報告_Nomura_玉晶光3406_光纖連接器CPO_20261005

PDF 原檔：報告_Nomura_玉晶光3406_光纖連接器CPO_20261005_original.pdf

圖片清單（已驗證 2026-10-07）

ingest 時建立的圖片索引，是 lib/ 嵌圖的唯一真相來源；<40KB 者未 Read、預設 logo／小圖。

原始內容

NOMURA

Genius Electronic Optical 3406.TW 3406 TT

EQUITY: TECHNOLOGY

Fiber connector and CPO an ideal leverage of its capabilities (Erratum)

Quantifying passive fiber connector and CPO opportunities; reiterate Buy

Action: Maintain Buy; raise TP to TWD1,415, implying ~34% upside

We expect AI optical interconnect and communication opportunities (e.g. co-packaged optics fiber array unit [CPO FAU], passive fiber connectors) to transform the traditional optical lens industry over the next 5 years, and believe Taiwanese handset camera lens makers (e.g. Largan [3008 TT, Buy] and Genius) possess inherent advantages to cater to these new markets, given their high-precision molding+plastic injection, optical design, active alignment, optical coating and inspection capabilities, as well as clean-room facilities and sufficient FCF from cash-cow handset lens businesses.

In addition to holding the position of Apple's (AAPL US, Not rated) No. 2 lens supplier, we believe Genius has obtained ferrule order for passive fiber connectors and will start shipping by end-2026F or early-2027F, as per our channel checks (Genius declines to comment, however). Genius appears well-positioned to supply various CPO FAU components (lens, prism, V groove, FA, etc.) and may potentially handle FAU assembly if requested by customers. We estimate the passive fiber connector and CPO-related new businesses will contribute <3%/16-20% earnings for Genius in 2027F/2028F (details in Fig. 10). Yet, given the high dollar content and long-term growth potential of CPO/FAU related products over the next five years, we believe optical lens makers with inherent advantages in this new field deserve stock re-rating; we thus raise our target P/E to 25x (near the high end of the historical range of 10-30x since 2020), from 17x. We reiterate Buy, and raise our TP to TWD1,415, based on 25x 2028F EPS of TWD56.61 (vs. prior TP of TWD568, based on 17x 2025F EPS of TWD33.4, set two years ago). Genius trades at 25x 2027F P/E.

Apple: Sunny's market share loss and continued camera upgrades in 2027-28F

Sunny (2382 HK, Neutral) earlier faced some yield issues with the new variable aperture (VA) main camera lens for the iPhone 18 Pro series, and this could benefit Largan (more directly) and Genius (indirectly) in the competitive landscape. It remains uncertain whether Genius will be introduced as a third lens supplier to iPhone Pro' main camera in the future, and this poses an upside risk to our forecasts. However, given the headwinds of rising memory costs and the potential cannibalization of the 18 Pro series by the iPhone Duo, we forecast slower iPhone lens shipments from end-2026F to 1H27F, before the 20 th anniversary iPhone (Pro Max) and the second-gen iPhone Duo are introduced in 2H27F. We expect further upgrades to the iPhone Pro main camera in 2H27F, and introduction of a new periscope camera for the Pro Max in 2H28F.

(Erratum: This note replaces an earlier published note which incorrectly focused on specific fiber connector. We apologise for any inconvenience this may have caused.)

Source: Company data, Nomura estimates

Implied upside

+34.1%

Market Cap (USD mn)

3,740.5

ADT (USD mn)

110.2

Relative performance chart

Source: LSEG, Nomura

Research Analysts

Taiwan Technology

Anne Lee, CFA - NITB anne.lee@nomura.com +886(2) 21769966

Eric Chen, CFA - NITB

eric.chen@nomura.com +886(2) 21769965

China Technology

Donnie Teng - NIHK

donnie.teng@nomura.com +852 2252 1439

Production Complete: 2026-10-05 18:29 UTC

Key data on Genius Electronic Optical

Cashflow statement (TWDmn)

Source: Company data, Nomura estimates

Source: Company data, Nomura estimates

Company profile

Genius is the second source in Apple's lenses, and is gaining share in recent years. VR is a new growth engine for the company, with Facebook and Apple being its key VR customers.

