InP EML supply duopoly — Lumentum vs Coherent
The merchant supply of InP-based electro-absorption modulated lasers (EML) — the source-laser device for 100G/200G/400G/800G/1.6T fiber-optic datacenter transceivers — is a structural duopoly. Lumentum and Coherent Corp (formerly II-VI / Finisar) together control ≥80% of merchant EML supply globally, with the remainder split among captive vendors (NTT Electronics, Sumitomo Electric — primarily telecom) and emerging Chinese fabs (HG Genuine, Accelink — primarily domestic-China supply with limited Western datacom penetration).
Why the duopoly is structurally stable
Three structural-moat ingredients:
- Capital intensity — InP wafer fabs run on 4-inch substrates, are vertically integrated through epitaxy, fabrication, packaging, and test, and require multi-year qualification cycles for hyperscaler-grade reliability. Greenfield InP EML capacity is not a 12-month build.
- Process know-how — 200G/lane EML modulation requires sub-micron MQW (multi-quantum-well) layer control, optical-coupling-loss optimization, and reliability engineering measured in 100,000+ hours of accelerated-life testing. Yield curves climb slowly with cumulative wafer starts, which advantages incumbents.
- Customer qualification — hyperscalers (and module-assembly vendors serving them) qualify EML chips at the device level for each module form factor and each speed-grade generation. Re-qualifying a new EML supplier is an 18–24-month project; the cost is high enough that customers prefer dual-source allocations within the existing duopoly to a true tri-source.
The structural argument is captured in the supply-constrained framing on Coherent’s Q1 FY2026 earnings call: datacenter revenue +23% YoY but “supply‑constrained by InP lasers” (Coherent press release November 5, 2025). Both vendors are sold out on the leading-edge node and pricing is rising.
Comparative position: Lumentum vs Coherent
| Metric | Lumentum | Coherent Corp | Notes |
|---|---|---|---|
| InP fab footprint | San Jose CA + Towcester UK | Sherman TX + Allen TX + global | Coherent has broader InP base post-Finisar/II-VI |
| 200G/lane EML production status | Volume shipping to multiple AI-customers | Ramping volume; supply-constrained | Lumentum claims first-to-volume on 200G/lane |
| Cloud & Networking / Datacom revenue (most recent quarter) | $665.5M total LITE Q2 FY2026 (Cloud & Networking dominant share) | $1.2B Coherent Datacom & Comms Q2 FY2026 | Coherent’s segment is broader (incl. systems, undersea, telecom transport) |
| YoY Cloud/Datacom growth (most recent) | +65%+ (LITE Q2 FY2026) | +33.6% (COHR Q2 FY2026 Datacenter & Comms) | LITE growing faster off smaller base |
| NVIDIA Series A Convertible Preferred | $2.0B / 2,876,415 shares at $695.31 | $2.0B (parallel transaction; terms similar in concept) | Symmetric NVDA dual-source structure |
| Vertical integration | Component → Cloud Light transceivers | Component → Finisar transceivers (deeper) | Coherent has a longer-running module business |
| Total revenue base (FY ending most recently) | $1,645M FY2025 | ~$5.7B (COHR FY2025) | COHR materially larger; LITE more pure-play to optics |
Coherent’s broader portfolio (laser systems, materials, networking) dilutes its pure-play exposure to AI-photonics. Lumentum is the higher-beta name on the InP-EML thesis specifically. Coherent’s LATAM/industrial segments offer ballast that Lumentum lacks.
Capacity-build timing
| Capacity event | Lumentum | Coherent |
|---|---|---|
| Sherman TX fab expansion (COHR) | n/a | Multi-phase build; volume contributing FY2025–FY2027 |
| San Jose new fab (LITE) | NVDA-funded; announced March 2026 with $2B partnership | n/a |
| Towcester UK expansion (LITE) | Ongoing; legacy Oclaro fab | n/a |
| 200G/lane EML capacity (industry view) | Ramping through CY2026 | Ramping through CY2026 |
| 1.6T module qual completion | In qual; volume CY2026 | EML and silicon-photonics 1.6T ramping CY2026 |
| 3.2T / 200G VCSEL-based 1.6T | VCSEL-based 1.6T 2H CY2026 | Coherent management framing 2H CY2026 |
| CPO commercial volume | 2027 NVDA pluggable, 2028 NVDA scale-up CPO ramp | Same industry timeline |
Note: capacity-build timelines are management-framed; actual capacity-coming-online dates are confidential and only inferable from quarter-on-quarter capex disclosures.
