Datacom transceivers — Cloud Light heritage and the 1.6T ramp

The August 11, 2023 acquisition of Cloud Light Technology Limited for ~$750M Lumentum 01_company / Cloud Light acquisition added vertically-integrated pluggable datacom-transceiver assembly capability to Lumentum’s prior component-only position. Cloud Light’s Hong Kong-headquartered organization with Thailand-based assembly serves hyperscalers directly with finished optical modules — a structural shift away from selling chips to module-builder customers (Innolight, Eoptolink) who in turn sell to hyperscalers.

The Cloud Light division is the load-bearing path for Lumentum’s 800G and 1.6T transceiver shipments through the AI-cluster build-out cycle.

Transceiver product portfolio

Lumentum publicly catalogues the following CloudLight datacom transceiver products (selection of the highest-volume AI-cluster SKUs):

Product	Data rate	Form factor	Reach	Lane structure	Technology	Source
800G 2×DR4 OSFP	800 Gbps	OSFP	500 m SMF	8× 106.25 Gbps PAM4 (=8×100G/lane)	Silicon photonics (SiPh) modulators with InP CW source lasers	Lumentum product page ✓
1.6T 2×DR4 TRO OSFP	1.6 Tbps	OSFP	500 m SMF	8× 212.5 Gbps PAM4 (=8×200G/lane)	DSP + silicon photonics (SiPh); transmit-side retiming	Lumentum product page ✓
1.6T DR8 TRO OSFP	1.6 Tbps	OSFP	500 m SMF	8× 200G EML lanes	InP EML × 8 transmitters; “Transmit-Retimed Optical” architecture	Industry product reporting Convequity 2026 ◐

Note: Lumentum’s transceiver portfolio uses a mix of technologies — the 1.6T DR8 TRO OSFP is built on 8× 200G/lane InP EMLs (the AI-photonics-thesis-load-bearing path), while the 1.6T 2×DR4 TRO OSFP and 800G 2×DR4 OSFP use silicon-photonics modulators with InP continuous-wave source lasers (a structurally different optical-engine architecture). This is an important nuance — Lumentum is not exclusively EML-based on the transceiver side; the SiPh-with-InP-source-laser path is meaningful for power-efficiency targets Lumentum 1.6T 2×DR4 product page ✓.

Detailed product specifications

800G 2×DR4 OSFP

Spec	Value
Aggregate data rate	800 Gbps
Per-lane rate	106.25 Gbps PAM4 (8 lanes)
Reach	500 m over single-mode fiber
Optical interface	2× MTP-12/APC connectors
Power consumption	<15.2 W typical
Operating temperature	0–70 °C
Optical-engine	Silicon photonics (SiPh) modulators

Lumentum 800G 2×DR4 OSFP product page ✓

1.6T 2×DR4 TRO OSFP

Spec	Value
Aggregate data rate	1.6 Tbps
Per-lane rate	212.5 Gbps PAM4 (8 lanes)
Reach	500 m over single-mode fiber
Optical interface	Dual MPO-16/APC interfaces
Power consumption	16 W typical
Architecture	DSP + silicon photonics; transmit-side retiming reduces power dissipation
Management	CMIS 5.3

Lumentum 1.6T 2×DR4 TRO OSFP product page ✓

The “Transmit-Retimed Optical” (TRO) designation indicates the OSFP module retimes the transmit-side signal but skips full DSP regeneration on the receive side, saving power versus a fully-retimed module while preserving signal quality across the 500 m reach Lumentum 1.6T product page ✓.

Form factors — OSFP, OSFP-XD, QSFP-DD

The current AI-datacenter pluggable form-factor stack:

Form factor	Typical use	Notes
QSFP-DD	400G / 800G	Earlier-generation form factor; backwards-compatible
OSFP	800G / 1.6T	The volume form factor for current AI-cluster spine; what Lumentum’s 1.6T DR8 TRO OSFP and 2×DR4 TRO OSFP target
OSFP-XD	1.6T+ (proposed)	Higher-power envelope for >1.6T modules; emerging
CPO (co-packaged optics)	1.6T+	Not a pluggable form factor; integrated with switch ASIC. See CPO roadmap

NVIDIA’s reference platforms for AI clusters (Spectrum-X, Quantum-X) define both pluggable interfaces (OSFP-based) and CPO platforms (Spectrum-X Photonics / Quantum-X Photonics). Lumentum supplies optical components into both paths.

