Huawei proposes network-compute-storage integration to tackle AI computing power bottlenecks

Huawei said at a recent media briefing that AI large models have moved from testing to commercial use, but the industry faces challenges like insufficient computing power, high adaptation costs, and high development barriers. Huang Jian, Huawei's internet industry general manager for China, noted that token usage is growing explosively, far outpacing industry adaptation, and predicted token economy will reshape all work frameworks by 2027 to 2030.

Huang cited data showing token consumption grew 6x and model calls increased 50-100x in the past six months, driving demand for computing, storage, and networking. He said network capacity is the biggest bottleneck. Huawei's Xinghe AI data center network and Ascend super nodes boost NPO bandwidth by 8x and token generation efficiency by 5x, reducing distributed communication losses through hardware, software, scheduling, and ecosystem coordination.

For the internet industry's high-concurrency needs, Huawei built a three-tier service system: dedicated teams of over 100 engineers for top companies, and standardized services with 24/7 support for SMEs via partners. Huawei also invests thousands of staff in adapting open-source software and contributing to communities, lowering barriers for large model training and deployment.