The China International Optoelectronics Summit Forum, held concurrently with the China International Optoelectronics Exhibition (CIOE) (China Optical Exposition), was recently successfully held at the Shenzhen International Convention and Exhibition Center. As a multi-level international exchange platform integrating industry, academia, research, and application, the forum focused on academic, industrial, and application-focused topics, offering a comprehensive and in-depth discussion of the latest technologies and research directions in the optoelectronics field.
During the forum, Zhang Haiyi, Director of the Institute of Technology and Standards at the China Academy of Information and Communications Technology, delivered a keynote speech titled "Progress and Outlook of High-Speed Optical Communication Technology in the AI+ Era." She pointed out that in the AI+ era, AI and optical networks will continue to mutually empower innovation and development, driving the accelerated evolution of high-speed optical communication technology towards ultra-high speed, high energy efficiency, all-optical, and intelligent technologies.
The AI+ era is driving new demands for computing-network convergence, with optical networks and AI enabling two-way collaboration.
Zhang Haiyi stated that multimodal AI big models are experiencing explosive growth and continuously progressing toward general intelligence, forming a three-tiered development landscape. General big models, based on the development path of "big models + big computing power + big data," are continuously approaching limited general intelligence in human-computer interaction environments. Industry big models, as specialized big models supporting specific industries, meet the needs of core business operations and specialized data analysis, enabling the digital and intelligent transformation of specific industries. Device-side big models are becoming a key focus for manufacturers of mobile phones, PCs, smart cockpits, and humanoid robots, actively seizing the entry point for AI applications.
Open-source AI big models such as DeepSeek are driving the widespread adoption of collaborative inference applications across the cloud, edge, and device, addressing the issue of insufficient local computing power. Distributed resource scheduling is becoming critical. This is also driving demand for inference clusters. Real-time inference applications rely on millisecond-level computing resource responses, requiring networks to provide high-bandwidth, low-latency connections to cloud computing power to create a local-like experience while addressing issues such as data security and computing power access costs. Against this backdrop, optical networks and AI are forming a mutually beneficial partnership. On the one hand, optical networks provide an all-optical foundation for high-quality connected computing and data access. These networks not only build data transmission highways within AI systems, enabling the rapid and stable flow of large amounts of data between devices and nodes, but also promote the sharing and collaboration of AI-related resources, supporting the development of distributed AI large-scale model training and inference applications.
On the other hand, AI technology is driving the advancement of optical network intelligence. The entire lifecycle of optical network operations and management (planning, construction, maintenance, optimization, and operation) is crucial to the network's intelligence level and user experience. Technologies such as digital twins, AI large models, and intelligent commissioning enable intelligence at the service, control, and device levels, building full-lifecycle intelligent operations and maintenance capabilities and effectively improving the efficiency of information and communication network operations.
High-speed optical communication technology is evolving along four major directions, and industry breakthroughs are continuing to advance.
Zhang Haiyi pointed out that in the AI+ era, high-speed optical communication technology is accelerating along four major directions: ultra-high speed, high energy efficiency, all-optical integration, and intelligence. Significant technological and industrial progress is being made in each area. Regarding higher speeds, 400G systems have entered large-scale commercial use, with domestic operators deploying nearly 20,000 400G high-speed ports across their networks.
WSS has achieved an integrated design for the C+L bands, and the integrated OTU and EDFA technical solutions and industrialization are still under accelerated development and verification. Coherent technologies for 800G and above are becoming a hot topic, with testing and standardization progressing simultaneously. Looking ahead, AI-powered intelligent computing optical interconnects continue to accelerate, with 3.2T likely to be the evolutionary target within the next three years within data centers.
Regarding higher energy efficiency, retiming, LPO, and CPO are developing concurrently, with my country's CPO requiring focused efforts to achieve breakthroughs. OIO supports computing and storage optical interconnects, expanding towards multi-node system architectures. Many industry giants are deeply engaged in silicon photonics integrated platforms, and China needs to accelerate its deployment of advanced integration. Furthermore, new media technologies are developing simultaneously: SDM fiber standards are steadily advancing, with petabit applications gradually being explored and expanded. Hollow-core fiber technology holds great potential, with application scenarios under continuous exploration. SDM is exploring submarine applications and deployment, and participation in international submarine cable standards is increasing. In terms of all-optical connectivity, within intelligent computing centers, the OCS offers low latency and flexible scheduling, optimizing interconnection efficiency. Between intelligent computing centers, an optoelectronic collaborative networking architecture is being established to improve interconnection quality. The OXC+ optoelectronic collaborative networking architecture enables network-based computing, facilitating high-speed interconnection of distributed intelligent computing clusters and helping to overcome computing power bottlenecks.
In terms of intelligence, the integration of optical, communication, and sensing has attracted attention, but standardization is still in its early stages. We need to actively embrace AI big models to accelerate network intelligent transformation. At the same time, intelligent entities are becoming a hot topic in network big models, and their applications require ongoing evaluation. Digital twins are facilitating application innovation and accelerating standardization.
She pointed out that in terms of standardization and monitoring, the China Academy of Information and Communications Technology is gradually improving its capacity index system and continuously conducting nationwide capacity monitoring. For example, the "Computing Network Capacity Index Assessment Report (2025)" is planned to be officially released at the "All-Optical Capacity Millisecond Computing" forum held at the Beijing PT Exhibition on September 25, 2025. Furthermore, in terms of practical advancement, the "Metro-Area Millisecond Computing" initiative was launched to support the development of computing applications with all-optical transport capacity. In terms of international collaboration, preliminary research on an international optical network technology standard framework oriented towards 2030 was jointly promoted.
Concluding his speech, Zhang Haiyi stated that, focused on the new demands and challenges of the AI+ era, the China Academy of Information and Communications Technology will continue to collaborate with various sectors within China, including industry, academia, research, and application, to jointly advance high-speed optical communication technology innovation, industrial ecosystem development, and network quality monitoring, fully supporting the high-quality development of my country's new information infrastructure.