Sino-Telecom Unveils Integrated AI Computing Infrastructure at 10th Future Network Development Conference

Sino-Telecom and Purple Mountain Laboratories jointly unveiled an AI computing infrastructure solution at the 10th Future Network Development Conference, held in Nanjing from October 15 to 17, 2026. Presented in the conference's innovation showcase area, the solution combines computing, networking and optical transport to support AI clusters built on domestically developed technology.

The release addresses the growing importance of cluster interconnect efficiency as AI models expand. Sino-Telecom describes this broader scope as a move from Scale Up toward Scale Across: extending coordination beyond a single node or rack to multiple nodes, racks and AI data centers. The approach places greater emphasis on bandwidth, latency, lossless transmission, flexible networking and domestic control of the infrastructure.

Sino-Telecom's integrated IP and optical architecture connects these requirements across three layers, covering networking within AI clusters and interconnection between AI data centers.

The AS8200 is based on Ascend AI processors and provides computing resources for AI training and inference. It combines high-density heterogeneous computing with coordinated hardware and software, serving as the computing layer of the overall solution for government, enterprise and industry AI applications.

The 51.2T AI switch supports networking within clusters and direct use of wavelength-specific optical transceivers. This approach is designed to simplify the network architecture by eliminating intermediate transport equipment, while supporting scalable, high-speed interconnection among cluster nodes.

The 800G data center interconnect (DCI) solution targets AI data center interconnection, distributed AI training and large-scale data migration. Designed for high bandwidth, low latency and lossless transport, it addresses the data synchronization and model-parameter exchange requirements of AI clusters spanning multiple data centers.

Within the combined architecture, the AS8200 supplies AI training and inference resources, the 51.2T switch connects nodes within clusters, and 800G DCI links AI data centers. The IP layer provides flexible scheduling and networking, while the optical layer supports high-speed, low-latency transmission. Together, the three layers form an integrated infrastructure option for domestically developed AI computing clusters.

Sino-Telecom plans to continue developing products for high-density AI computing, edge AI and industry applications based on domestic IT, following an open, decoupled approach. Its stated direction is to connect computing hardware, software tools and industry solutions as AI cluster deployments expand.