Skip to content
primarysourced Photonics sector Lumentum
LITE
~7 min read · 1,559 words ·updated 2026-04-29 · confidence 58%

Hyperscaler datacom customers

Lumentum’s hyperscaler customer base has two distinct surfaces: component-level supply (InP EML laser chips that flow into module-builders’ transceivers and ultimately into hyperscaler data-centers, often without disclosed end-customer attribution) and module-level direct supply (finished pluggable transceivers shipped from Lumentum directly to the hyperscaler). The module-level direct surface only exists post-Cloud Light (acquisition closed November 7, 2023).

A persistent caveat: hyperscalers rarely confirm component-supplier relationships in their own 10-K disclosures, and Lumentum likewise cloaks named-customer revenue in its 10-K via redacted “Customer A / Customer B” labels. Most hyperscaler-attribution claims below are therefore ◐ industry-attributed via trade press (LightCounting, Cignal AI, Dell’Oro, 650 Group) rather than ✓ verified-primary.

Customer matrix (post-Cloud Light)

HyperscalerComponent supply (InP EML chips)Module supply (finished transceivers)Confidence
NVIDIALasers for Spectrum-X, Quantum-X CPO, GPU-cluster opticsMulti-billion purchase commitment, Mar 2026✓ verified-primary (LITE 8-K + NVDA press release)
Microsoft AzureEML chips into Azure-bound transceiversCloud Light heritage direct customer◐ industry-attributed (SCMP / Cignal AI)
MetaEML chips into Meta-bound transceiversCloud Light heritage direct customer◐ industry-attributed
GoogleEML chips into Google-bound transceiversLimited; Google has internal optics investments◐ industry-attributed; LITE has confirmed shipping to “all three” major hyperscalers
Amazon AWSEML chips into AWS-bound transceiversLimited; AWS uses internal silicon photonics + multiple merchant suppliers◐ industry-attributed
Oracle Cloud / xAI / CoreWeave (neoclouds)EML chips flow indirectly via NVIDIA Spectrum-X reference designsn/a — these neoclouds buy NVIDIA reference systems⚠ inferred

NVIDIA — the load-bearing anchor

NVIDIA is the only hyperscaler-class customer for which Lumentum has disclosed a primary-source-confirmed direct relationship. The March 2, 2026 8-K and the joint press release establish:

  • A “multi-billion purchase commitment” (exact dollar amount not disclosed)
  • Future capacity access rights for advanced laser components
  • Multi-year, non-exclusive, technology-neutral
  • Tied to Greensboro NC fab build-out for InP-laser production

Lumentum’s existing public product-page disclosures confirm laser supply into NVIDIA’s silicon-photonics-based switches (Spectrum-X, Quantum-X CPO) — see datacom transceivers. Detailed treatment in nvidia partnership.

Microsoft Azure — Cloud Light heritage

Cloud Light Technology Limited, founded 2018 in Hong Kong with manufacturing in Dongguan and SE Asia, was widely reported in industry trade press as a Microsoft-centric high-speed transceiver supplier in the years before its November 2023 acquisition by Lumentum. ◐ This characterization comes from:

  • South China Morning Post coverage of the acquisition (October 2023): “AI seen as driving further growth at Hong Kong fibre optics start-up Cloud Light after acquisition by US firm”
  • Cignal AI / LightCounting market notes characterizing Cloud Light as a top-3 800G transceiver supplier with concentrated hyperscaler exposure
  • Lumentum’s own filings disclose customer concentration only as numbered customers — neither Microsoft nor Meta is named in the LITE 10-K

Lumentum disclosed that Cloud Light booked “more than $200M revenue in the past 12 months” pre-acquisition, with nearly all derived from 400G+ transceivers and over half (in the latest pre-deal quarter) from 800G modules. ✓ verified-primary via LITE acquisition press release. The hyperscaler-customer concentration is confirmed (the customer base was concentrated to a small number of hyperscalers); the identity of those hyperscalers is ◐.

Meta — Cloud Light heritage

Meta is consistently named alongside Microsoft as a Cloud Light heritage customer in industry trade-press coverage (LightCounting research notes, optics.org coverage). ◐ The relationship is described as flowing through Meta’s optical-transceiver consumption for the company’s AI-training-cluster build-out post-2023 (the GenAI capex inflection that drove Meta’s 800G transceiver consumption to multi-million-unit annual volumes by 2025).

Independent confirmation surfaces:

  • LightCounting’s research note “November 2023 Lumentum is back in. Intel is out.” (cited in industry summaries) framed Cloud Light as restoring Lumentum’s hyperscaler module-supply position
  • Meta’s own infrastructure disclosures (open-compute, AI-cluster announcements) reference 800G transceiver consumption but never name suppliers

Google — limited direct, indirect via NVIDIA reference

Google has been the most strategic-internal of the major hyperscalers in optical-transceiver supply, with disclosed internal silicon-photonics work (the Mineral / Apollo / Argos optical-circuit-switch infrastructure) and a documented preference for multi-sourcing pluggables at the module level. Lumentum’s disclosed shipment to “all three” major hyperscale cloud customers in fiscal Q4 2025 commentary is read by analysts as Microsoft + Meta + Google or Microsoft + Meta + Amazon — the exact identity of the third is ⚠ inferred.