Valuation Methodology

Our TP of TWD1,415 is based on 25x 2028F EPS of TWD56.6. Our target P/E of 25x is near the high end of Genius' historical range of 10-30x since 2020. The benchmark index for this stock is TAIEX.

Risks that may impede the achievement of the target price

Downside risks are: 1) worse-than-expected iPhone demand, 2) rising ASP pressure due to competition, 3) worse-than-expected progress in market-share gain in iPhone rear cam lenses, and 4) slower-than-expected penetration into other Android customers.

ESG

The company does not disclose any ESG information. However, based on industry features, we believe Genius is not a labor-intensive corporation and it should not cause severe environmental issues.

Optical communication - an ideal leverage of lens makers' capabilities

Optical interconnect - from MPO to MMC and contact-less fiber connector

Multi-fiber push-on (MPO) is the current standardized physical-contact optical interconnect architecture designed to aggregate high-density ribbon fibers (typically 8, 12, 16, 24 or 32 cores per ferrule) into a single compact interface. MPO connectors leverage precise physical contact enforced by a high mechanical spring force (around 10-20N) and steel guide pins for alignment to achieve ultra-low insertion loss (0.20-0.35dB) across clean fiber end-faces.

The micro multi-fiber connector (MMC) is another proprietary physical-contact optical interconnect standard beyond MPO developed by US Conec (unlisted) for high-density AI interconnects to directly address the server faceplate and panel space constraints. The MMC introduces a very small form-factor (VSFF) multi-fiber design that utilizes a miniaturized TMT ferrule (patented by US Conec; Fig. 1), and according to US Conec, the width of the TMT ferrule is similar to that of conventional MT ferrule yet with ~30% smaller in thickness and ~50% shorter in length to shrink the physical footprint to roughly one-third the size of an MPO. The spatial reduction consequently unlocks up to a 3x density advantage in patch panel ports and transceiver faceplates compared to standard MPO footprints (Fig. 2). Meanwhile, the MMC delivers a crucial density jump while preserving high optical performance, offering low insertion loss and supporting up to 250um pitch multi-fiber ribbon arrays (including 12-, 16-, 24-, and 32-fiber configurations) to ensure interoperability with MPO ecosystems. The MMC also improves operational serviceability through US Conec's DirectConec push-pull latching mechanism to allow technicians to plus and unplug individual cables even at a highly populated panel without external retraction tools.

Fig. 1: TMT ferrule vs. conventional MT ferrule

Source: US Conec, Nomura research

Although MPO and MMC are the most popular standards of optical fiber connectors nowadays, they both rely on direct physical contact between ferrules and thus require higher spring pressure and precision alignments when connecting the cables. Also, physical contacts leave MPO and MMC's exposed optical interfaces inherently susceptible to wear and tear over repeating mating cycles as well as potential signal degradation from dust and debris contamination (and therefore needs frequent cleaning and inspection).

Expanded beam optical (EBO), for example, is an innovative passive fiber connector technology promoted by an industry group 'EBO Multi-Source Agreement Group (EBO MSA)' that could reshape optical interconnect technologies by addressing pre-mentioned challenges of MPO or MMC, since this type of connector replaces traditional physicalcontact end-face matings with non-contact, lens-based beam expansion (Fig. 4). The passive fiber connectorutilizes integrated total internal reflection (TIR) mirrors within precision polymer ferrules. As the light exits an optical fiber core and enters the ferrule, it

Fig. 2: MMC has a smaller footprint than MPO to enable a 3X optical interconnect density

Source: US Conec, Nomura research

strikes the internal TIR mirror that collimates and expands light beams up to 10x their core diameter before travelling through an intentional, tiny air gap to the mating connector. An identical, paired TIR mirror is integrated within the receiving ferrule to capture the expanded beam and focuses it back into the receiving fiber core. According to the EBO MSA, the ferrule is vapor-coated with a special anti-reflection coating that efficiently reduces the reflection of light beams (Fig. 5).