Hyperscaler customer base — caveats

Cloud Light’s heritage customer base prior to the August 2023 Lumentum acquisition was widely reported in trade press as Microsoft- and Meta-centric. Post-acquisition, Lumentum has not formally disclosed customer-level concentration — but FY2025 10-K disclosures and earnings commentary point to large concentration among a small set of hyperscalers:

Hyperscaler	Status (post-Cloud Light)
Microsoft Azure	Cloud Light heritage customer; ongoing direct-supply relationship ◐
Meta	Cloud Light heritage customer; ongoing direct-supply relationship ◐
Google	Limited; Google has substantial internal optics capability ◐
Amazon AWS	Limited; AWS uses internal + multiple merchant suppliers ◐
NVIDIA	Anchor customer post-March 2026 strategic partnership and multi-billion purchase commitment NVIDIA 2026-03-02 release ✓

◐ Hyperscaler-level customer attribution is not formally disclosed by Lumentum or by the customers; reconstructed from industry trade press and analyst commentary. Treat as ◐ partial / aggregator. The NVIDIA relationship is now a primary-source-disclosed direct customer relationship via the strategic partnership.

The FY2025 10-K does formally identify a 10%-of-revenue concentration class: customer concentration above 10% of revenue is disclosed as a single hyperscaler 10-K Item 1 / Item 7A ✓ — but the specific identity is not named.

Vertical integration — the Cloud Light strategic rationale

The competitive significance of the Cloud Light acquisition is best understood as closing the vertical-integration loop:

Layer	Pre-Cloud Light (Lumentum 2022)	Post-Cloud Light (Lumentum 2026)
InP wafer epitaxy	In-house (San Jose, Caswell)	In-house
EML chip fab	In-house	In-house
Optical sub-assembly	In-house (TROSA, COB)	In-house
Pluggable transceiver module	Sold to module-builders (Innolight, Eoptolink, etc.)	Direct to hyperscaler (Cloud Light Thailand assembly)
Hyperscaler relationship	Indirect via module-builders	Direct

The structural shift means Lumentum captures more of the gross-margin stack on AI-cluster transceivers and is exposed directly to hyperscaler ramp dynamics rather than being intermediated by module-builders that compete on price.

The trade-off: as a hyperscaler-direct module supplier, Lumentum now competes head-to-head with Innolight and Eoptolink (Chinese module-builders that buy InP EML chips from Lumentum and Coherent and assemble pluggables for hyperscaler bids). This is partially a customer-vs-competitor blurring — Lumentum continues to sell merchant InP EML chips to Innolight while competing with them on finished modules through Cloud Light.

Production capacity narrative

Lumentum’s public framing as of 2026-Q1:

1.6T transceivers entering production for NVIDIA and hyperscale applications Convequity 2026 deep-dive ◐
“Lumentum touts AI optics boom, $2B quarterly target, new fab and 1.6T transceiver shipments” Daily Political 2026-03-21 ◐
2028 EPS aspirational target of $30 per share, anchored on AI-photonics revenue mix Financial Content 2026-03-30 ◐

◐ Forward-revenue quarterly targets and 2028 EPS aspirations are sell-side / management commentary; not formally GAAP-guided figures. The $2B quarterly run-rate ambition first surfaced in management commentary spring 2026 — confirm against future formal guidance.

Cross-thesis links

InP EML process — the merchant supply that feeds Lumentum’s own 1.6T DR8 TRO transceivers; vertical integration on the source-laser side. See InP EML process.
NVIDIA — anchor customer for both pluggable transceivers and (via component supply) CPO modules. See 03_ecosystem / NVIDIA partnership.
Cloud Light acquisition — the M&A event that built this division. See 01_company / Cloud Light acquisition.
Marvell DSP — Marvell’s PAM4 DSP feeds many of Cloud Light’s transceiver builds; complementary partnership.
Innolight / Eoptolink — competitors-and-customers blurring; both buy Lumentum EML chips and compete with Lumentum modules in hyperscaler bid cycles.