon photonics fabs, but no specific TSMC COUPE link to re-thesize
6925 USHIOAlready not a CPO play; no update needed

Sources