China Mobile Hubei, Huawei Validate AI Inference Solution for Carrier Networks

Huawei Launches Xinghe Intelligent Data Center Network to Power the AI Era

China Mobile Hubei and Huawei validated China’s first carrier AI inference acceleration solution, achieving up to 372% higher token throughput and advancing efficient AI infrastructure deployment for carriers.

China Mobile Hubei and Huawei Improve AI Inference

China Mobile Hubei and Huawei have completed China’s first carrier industry validation of an AI inference acceleration solution, marking a significant milestone in improving AI computing efficiency for telecommunications networks.

The live-network validation was announced during MWC Shanghai 2026 and demonstrated substantial performance gains for long-sequence AI inference workloads.

The solution combines Huawei’s OceanStor A800 storage, Ascend A3 SuperPoD, and Unified Cache Manager (UCM), delivering up to a 372% increase in token throughput.

Huawei said the technology addresses one of the biggest challenges in AI inference by overcoming the memory limitations that restrict KV cache capacity during long-context processing.

Introduced in 2025, UCM uses external high-performance storage to expand KV cache capacity while managing cache throughout its lifecycle. This enables longer context windows for single-turn conversations and allows historical KV cache to be reused during multi-turn dialogues, reducing redundant computation and lowering inference costs.

The validation used the vLLM-Ascend framework in China Mobile Hubei’s live network with sequence lengths ranging from 8,000 to 190,000 tokens across MiniMax M2.5 and GLM-5.1 models. Results showed significant improvements in both time to first token (TTFT) and tokens per second (TPS), with the largest gains recorded in longer sequence environments.

China Mobile Hubei said the validation highlights the value of storage, computing, and network collaboration for AI services such as AI agents and code generation.

Huawei added that growing demand for AI agents and token-based services will require more efficient AI infrastructure, with the validated solution helping carriers improve performance while reducing operational costs and supporting large-scale AI deployment.