Co-packaged optics (CPO) roadmap

Co-packaged optics (CPO) is the structural shift that moves the optical engine — the laser source, modulator, photodetector, and driver electronics — from a pluggable module on the front panel of a switch to a position co-packaged on the substrate next to the switch ASIC. The motivation: as switch ASICs scale toward 100 Tbps and beyond, the electrical SerDes between ASIC and front-panel pluggables become the dominant power and reach bottleneck. Bringing the optical engine on-substrate eliminates the host-side SerDes and enables substantially better power-per-bit at the system level.

For Lumentum, the CPO transition is not a revenue cliff — it’s a vector that absorbs the same InP wafer capacity, reconfigured for the new physical architecture. The key change is the laser source: in CPO, the laser sits outside the optical engine as an “External Laser Source” (ELS), and the modulators inside the engine use silicon photonics rather than monolithic InP EML.

Lumentum’s CPO position — External Laser Source supplier

Lumentum’s flagship CPO product is the ultra-high-power (UHP) 1310 nm continuous-wave (CW) DFB laser — designed specifically as the External Laser Source for NVIDIA’s CPO platforms.

Per Lumentum OFC 2025 release: “Lumentum’s ultra-high-power (UHP) 1310 nm DFB lasers are engineered for the demanding requirements of co-packaged optics” and were integrated into NVIDIA’s “Spectrum-X Photonics and Quantum-X Photonics networking switches” Semiconductor Today 2025-04-02 ✓.

UHP CW DFB lasers differ from EMLs in two important ways:

Attribute	EML (200G/lane)	UHP CW DFB (CPO ELS)
Function	Laser source + integrated modulator	Laser source only — modulation done by separate SiPh modulator
Output power	Modest (sufficient for direct-attached SiPh-equivalent drive)	Ultra-high — must drive multiple SiPh modulators downstream after fan-out and splitting
Wavelength	Various (1271, 1291, 1311, 1331 nm CWDM4 grid)	1310 nm
Operating mode	Modulated CW	Pure CW (continuous wave)
Fab process	InP EML monolithic integration	InP DFB (simpler structure but higher output power)

The UHP DFB is in many ways the more fab-process-demanding device — it pushes InP gain-region engineering to higher power densities. But it is structurally simpler than the EML (fewer integrated layers, no EAM section), so it can scale to higher output volumes per wafer.

NVIDIA Spectrum-X / Quantum-X Photonics — the lead CPO vehicle

NVIDIA announced its first CPO networking switch products in March 2025, framing them as the optical-fabric backbone for next-generation AI factories scaled to “millions of GPUs”:

Spectrum-X Photonics — Ethernet-based CPO switches
Quantum-X Photonics — InfiniBand-based CPO switches

Per NVIDIA’s announced ecosystem, partners include “TSMC, Browave, Coherent, Corning Incorporated, Fabrinet, Foxconn, Lumentum, SENKO, SPIL, Sumitomo Electric Industries and TFC Communication” NVIDIA Spectrum-X Photonics announcement 2025-03 ✓.

Lumentum’s specific role: External Laser Source supplier — UHP 1310 nm DFB lasers fed externally into the CPO switch package.

Per Nvidia’s VLSI Conference 2025 disclosure: NVIDIA highlighted Lumentum for “single high-power DFBs among its laser partner ecosystem for CPO solutions” HPCwire 2026-04-20 ◐.

The 2028 commercial-volume thesis

Lumentum (and the broader CPO ecosystem) frames CPO as a 2028+ commercial-volume opportunity. The aligning narrative across multiple sources:

Milestone	Timing	Source
First Spectrum-X/Quantum-X CPO switches sampling	2025-2026	NVIDIA 2025-03 ✓
Lumentum Greensboro fab production ramp	mid-2028	Lumentum 2026-03-26 ✓
Nvidia Feynman GPU generation with NVL1152 scale-up	2028 — CPO integrated directly into NVLink switch	SemiAnalysis 2025-2026 ◐
Lumentum 2028 EPS target	$30/share aspiration	Financial Content 2026-03-30 ◐

⚠ The 2028 timing alignment between Lumentum’s Greensboro production ramp, NVIDIA’s Feynman generation, and the CPO scale-up phase is not coincidental — the capacity build-out is structurally aimed at this window.