Fig. 3: An EBO ferrule

Source: EBO MSA, Nomura research

Fig. 5: A cross-section illustration of EBO

Modern EBO Technology for data-centers

Source: EBO MSA, Nomura research

Although one of the underlying issues with contact-less passive fiber connector is a higher insertion loss (0.35-0.75dB), the contact-less passive fiber connector system still provides a few advantages compared to traditional physical-contact connectors:

• Physical-contact connectors require frequent inspection and cleaning cycle before mating, while the contact-less connectoris nearly immune to airborne particle contamination thanks to the embedded lens system and no direct surface contacts. An easier handling with reduced maintenance efforts therefore significantly lowers the installation time (as much as by 85%, according to industry statistics) and labor cost. ·
• Contact-less passive fiber connector can ensure consistent signal integrity across thousands of mating cycles due to contact-less designs to avoid face-wear and microscratching. By contrast, physical connectors grind against each other and ultimately lacks long-term durability. ·
• Traditional connectors require precise fiber-to-fiber alignments, whereas the contactless passive fiber connector system can be used in a hermaphroditic condition (socket/connector agnostic) given identical connector at both ends. As such, the risk of installation error is largely eliminated, even without the need for fiber-optic ·

Fig. 4: Connection of two EBO ferrules

Source: Rosenberger, Nomura research

specialists.

• The contact-less fiber connector only demand a fraction (e.g. one-tenth) of the high spring pressure that compresses glass end-faces together in physical connectors, thereby minimizing mechanical strains on chassis frames. ·
• The contact-less, high density, blind mating, and low spring pressure features make the connector potentially an ideal optical connector for optical back-plane interconnect applications (similar to the function of orthogonal backplane or midplane in nVidia's [NVDA US, Not rated] Kyber architecture), in our view. ·

Fig. 6: EBO ferrule offerings (12F/16F)

Source: EBO MSA, Nomura research

The EBO MSA was established in March 2026 to promote EBO as an open, industry-wide standard for AI cluster infrastructure. The primary objective is to standardize EBO connector mechanical form factors, latching mechanisms, and optical fiber interface dimensions to allow hyperscalers to source interchangeable EBO connectors from multiple vendors rather than relying on one single supplier. According to EBO MSA, there are 9 end-user companies (including Microsoft [MSFT US, Not rated], Meta [META US, Not rated], nVidia, AMD [AMD US, Not rated], etc.) and 48 vendors in the membership by the end of September 2026 (Fig. 7). Members vote on specific connectors to develop and contribute design specifications - there are 18 types under discussion, and four are on the roadmap:

• 128f (8x16f) design specification is underway. ·
• 16f MPO design is under discussion on specification. ·
• 16f VSFF and external laser small form factor pluggable (ELSFP) are planned, and we note that EBO development for ELSFP could be challenging since high-power laser could damage the AR coating and polymer ferrules. ·

Fig. 7: EBO MSA member companies

Source: EBO MSA, Nomura research

Of course, all the key members of EBO MSA are potential users of EBO connectors. But for more concrete industry developments, Microsoft was the first announced hyperscale cloud provider to deploy EBO technology in its Azure cloud (link). Microsoft's early use of EBO technology has shown the potential to reduce network deployment timelines in certain environments. The technology has also demonstrated strong signal performance in live data center conditions, where dust exposure and routine handling are typical during installation and maintenance. The technology will help Microsoft reduce the need for frequent cleaning and inspection while supporting reliable optical performance in dense, high-volume deployment environments.

Meta joined EBO MSA in May 2026, and from our industry checks, we think Meta is likely evaluating the technology for future ASIC mass-volume optical interconnect demand.

Why optical plastic lens companies have advantages in producing ferrules for passive optical connectors?

