Huawei has proposed redesigning the mobile network core to route AI-agent requests by their requirements and coordinate computing capacity, a plan that would extend operators’ role beyond providing connectivity.
The blueprint adds a Data & AI Plane to existing network control and user functions, and places AI decisions inside procedures that govern network operations. Huawei says devices could use an intent-based interface to describe what a task needs. The proposal includes Agent Over New Radio (AoNR) for latency-sensitive tasks, a wide-area computing service to connect computing resources, and an Agent Computing Network for collaboration among agents across domains.
In a Sept. 8 technical blog post, Huawei set two AoNR targets: transmitting all tokens for a single AI task within 50 milliseconds and keeping end-to-end task latency below 400 milliseconds. For the second measure, the clock starts when a user submits a task and stops when the first token arrives. These are proposed targets, not results from an operating network.
Huawei’s Intelligent World 2035 report forecasts that 900 billion AI agents could be served by 2035, alongside a global population of 9 billion. The 900 billion estimate is Huawei’s projection. The company’s blog points to limits in portable devices’ computing power, battery capacity, size and cost as reasons some work may need to be handled elsewhere. Huawei also presents task-oriented network services as a potential offering for operators, but has not specified who would pay to build them.
Huawei frames the proposal as another step in network design. Its blog describes 2G and 3G core networks as IP-based, 4G as using a distributed design with control and user functions separated, and 5G as cloud-native.
A Huawei blog post called for industry work on AI-native architecture through the 3rd Generation Partnership Project (3GPP). The group’s portal lists its 6G system architecture study as a draft, with version 0.9.0 uploaded Sept. 21. The listing does not show that Huawei’s proposed architecture has been adopted, and no date for a standards decision has been announced.
