A research report from East Money Securities indicates that Nvidia's GTC conference outlined the company's roadmap for computing and networking product iterations over the next two years, with multiple interconnect pathways expected to benefit. In the short to medium term, copper interconnect solutions are likely to be adopted by end customers due to advantages in cost and stability. Over the long term, optical interconnect solutions such as CPO and NPO are expected to gradually increase in adoption. From a cost perspective, Nvidia has retained short-distance copper interconnect versions between chips in its Rubin Ultra and Feynman series, offering cloud providers flexibility in choosing cost-effective options. This suggests that copper interconnects will remain viable at least until the Feynman generation. The cost-performance advantage of copper connections makes them particularly suitable for cost-sensitive applications such as AI inference. Key insights from East Money Securities are as follows:
Cluster-level products are targeting a trillion-dollar market, with the Feynman architecture preserving a role for both optical and copper interconnects. At the GTC 2026 event held in San Jose, California, Nvidia’s founder and CEO Jensen Huang projected that revenue opportunities for Nvidia’s AI chips—including the Blackwell and Rubin series—could reach at least $1 trillion by 2027. The company introduced the Vera Rubin POD-level AI platform and revealed a technical prototype of the next-generation Feynman architecture. Nvidia’s product roadmap confirms a dual-path network ecosystem strategy that incorporates both copper cabling and CPO technologies, indicating increased demand for copper cables, optical chips, and CPO production capacity.
The coexistence of optical and copper interconnects is expected to continue. East Money Securities believes that system interconnect architecture will not rely on a single technological pathway. Instead, a combination of copper and optical interconnects will be deployed based on factors such as transmission distance, power efficiency, cost-effectiveness, and reliability. Copper cables remain the core solution for short-distance interconnects. Nvidia will continue to advance copper interconnect technology, while optical interconnects will be applied in both scale-up and scale-out scenarios. Both capabilities are considered essential to Nvidia’s technology strategy, and the Feynman series will support both types of connections.
Continued attention should be paid to the application progress of CPO products. Nvidia’s CPO Spectrum-X switch has entered full mass production and is manufactured by TSMC. However, due to its highly integrated nature, CPO carries risks of vendor lock-in. Additionally, cloud providers remain cautious about early-stage CPO manufacturing yields and maintenance models. While CPO technology is expected to mature over time, widespread adoption will depend on validation and acceptance by cloud service providers.
In the short to medium term, cost advantages within server racks may sustain the relevance of copper interconnects. By retaining short-distance copper interconnect options in the Rubin Ultra and Feynman series, Nvidia provides cloud providers with cost-effective alternatives, ensuring copper interconnects remain viable at least through the Feynman generation. The favorable cost-performance ratio of copper connections makes them especially advantageous in cost-sensitive applications such as inference. In high-speed scenarios, companies such as Credo and Huagong Zhengyuan have introduced 1.6T AEC/ACC products capable of meeting certain high-bandwidth demands.
In terms of investment targets, it is recommended to monitor data center optical communication companies including Zhongji Innolight, Eoptolink, Accelink, HG Tech, TFC Optical, Sicoya, Gigalight, JIayuan Technology, and Uscell; copper interconnect suppliers such as Zhaolong Interconnect and Recodeal; and switch manufacturers including Ruijie Networks and Phylink.
Risks to consider include customer concentration, order fulfillment uncertainties, technological iteration challenges, and overseas investment risks.
Comments