From MPO to MMC, the fiber density increases to 3x, and we notice that the precision level requirements for MMC TMT ferrules also increase substantially versus MPO MT ferrules. The high precision levels (<5um, or even <3um) of MMC TMT ferrules usually exceed the traditional optical connector makers' plastic injection capability (usually >10um) and traditional connector makers need to invest new, higher precision level tools to deal with those higher precision requirements. However, those precision levels are about the right levels (or even lower-end than) which plastic camera lens makers can do without much new investment in machine tools. This is why we think plastic lens makers have advantage in the advanced passive optical connectors. The reason why we do not see many lens makers enter MMC ferrule market at this moment is because it is a patented technology by US Conec, and US Conec only licenses to limited suppliers, such as Sumitomo (5802 JP, Buy), etc. We estimate the production difficulty and price of passive fiber connector ferrule is around the levels of mid-range handset camera lenses.

Our base assumption for Genius' fiber connector ferrule revenue contribution is about 2-3% for 2027F and 2028F, by assuming an ASP of USD0.5-1 with 20-35mn units annual volume. We assume the GM for this new product to be higher than corporate average.

The upside of fiber connector market will depend on the adoption and penetration of the technology into hyperscalers' AI optical interconnect roadmap. By our estimate, every rack (containing 60-80 units of GPUs or AI ASICs), if adopting fiber connector for optical scale-up for intra-rack or cross-rack interconnect technology, will potentially require 600-1,000 units of fiber connector ferrule components (300-500 pairs). Our base volume assumption for Genius' fiber connector shipments require one major CSP to adopt fiber connector in one major AI platform, plus global sampling, yield loss, trial run demand and necessary inventory buffer. If there are more than one major AI projects adopt fiber connector in 2028F, it will represent potential upside to our forecasts.

CPO opportunities could be bigger than fiber connector ferrule, but less clear at this moment

Compared to fiber connector ferrule, CPO opportunities to Genius remain more uncertain in details, such as what products to offer, what are the business models, and who are intermediate customers and/or end customers. Not to mention, the ramp-up of CPO market itself also has high uncertainties.

We believe Genius is working with FOCI (3363 TT, Not rated) in supplying key components for CPO FAU, such as V groove, prism, and/or fiber array (FA). It is also likely in talk with other FAU makers for various component supply (including lens, V groove, prism, FA, etc.). These FAU makers tend to do FAU final assembly in-house, and hence, may treat Genius as a component supplier to fill its internal capability or outsourcing partners' gap. Genius' flexibility in various component offering for CPO FAU could make it an easier partner for FAU customers, versus Largan (it tends to do more FA or FAU, rather than just components), and may see revenue contribution earlier than Largan. However, we assume USD25-35 ASP for Genius' CPO FAU component offering in our base assumptions, given that it may supply more components rather than doing finished FAU products (ASP USD>100).

In the other side, we also see opportunities for Genius to get FAU module business from customers like Corning (GLW US, Not rated), but the progress of Genius to do such full assembly lines could be slower than Largan, as Genius still relies more on external equipment supply, unlike Largan having strong internal equipment design and manufacturing capability.

We assume CPO FAU to contribute Genius' EPS by <1% in 2027F and 17% in 2028F. For detailed assumptions, please refer to Fig. 10.

Fig. 8: The FAU typically includes optical components such as the V-groove, mirror and optical fibers

Source: TSMC, Nomura research

Fig. 9: Prism/MLA is the mirror directing the beam to PIC

Source: AGC, Nomura research

Fig. 10: Our estimates of CPO market and earnings contribution to Genius

Source: Nomura estimates

Fig. 11: Genius: Earnings forecast revisions

Source: Nomura estimates

Fig. 12: Genius: Quarterly financial forecasts

Source: Company data, Nomura estimates

Fig. 13: Genius: P/E bands

Fig. 14: Genius: P/B bands

Source: TEJ, Nomura estimates

Source: TEJ, Nomura estimates

Appendix A-1

This report has been produced by Nomura International (Hong Kong) Ltd., Taipei Branch (NITB), Taiwan.

See Disclaimers for Nomura Group entity details.