The capacity built today for 200G/lane EML pluggable transceivers is positioned to feed both pluggable form factors and CPO (via the UHP DFB derivative process) when the CPO transition lands. The 6-inch InP wafer scale at Greensboro is what makes the volume transition economically viable Semiconductor Today 2026-03-26 ✓.

How 200G/lane EML capacity feeds CPO

The structural relationship between the two technology nodes:

200G/lane EML is the pluggable-transceiver workhorse for 1.6T modules — the load-bearing AI-photonics revenue today.
UHP CW DFB is the CPO ELS — the load-bearing CPO revenue in 2028+.

Both devices are fabricated on the same InP wafer process, in the same fabs, with much of the same process equipment. The MOCVD epitaxy, the active-layer growth, the metallization, the testing — all share infrastructure. Capacity built for 200G/lane EML has substantial process commonality with UHP DFB capacity.

This is why the NVIDIA $2B investment + Greensboro fab acquisition + San Jose UHP-laser expansion are all part of one strategic capacity buildout rather than three separate decisions:

NVIDIA $2B (March 2 2026) — capital injection
San Jose UHP laser capacity expansion (August 2025 announcement) — UHP DFB scale-up Semiconductor Today 2025-08-08 ✓
Greensboro NC fab acquisition (March 26 2026) — 6-inch InP capacity for both 200G EML and UHP DFB

Competitive positioning vs. NVIDIA-internal CPO modules

A persistent question for the Lumentum thesis: does NVIDIA eventually internalize the optical engine and reduce dependence on Lumentum?

The answer based on disclosed structure:

The CPO module — silicon photonics modulators, photodetectors, electrical-to-optical conversion — is being built in NVIDIA’s reference platform with TSMC and Coherent as named SiPh and integration partners.
The External Laser Source — the UHP CW DFB — is the InP-fab-dependent part that requires a III-V semiconductor process Lumentum and Coherent operate. NVIDIA is not building captive III-V fabs.
The multi-billion-dollar purchase commitment in the March 2026 strategic agreement structurally locks NVIDIA-Lumentum component flow for the CPO ramp window.

So the long-term Lumentum-vs-NVIDIA-internal risk specifically attaches to:

SiPh modulator supply (NVIDIA could shift between Coherent and Lumentum, or to TSMC-internal SiPh)
Pluggable-transceiver assembly (NVIDIA could shift between Cloud Light, Innolight, Eoptolink, or new module-builders)

Not to the InP source-laser layer, which remains a structurally constrained III-V semiconductor capability that only Lumentum, Coherent, and a small set of telecom-internal players (NTT, Sumitomo) possess at the required process maturity.

⚠ Risk caveat: the long-term competitive position depends on NVIDIA not entering captive III-V manufacturing. As of 2026-Q1, NVIDIA has not announced any captive InP fab capability, but this should be monitored as a potential structural shift.

Cross-thesis links

NVIDIA Spectrum-X / Quantum-X Photonics — anchor CPO vehicle. See 03_ecosystem / NVIDIA partnership.
InP EML process — the technology root that feeds both pluggable EML and CPO ELS UHP DFB. See InP EML process.
LWLG (Lightwave Logic) — electro-optic polymer modulators are an alternative SiPh modulator approach for CPO; they could in principle replace the SiPh-modulator part of the CPO module while leaving the Lumentum UHP DFB layer unchanged. The two stacks may complement at the system level.
Coherent Corp — both the InP EML duopoly counterparty and the named NVIDIA SiPh-integration CPO partner. The Lumentum-Coherent dynamic is multi-layered (cooperate as duopoly suppliers; compete on CPO module integration).
TSEM / GFS Fotonix — silicon-photonics foundries that may participate in CPO modulator fabrication; relevant to the SiPh layer but not the InP source-laser layer.