For NVIDIA-anchored AI-cluster deployments inside Google Cloud (the GCE H100 / H200 / GB200 instances), Lumentum’s lasers flow indirectly via NVIDIA reference designs — meaning Google Cloud effectively consumes Lumentum InP EMLs even where Google’s hyperscaler-grade in-house networks do not.

Amazon AWS — limited direct

AWS has been the most aggressive hyperscaler in vertically integrating optical-transceiver supply — most notably via internal silicon-photonics teams (the Annapurna Labs Nitro extension into networking ASICs, the AWS-internal optical-transceiver effort widely reported but not publicly disclosed). AWS multi-sources at the module level among Innolight, Eoptolink, and (post-Cloud Light) Lumentum. ◐ — no LITE disclosure names AWS.

For AWS’s NVIDIA-anchored Trainium-vs-NVIDIA mixed compute, Lumentum InP EMLs again flow indirectly via NVIDIA reference designs.

Pre/post Cloud Light contrast

PeriodHyperscaler exposureChannelRevenue attribution
Pre-Cloud Light (FY2015-FY2023)Indirect — chips into module-buildersComponentModule-builder customers (Finisar/Coherent, Innolight, Eoptolink, etc.) booked the hyperscaler revenue
Post-Cloud Light (Nov 2023 onward)Direct module supply to top-3 hyperscalers + indirect chip supply to allModule + componentLumentum books hyperscaler-direct module revenue + downstream chip revenue
Post-NVIDIA (Mar 2026)Direct chip and module supply to NVIDIA-anchored AI clustersModule + componentLumentum books direct NVIDIA revenue + retained third-party hyperscaler revenue

The Cloud Light deal moved Lumentum up the value chain from a component supplier (selling EML chips to module-builders for $X / chip) to a finished-module supplier (selling 800G transceivers for $1,000-$2,000 each, with the EML chip embedded). This is the financial arithmetic that makes the Cloud Light acquisition so important: it more than doubled Lumentum’s cloud-data-center infrastructure revenue in the 12 months post-close. ✓ verified-primary via LITE press release.

Customer concentration disclosed in 10-K

Lumentum’s fiscal 2025 10-K (filed August 2025) disclosed customer concentration in the standard SEC-required form: customers representing 10% or more of net revenue, identified as “Customer A,” “Customer B,” etc. without naming. ◐ Trade-press attribution maps:

  • One unnamed hyperscaler reportedly hit 29% of Lumentum revenue in a single quarter of FY2024 — consistent with a single Microsoft-centric or Meta-centric Cloud Light customer
  • Ciena and Cisco are widely cited as historical 10%+ customers on the telecom-OEM side (see telecom oems)
  • The fiscal 2025 10-K reflects a transition: the historical Ciena/Cisco concentration is being replaced by hyperscaler concentration as Cloud Light scales

Implications for the AI-photonics thesis

For investors, the hyperscaler-customer surface is the least transparent layer of the Lumentum thesis — and the part most exposed to misattribution. The clearest signal is the NVIDIA partnership, which is fully primary-sourced. Everything else around Microsoft, Meta, Google, AWS direct-supply attribution is ◐ trade-press confidence and should be treated as such until Lumentum or a hyperscaler confirms in primary disclosure.

The structural take-away: Lumentum’s hyperscaler exposure is multi-vendor concentrated — they ship to all the big hyperscalers via NVIDIA’s reference designs (and so participate in every NVIDIA-anchored AI cluster regardless of the host hyperscaler’s identity), and they also ship direct-to-Microsoft and direct-to-Meta via the Cloud Light channel. This dual-channel structure makes hyperscaler customer concentration risk less binary than it appears: a single hyperscaler’s transceiver-consumption pause hurts, but the NVIDIA-aligned indirect surface continues regardless.

Caveats and confidence levels

  • All hyperscaler-direct relationships except NVIDIA carry ◐ confidence
  • Cloud Light’s pre-acquisition customer roster has been described in trade press but never confirmed in primary disclosure by either Lumentum, Cloud Light (pre-deal), or any hyperscaler
  • Lumentum’s “all three major hyperscale customers” statement in Q4 FY2025 is verifiable from earnings-call transcript but does not name the three
  • The 10-K customer-concentration table names customers by letter, not by identity
  • nvidia partnership — the only ✓ verified-primary hyperscaler-class direct relationship
  • cloud light acquisition — the $750M deal that created the direct-to-hyperscaler module channel
  • datacom transceivers — the 400G/800G/1.6T product portfolio shipped to these hyperscalers
  • competitors — Innolight and Eoptolink, which compete head-to-head with Cloud Light at hyperscalers
  • telecom oems — the historical Ciena/Cisco/Nokia revenue base being replaced by hyperscaler concentration
  • customers — Marvell’s hyperscaler set + DSP that pairs with Lumentum’s transceivers in the module
  • customers photonics — TSEM PH18 customer set; module-builders consuming both InP from Lumentum and silicon-photonics PICs from TSEM

Sources