Huawei Debuts Stellar AI Network to Address Exploding Token Demand

The updated Stellar AI Fabric 2.0 introduces the SF9300 UBG switch series, which utilizes a two-layer multi-plane architecture to reduce network deployment costs by 30%. By unifying protocols, Huawei has successfully lowered end-to-end latency from 20 μs to 11 μs. Furthermore, the implementation of Link-Layer Retransmission ensures zero packet loss during link flaps, potentially saving 100 hours of cluster downtime annually. For hardware efficiency, the new CloudEngine XH9300-EN switches employ a centralized light-source design that reduces interconnect power consumption from 1,000 W to 600 W.

Security remains a core focus, with a three-layer guardrail system designed to mitigate risks like prompt injection and model poisoning. In initial deployments, this security suite identified 415 high-risk vulnerabilities and over 300 threat events within a single week. The broader Stellar AI WAN solution complements these efforts by enabling lossless computing power delivery over distances exceeding 1,000 km. Through layerwise model partitioning, Huawei claims it can reduce branch xPU costs fourfold while maintaining raw data security on-premises.

For enterprise environments, the Stellar AI Campus Network 2.0 addresses the shift toward autonomous operations. By embedding AI micro-models directly into network switches, the system reduces threat detection time for endpoint terminals from five minutes to under 10 seconds. This autonomy is further supported by the Deep Causal Pruning algorithm, which increases troubleshooting accuracy to 95% and helps slash mean-time-to-repair to under five minutes.