Optical devices, optical fiber and indium phosphide substrates for data-center scale-up Report Interpretation

Overview

Nomura examines how optical scale-up in AI data centers could bring forward demand for optical devices and indium phosphide substrates. Its central view is that NPO adoption by companies competing with Nvidia’s CPO approach could create substantial incremental demand from late 2027 or early 2028, benefiting Japanese wire, cable and materials suppliers.

Core views

Nomura’s central argument is that the market should distinguish scale-out from scale-up. Scale-out replaces existing optical-transceiver connections with CPO and remains supportive because it raises the value added and density of optical links. Scale-up, by contrast, replaces copper connections within a rack with optical connections. Nomura therefore views scale-up as completely fresh demand for optical devices and fiber, whether the architecture uses CPO or NPO. The report draws on August comments from Lumentum, Coherent and Applied Optoelectronics after Japanese wire and cable shares had corrected on concerns over slow CPO adoption. Coherent said it expected demand to emerge earlier than it had expected three months previously and saw no evidence of CPO demand being delayed; customer requests were instead being pulled forward. Lumentum said Nvidia had secured substantial CPO-component supply through long-term agreements, helping explain why other companies were becoming more constructive on NPO. Lumentum’s 27 August briefing is the key catalyst in Nomura’s reasoning. The company said NPO can offer greater near-term business opportunity than CPO and revised its expected NPO-demand start from late 2028 or early 2029 to late 2027 through early 2028. Nomura interprets this as competition-driven adoption by customers outside Nvidia that want an alternative to Nvidia-led CPO. Lumentum also expects the scale-up market to begin in 2028, with CPO accounting for roughly half, mainly for Nvidia, and NPO making up the remainder for other customers. It noted that lasers per GPU in scale-up could be 5–10x those in scale-out. Nomura explains that CPO places semiconductor chips and optical engines in the same package, whereas NPO places the optical engine close to, but outside, the chip package. NPO can use slightly lower-output CW-LDs than CPO and retains longer copper electrical wiring between the optical engine and the GPU, but it still replaces part of the rack’s previously all-copper connections with optical links. This mechanism supports demand for optical devices and fiber under both approaches rather than making CPO and NPO mutually exclusive demand outcomes. The report also identifies scale-across—data-center interconnects between geographically dispersed data centers—as an underappreciated source of demand. Lumentum links this opportunity to the construction of smaller power-generation facilities and the geographic distribution of data centers to address power constraints. Nomura sees a further opportunity for Sumitomo Electric’s integrated tunable laser assemblies, used in long-distance transmission, where the company has a high market share. For Japanese suppliers, Nomura sees major opportunities for Sumitomo Electric, Furukawa Electric and Fujikura as CPO and NPO expand. Sumitomo Electric’s high-output CW-LDs were in final sample certification as of its 22 May briefing, with pilot shipments scheduled for the end of FY2027/3 and a ramp in FY2028/3. Nomura believes its opportunity extends beyond optical devices to optical IC connection components, NPO demand and scale-across applications. The report also notes Sumitomo Electric’s discussions with Nvidia and multiple other customers. InP-substrate availability is a separate demand and supply-chain thread. Lumentum said it was continuously seeking InP-substrate suppliers because its long-term arrangements with a Japanese supplier and AXT might not be sufficient. Nomura sees this as a potential new opportunity for Sumitomo Electric and JX Advanced Metals. Sumitomo Chemical began mass production and sales of InP epiwafers on 31 August and targets around ¥10bn of sales by the mid-2030s. Nomura considers the profit impact on Sumitomo Chemical limited relative to its ¥208.4bn of FY2026/3 core operating profit, but views the launch as evidence of industry demand for stable InP-epiwafer procurement. It estimates, without confirmation from Sumitomo Chemical, that JX Advanced Metals may supply InP substrates for that production. Nomura’s data-center demand framing indicates the possible scale of the opportunity: GPU-cluster expansion could raise fiber and wiring-material needs by about 1.5x once clusters exceed 130,000 GPUs; bandwidth upgrades from 400G to 3.2T could raise fiber requirements by 2–4x depending on laser technology; and full optical conversion of in-rack wiring could require about 10x more fiber per GPU. The report also cites Corning’s view that passive-photonics opportunities could amount to $10bn by 2030.

Analysis framework

Nomura combines recent company earnings and investor-briefing comments with announced long-term contracts and a supply-chain mapping of optical devices, fiber, laser components and InP substrates. It compares the demand implications of CPO, NPO, scale-out, scale-up and scale-across, then links these architectural changes to specific Japanese suppliers and production timelines.

