Chip-to-chip interconnection, a core component of global AI computing infrastructure, is undergoing a historic transition from electrical to optical technology. On March 16 local time, NVIDIA unveiled its Feynman chip, marking the first integration of optical communication for chip interconnects, which can reduce communication energy consumption in AI data centers by over 70%.
Analysts believe that with the establishment of overseas technological pathways and increased domestic industrial policy support, A-share companies deeply embedded in the global computing supply chain are gaining attention. Notably, while industry giants advance rapidly, domestic policy levels have injected strong momentum into China's optical communication technology development. The "Steady Growth Action Plan for the Electronic Information Manufacturing Industry (2025-2026)" jointly issued by the Ministry of Industry and Information Technology and the State Administration for Market Regulation explicitly proposes "conducting technological research in key areas of photonics, increasing R&D investment in high-speed optical chips and co-packaged optics (CPO), and promoting the integration of optical architecture with existing electrical architecture ecosystems." This policy statement precisely targets the core segments of the CPO technological pathway, providing clear direction for domestic companies' technological advancements.
Local governments are also taking frequent actions. In March, the General Office of the Guangdong Provincial Government issued the "Action Plan for Accelerating the Cultivation and Development of New Tracks to Lead the Construction of a Modern Industrial System (2026–2035)," proposing to "strengthen R&D of key materials, equipment, and processes for optical chips in sub-sectors such as optical communication, optical sensing, and optical computing." It also emphasizes increasing the demonstration and application of optical chip products in scenarios like next-generation information communication, data centers, and intelligent computing centers.
Supply Chain Deep Integration Even two weeks before the GTC conference, NVIDIA had already signaled strong commitment to optical technology. In early March, NVIDIA announced on its website strategic agreements with Lumentum and Coherent, investing $2 billion in each optical technology company. This total strategic investment of $4 billion was interpreted by the market as NVIDIA's move to secure key upstream materials in the CPO industry chain.
In fact, leading A-share listed companies are already deeply embedded in NVIDIA's supply chain. China Merchants Securities research team stated that Zhongji Innolight Co.,Ltd. holds a first-mover advantage in the 800G high-speed optical module sector and has begun laying out 1.6T products. Wang Yihong, Chief Communications Analyst at TF Securities, noted that Eoptolink Technology Inc.,Ltd. is accelerating shipments of silicon photonics products, with the proportion of silicon photonics products expected to significantly increase by 2026. The company has successfully launched 400G, 800G, and 1.6T product series based on silicon photonics solutions. With the mass production of NVIDIA's Rubin platform in the second half of the year and the clarification of the Feynman architecture's technological pathway, the product iteration cycle from 800G to 1.6T and even 3.2T is accelerating.
At the concurrently held Optical Fiber Communication Conference and Exhibition (OFC 2026), Chinese optical module manufacturers were also highly active. At the exhibition, 1.6T optical modules, silicon photonic integration, and CPO technology solutions became focal points. Industry insiders believe that optical interconnection is becoming a core variable determining the performance ceiling of AI infrastructure. Multiple technological curves, including product iteration from 800G to 1.6T, architectural evolution from pluggable to CPO, and short-distance replacement of copper cables with optics, are advancing simultaneously, collectively forming the technological foundation of this round of optical interconnection evolution. As AI cluster scales continue to expand to hundreds of thousands of GPUs, the optical module industry is facing structural growth opportunities. Chinese manufacturers, leveraging deep optical technology accumulation and rapid iteration capabilities, are expected to continuously increase their market share in the high-end optical module sector.
Industry Chain Value Reassessment With NVIDIA's formal introduction of optical communication for chip interconnects at GTC 2026, the industrial pathway for CPO technology has become clear. Analysts generally believe that this technological breakthrough not only marks the entry of AI computing infrastructure into a new cycle of architectural upgrades but also signifies a reshaping of the value distribution pattern in the optical communication industry chain. From upstream optical chips and midstream optical modules to downstream data center applications, the large-scale commercial use of CPO is opening new growth spaces.
Xiao Qunxi, Chief Machinery Industry Analyst at Guotai Junan Securities, stated that Rubin is no longer just a single GPU product but an integrated AI supercomputing platform comprising CPU, GPU, interconnection, networking, and system components. NVIDIA is elevating the delivery unit of AI infrastructure from board-level to entire cabinet systems. To achieve such high-density interconnection, Rubin may adopt a two-layer network topology and implement "optical replacing copper" within cabinets.
Xiao further explained that at the interconnection level, CPO and silicon photonics are becoming important directions for hyperscale AI systems. Future data centers will gradually transition from traditional copper interconnects to optical connection systems with higher bandwidth density and lower loss. In terms of cooling, air cooling is losing suitability for ultra-high-power computing platforms, and liquid cooling is gradually shifting from an optional solution to a standard configuration.
Cheng Qiang, Director of the Research Institute and Chief Economist at Tebon Securities, stated that from an industry development trend perspective, CPO is accelerating its transition from the technology validation phase to early commercialization. Global foundry giants are accelerating their layout in silicon photonics chip manufacturing. Previously, Tower Semiconductor announced plans to double its silicon photonics manufacturing capacity, GlobalFoundries acquired a Singapore-based silicon photonics wafer foundry, and United Microelectronics Corporation collaborated with IMEC to achieve CPO-compatible silicon photonics processes. Domestically, foundries such as Yandong Microelectronics and Sai Microelectronics are also持续推进 the process development and wafer manufacturing layout for silicon photonics chip series.
Ou Zixing, Communications Industry Analyst at Soochow Securities, believes that future optical interconnection will be driven by diverse network connection scenarios, and the overall market space of the industry is expected to maintain rapid expansion. Various technological pathways are not entirely substitutive but form differentiated positioning in terms of technical characteristics, cost structures, and adapted application scenarios.
Combining market cycles and industrial evolution stages, Ou suggested focusing on three core themes: first, core optical modules, where industry leaders will fully benefit from the trend towards higher speeds and larger bandwidths, offering both growth and certainty; second, second-tier optical modules, where as downstream demand spreads from leading customers, second-tier manufacturers are expected to gain more opportunities to break into the supply chain; third, emerging optical interconnection technologies, such as CPO and silicon photonics, which are in the early industrialization phase from zero to one, and current strategic positioning may yield incremental benefits from technological iteration.
Comments