-+ 0.00%
-+ 0.00%
-+ 0.00%
Northeast Securities: NPO is the new main line of AI optical interconnection that balances performance benefits and engineering feasibility as a core path
Share
Listen to the news

The Zhitong Finance App learned that Northeast Securities released a research report saying that NPO is not a simple CPO transition plan, but a core path that takes into account performance benefits and project viability in the current industrial stage. As a new battleground for domestic computing power, leading cloud vendors are speeding up commercialization. Domestic cloud vendors, equipment vendors, chip manufacturers, and optical communication companies have already participated in the formulation of standards such as OPENNPO, and Chinese companies are expected to further switch to co-defining new optical interconnection architectures. NPO is the “Tao” in the field of optical connectivity. Cloud vendors at home and abroad are simultaneously embracing NPO. It is also another major opportunity for China's industrial chain and continues to recommend relevant targets.

The main views of Northeast Securities are as follows:

NPO takes into account performance, yield and maintainability, and is currently the next-generation optical interconnection solution with the most implementation conditions

As SerDes evolves from 112 Gb/s to 224 Gb/s, long electric channels with traditional pluggable architectures face loss, power consumption, and heat dissipation constraints, and the photoelectric conversion position continues to migrate to the near end of switching chips and computing power chips. The CPO integrates the ASIC and optical engine in the same package, and has higher bandwidth density and energy efficiency limits, but places higher requirements on advanced packaging, joint yield, thermal management, and system maintenance; NPO deploys an independently packaged optical engine near the ASIC to exchange a smaller electrical performance gap for higher manufacturing yield and maintainability. Therefore, NPO is not a simple CPO transition plan, but a core path that takes into account performance benefits and engineering viability in the current industrial stage.

AI clusters continue to expand, and scale-up becomes the largest new application scenario for optical connectivity

Long context, multi-modality, MoE, and complex inference continue to amplify inter-chip synchronous communication requirements, and the competitive focus of computing power systems is shifting from peak performance of a single card to effective computing power in clusters. Supernodes have evolved from tens of cards to hundreds of cards and thousands of cards, and the scale-up interconnect range also extends from within the aircraft and within the rack across racks. The bandwidth of a single XPU interconnect is about 9-10 times that of scale-out. Compared with scale-out networks that already widely use optical modules, Scale-Up currently still relies heavily on PCBs, copper cables, and high-speed connectors, and the optical interconnect penetration rate is low; as high-speed copper interconnect gradually approaches the upper limits of distance, power consumption, and wiring density, “light to copper retreat” will extend from cabinet to cabinet, device, and near end of the chip, further opening up the market limit for optical connectivity.

NPO has become a new battleground for domestic computing power, and leading cloud vendors are speeding up commercialization

Domestic computing power also needs to release effective computing power by expanding the tightly coupled computing domain and improving system interconnection efficiency. NPOs do not need advanced packaging of ASICs and optical engines, and are more suited to the current domestic technical base and mass production capacity. Huawei formed a collaborative layout of Hi-One high-density optical engines, OPENNPO standards, and supernode systems around the “Tao Law”; Ali promoted the 3.2 TNPO pilot deployment and evolved to 6.4 TUPO; Tencent simultaneously verified the 3.2T silicon light and VCSEL solutions, and it is expected that the NPO supernode solution will be deployed in 2026Q4, and commercialization will continue to accelerate.

NPO can fully reuse the existing domestic optical communication industry chain, providing an important opportunity for Chinese enterprises to participate deeply in the next generation of optical connectivity

NPO retains independent optical engines, PIC, EIC, lasers, FAU, high-density connectors, in-board optical fibers, and advanced packaging links. The accumulation of domestic manufacturers in traditional pluggable optical modules, silicon light design, precision coupling, passive devices, optical chips, and large-scale manufacturing can be directly transferred. Currently, NPO mechanical, electrical, optical, management and test interfaces are still in the unification stage. Domestic cloud vendors, equipment vendors, chip manufacturers, and optical communication companies have already participated in the formulation of standards such as OPENNPO, and Chinese companies are expected to further switch to common definers of new optical interconnection architectures.

Investment advice

NPO is the “Tao” in the field of optical connectivity. Cloud vendors at home and abroad are simultaneously embracing NPO. It is also another major opportunity for China's industrial chain and continues to recommend relevant targets. (1) Light engine: Zhongji Xuchuang, Xinyisheng, Guangxun Technology, Huagong Technology; (2) High-density optical connectivity: Taichenguang, Changxin Bochuang, Shijia Photonics, Zhishang Technology; (3) FAU, semiconductor optics: Tianfu Communications, Optical Bank Technology, JPT, Juguang Technology; (4) Light source and upstream materials: Yuanjie Technology, Dongshan Precision, Changguang Huaxin, Yunnan Germanium; (5) Advanced packaging: Changdian Technology; (6) Domestic computing power and line switching chips: Moore Cheng, Mu Xi Co., Ltd., Shengke Communications, ZTE.

Risk warning: AI computing power capital expenditure falls short of expectations; NPO product verification and large-scale deployment progress falls short of expectations; domestic computing power and supernode construction progress falls short of expectations; market competition intensifies.

Disclaimer:Webull uses external vendor Google Translation Service for news translations where we endeavour to ensure these are correct, however, we recommend that you please double-check this information accordingly. Webull is not responsible for translation errors or issues.
What's Trending