Huawei upgrades its Stellar AI Fabric interconnect solution for efficient AI computing production networks, announced at HUAWEI CONNECT 2026.

At HUAWEI CONNECT 2026, Huawei hosted an AI DC Innovation Summit convening over 600 industry leaders and researchers. The company unveiled its upgraded Stellar AI Fabric Solution, designed to deliver continuous uptime and maximum compute utilization for production AI clusters.

Erick Zhang, President of Data Center Network Domain at Huawei Data Communication Product Line, explained that AI is transforming traditional data centers into compute factories. In this emerging era of agentic AI, the underlying network fabric directly determines token throughput and overall compute yield.

The upgraded Stellar AI Fabric integrates four core components: Rock-Solid Architecture 2.0, StarryWing Digital Map 2.0, Hyper-Converged Fabric (HCF), and Network Packet Load Balancing (NPLB). The enhanced NetMaster network AI agent automates root-cause analysis for 95% of network faults within minutes, maintaining consistent production uptime.

Huawei launched two new switch families optimized for high-density computing. The CloudEngine SF9300 (UBG switch) employs a two-layer multi-plane architecture that reduces buildout costs by 30%. Its simplified UB forwarding lowers end-to-end latency by 40%, while Link-Layer Retransmission (LLR) enables microsecond-level recovery, eliminating packet loss during link flapping.

The CloudEngine XH9300 (NPO switch) delivers proprietary 100T and 51.2T capabilities. A centralized light source and proprietary 3.2T Optical Engine lower interconnect power consumption by 40%. By eliminating the optical DSP stage, the architecture cuts forwarding latency by 26%, while snap-in pluggable optical engines accelerate maintenance tenfold.

The Beijing FinTech Industry Alliance, Huawei, and 17 partners released the Technical Specification for Network High Availability of Financial Data Center, establishing technical baselines and testing methods across two categories and five network types for both intra-DC and cross-DC environments, evaluating architectural redundancy, fault recovery, and proactive prevention.

At the Institute of High Energy Physics under the Chinese Academy of Sciences, Computing Center Director Qi Fazhi reported that deploying the Stellar AI Fabric increased computing efficiency by over 10%. Network-wide load balancing resolved training traffic imbalances, enabling uninterrupted throughput for massive telemetry data and accelerating deep-space signal analysis.