Pricing discipline mechanics
In a structurally supply-constrained duopoly with fungible end customers, pricing power is asymmetric:
- Spot ASPs rise when customer demand growth (running ahead of capacity adds) outpaces supply growth — this is the current regime, with double-digit ASP increases on 200G EML in CY2026 (per industry trade-press analyst estimates — ⚠).
- Volume contracts (multi-year) lock in floor pricing — NVDA’s $2B equity investment plus multibillion-dollar purchase commitment to Lumentum is best read as a volume pre-purchase against capacity that Lumentum is dedicating to NVDA. The implied ASPs on this committed-volume tier are confidential; reasonable bracketing puts NVDA’s effective per-chip pricing somewhere between aggressive volume discounts and merchant ASPs, with the pricing discipline preserved by the structure that NVDA’s volume is incremental, not displacing existing merchant commitments.
- Mix shift to higher speeds drives ASP growth — 800G EML chip ASPs are higher than 400G; 1.6T EML ASPs higher still. As the mix tilts to leading-edge speed grades, blended ASPs rise even before considering supply-tightness premium.
The pricing discipline is fragile in two scenarios: (a) hyperscaler demand inflects negative on AI capex digestion, leaving capacity adds stranded; (b) NTT/Sumitomo decide to enter merchant datacom EML supply at scale, which would convert the duopoly to a tri-opoly with attendant margin compression. Neither has materialized as of April 2026.
NVDA’s dual-source structure preserves the duopoly
NVDA’s parallel $2B equity investments in both Lumentum and Coherent (CNBC March 2, 2026) is the most informative signal on the duopoly’s persistence. If NVDA had concluded one supplier was superior, a unilateral investment would have been the move. Instead NVDA chose the symmetric structure — bilateral capacity dedication, dual-source supply continuity, and no obvious tilt of competitive dynamics between the two suppliers. The duopoly is the partnership architecture.
This is also a tell against the “NVDA in-sources optics” risk. Owning equity in both incumbent merchant EML suppliers is consistent with NVDA’s view that the merchant model serves NVDA better than vertical integration would — at least through the 2027–2028 capacity cycle.
What would break the duopoly
- A credible third-source emerges with hyperscaler qual — most plausible candidates: Sumitomo Electric (Japanese; primarily telecom; could pivot), HG Genuine (Chinese; export-control-constrained for Western shipments), Marvell-internal silicon photonics (sidesteps InP entirely with photonic-integrated-circuit alternatives).
- Hyperscaler in-sources — least likely; the qualification + capex burden is high relative to merchant supply availability. AWS has the largest in-house optics team, but even AWS qualifies merchant EML for the bulk of its volume.
- Silicon-photonics + heterogeneous-integration approaches displace InP at the source-laser layer — currently silicon photonics relies on InP gain-chip flip-chip-bonded onto SiPh waveguide; the InP component is still merchant-sourced. Disruption would require monolithic InP-on-Si or quantum-dot-on-Si lasers reaching production — multi-year horizon.
Cross-link
- AI capex cycle — top-down demand
- Datacenter optics TAM — share allocation framework
- 03_ecosystem competitors — Coherent and other rivals
- 03_ecosystem NVIDIA partnership — bilateral structure detail
- 02_technology InP EML process — node-level capability
Sources
- Lumentum 8-K — March 2, 2026 Series A Convertible Preferred Stock placement ✓
- Coherent Q1 FY2026 press release — supply-constrained framing ✓
- Coherent Q2 FY2026 press release ✓
- CNBC — NVIDIA $4B parallel investment in Coherent and Lumentum ◐
- TrendForce — laser shortage analysis December 2025 ◐
- Lumentum FY2025 10-K segment disclosures ✓