Nomura: CPO Commercialization Enters Supply-Demand Validation Phase; Market Underestimates Progress
Nashnova编辑部
Nomura compiled management signals from five optical-component companies and concluded that CPO — co-packaged optics — is shifting from 'can the technology work?' to 'who can lock capacity and clear qualification?' The market has not yet priced this transition adequately.
Why does scale-up multiply optical-component value by an order of magnitude?
Lumentum quantified it for the first time: in scale-up networks — high-bandwidth interconnects between GPUs inside a rack — laser demand per GPU is 5–10× that of scale-out (rack-to-rack connections).
This means → scale-up is not about moving pluggable modules inside the rack. It is a full redistribution of the interconnect architecture — external laser sources, fiber arrays, connectors, and silicon-photonics packaging all upgrade together.
Corning's data corroborates: once intra-rack connections go fully optical, fiber count per GPU could reach 10× current levels. When a cluster exceeds 130,000 GPUs and the network expands from two tiers to three, demand rises another ~50%.
Corning also flagged that passive optical components — parts that guide light signals without needing power — connecting external lasers, optical engines, and silicon-photonics chips create an entirely new market, potentially worth ~$10 billion by 2030.
How tightly have the timelines converged?
Multiple companies point to nearly the same cadence: 2026 for validation and pilot production, 2027 for supply expansion and customer onboarding, 2028 for volume ramp.
AAOI began small-batch CPO production in Q1 2026, with full launch targeted for H2 2026–2027. It plans to raise ELSFP monthly capacity to 400,000 units by 2028; expansion lead time is 21–24+ months.
Coherent expects ultra-high-power CW lasers (CW-LD — chips that emit a continuous, stable light signal) for scale-out CPO to ramp in Q4 2026, with scale-up lasers following in H2 2027. Management stressed: no demand delays observed — customers are actually pulling schedules forward.
Lumentum sees CPO laser-chip demand rising in H2 2027; scale-up CPO shipments begin in the same period and enter a clearer ramp in H1 2028. Sumitomo Electric's high-power CW-LDs are in final sample qualification, with trial supply planned by FYE March 2027 and ramp through FYE March 2028.
What do long-term agreements and InP-substrate lockups signal?
Roughly 75% of Lumentum's business is covered by long-term agreements; Sumitomo Electric's optical-device LTA share is rising from ~50% toward 60–70%. This means → in an industry with long capacity-expansion cycles, the leaders have effectively pre-sold their output.
In plain terms = LTAs serve three purposes: guarantee supply of critical components, underwrite suppliers' capex, and raise the barrier for competitors to enter on short notice. But an LTA is not a revenue guarantee — its real value depends on prepayments, minimum purchase volumes, and fulfillment pace.
InP substrates — indium phosphide wafers used to make high-power laser chips — have been elevated from a commodity input to a strategic resource. Lumentum signed a seven-year InP-substrate LTA with a Japanese supplier and holds a substantial share of that supplier's capacity; it has also signed LTAs with three InP suppliers total, is negotiating with others to diversify away from China-sourced supply, and has extended its AXT agreement through 2031.
Based on the unusual contract length, Sumitomo Electric's previously disclosed 3–7-year agreements, and its global market position, Nomura infers the Japanese LTA counterparty is most likely Sumitomo Electric. Nomura explicitly states this is an analyst inference, not a confirmation by either party.
Will CPO be winner-take-all?
Lumentum expects that when the scale-up market launches in 2028, CPO will account for roughly half (primarily Nvidia-aligned architectures); the other half may be served by NPO — near-package optics, where optical modules sit close to the chip package but are not co-packaged onto it. Key customers' NPO products may ship one quarter ahead of CPO.
This reflects a market where customers choose paths based on architecture fit, reliability, and supply-chain maturity — CPO will not suddenly replace all pluggable modules at a single inflection point.
In plain terms = high-power lasers, InP substrates, fiber, and connector assemblies are used in both CPO and NPO. Demand hinges on optical-interconnect penetration rates and port counts, not on which packaging label ultimately "wins."
What exactly does Nomura think the market is underestimating?
Nomura's core thesis: the next phase of outperformance depends less on whether a company has a CPO story and more on customer-qualification progress, supply security, volume-production yields, and multi-product synergies.
This means → these four dimensions are precisely the ones current market pricing has not fully reflected. Near-term share prices may front-run long-dated potential, but earnings delivery still requires clearing four gates: qualification → capacity expansion → yield ramp → customer deployment.
Among the three Japanese wire-and-cable companies Nomura covers, Sumitomo Electric is seen as the most likely beneficiary — because it operates simultaneously across InP substrates, optical devices, optical-interconnect assemblies, and fiber materials.
Content is for reference only, not financial advice.