Methodology notes

• Industry AnalysisSupply-demand framework

  Optical-component demand and supply analysis

  The report uses company demand outlooks, customer contracts, production plans and substrate sourcing comments to assess when demand may emerge and where supply constraints could create opportunities.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Optical-device and InP-substrate supply-chain mapping

  Nomura traces how data-center optical adoption can flow through optical devices, fiber, connection components, epiwafers and InP substrates to Japanese suppliers.

• Industry AnalysisVolume-price decomposition

  Fiber and optical-link volume drivers

  The report separates demand drivers such as GPU-cluster size, bandwidth migration and conversion of copper links to estimate changes in required fiber and optical connectivity.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Sumitomo Electric Industries [5802]

  Nomura sees major opportunities from CPO and NPO optical devices, optical IC connection components, InP substrates and scale-across transmission.

  Strengths

  High-output CW-LDs were in final sample certification; the company has a high share in iTLAs and is in discussions with Nvidia and other customers.

  Weaknesses

  CPO optical-device sample accreditation gave the impression it was slightly behind US rivals.

  Comparison

  Nomura compares its optical-device opportunity with US suppliers including Lumentum and Coherent.

  Risks

  Timing of CPO and NPO deployment and the company’s ability to convert qualifications and customer discussions into volume shipments.

• Furukawa Electric [5801]

  Expected beneficiary of rising optical-device, fiber and cable demand from CPO, NPO and data-center expansion.

  Strengths

  Nomura includes it among the three Japanese wire and cable companies with substantial opportunities from increased optical adoption.

  Comparison

  Considered alongside Sumitomo Electric and Fujikura as a Japanese wire and cable beneficiary.

  Risks

  Demand timing remains tied to the pace of data-center optical deployment.

• Fujikura [5803]

  Expected beneficiary of increasing optical fiber and cable demand as data-center connections become more optical.

  Strengths

  Nomura includes it among the three Japanese wire and cable companies positioned for CPO and NPO-related demand.

  Comparison

  Considered alongside Sumitomo Electric and Furukawa Electric as a Japanese wire and cable beneficiary.

  Risks

  Demand timing depends on deployment of optical architectures in data centers.

• JX Advanced Metals [5016]

  Potential beneficiary of increased InP-substrate demand and possible supply to Sumitomo Chemical’s InP-epiwafer production.

  Strengths

  Planning to increase InP-substrate output at its Isohara plant and Hitachinaka works.

  Weaknesses

  The possible supply relationship with Sumitomo Chemical is Nomura’s estimate rather than a company-confirmed fact.

  Comparison

  Considered with Sumitomo Electric as a possible Japanese supplier to address additional InP-substrate demand.

  Risks

  Supplier relationships and demand conversion remain unconfirmed.

Key data

• NPO demand timingLate 2027 through early 2028Lumentum’s revised expectation, versus its prior expectation of late 2028 or early 2029.
• Scale-up market timing2028Lumentum expects the market to begin then, with CPO roughly half and NPO the remainder.
• Lasers per GPU in scale-up5–10x scale-outLumentum estimate cited by Nomura.
• Optical fiber per GPU under full in-rack optical wiring10x current levelNomura’s presentation of Corning briefing materials.
• Bandwidth-related fiber increase from 400G to 3.2T2–4xDepends on the single-wave laser technology used.
• Sumitomo Chemical InP epiwafer sales targetAround ¥10bn by the mid-2030sAnnounced on 31 August.
• Sumitomo Chemical core operating profit¥208.4bn in FY2026/3Nomura uses this to characterize the epiwafer initiative’s limited group-profit impact.
• Passive-photonics opportunity$10bn by 2030Potential new business opportunity cited from Corning materials.

Impact & implications

Nomura believes the shift from copper to optical links within racks expands the addressable market beyond conventional scale-out. It sees NPO as an especially important near-term route to incremental demand, while scale-across can add longer-distance transmission opportunities. Stable InP-substrate procurement is presented as a critical supporting requirement for this expansion.

Risks

• Slower-than-expected deployment or uptake of CPO and NPO could defer the anticipated optical-demand expansion.
• Lumentum indicated that existing long-term arrangements with a Japanese supplier and AXT may not provide sufficient InP-substrate supply, highlighting supply availability as a constraint.
• Nomura’s inferred supply links, including a possible JX Advanced Metals relationship with Sumitomo Chemical, are not company-confirmed.

What to watch

• Progress in NPO deployment, which Lumentum expects from late 2027 through early 2028.
• Sumitomo Electric’s pilot shipments planned for the end of FY2027/3 and ramp-up in FY2028/3.
• Company updates from Lumentum, Coherent, Applied Optoelectronics, Corning and AXT on optical-device and InP-substrate demand.
• Further long-term optical-product and raw-material procurement agreements between hyperscalers, optical suppliers and substrate producers.