Beijing – October 15, 2026 -- Huawei projects that the number of AI agents worldwide could reach 900 billion by 2035, roughly 100 times the global population, and argues that mobile core networks must be rebuilt to handle them.
The projection comes from Huawei's Intelligent World 2035 report. The company says core network architecture must be able to support long-term growth in new connections and services. It describes the shift as a move from transporting data bits to enabling task-oriented tokens, with agentic services becoming typical mobile network scenarios as AI terminals and applications spread.
Huawei defines three agentic service scenarios for experience assurance
Agent Over New Radio (AoNR) targets task latency. Portable AI devices are limited by computing power, battery capacity, size, and cost, so end-to-end task latency must stay below demanding thresholds.
The Wide-Area Computing Network (WCN) targets computing efficiency. It interconnects computing resources through mobile networks, enabling task-based computing scheduling and standardized computing services.
The Agent Computing Network (ACN) targets collaboration security. Huawei says complex tasks will involve multiple specialized agents working across domains, enterprises, and ecosystems rather than a single agent. A single agent is constrained by model capabilities, context limitations, role conflicts, and sequential processing. The ACN rests on three capabilities: cross-domain trusted digital identity, skill-based agent discovery and routing, and task-driven dynamic networking.
Next-generation core adds a Data & AI Plane and intent-based interfaces
Huawei sets out three characteristics for the architecture:
- Interfaces: Intent-based interfaces replace signaling-centric ones. Terminals pass personalized task requirements, including networking, computing, and experience needs, directly to the network.
- Functions: A dedicated Data & AI Plane (DAP) sits alongside the traditional control and user planes. It interprets task intents and aligns resources dynamically in real time.
- Procedures: AI decision-making is embedded in fundamental network procedures, moving from an AI overlay to an AI-native design that coordinates connectivity and computing.
Two-stage evolution moves operators from AI overlay to AI-native core
Stage one overlays AI capabilities on existing networks to enhance customer experience. Stage two is the AI-native next-generation core, which Huawei says delivers two capabilities. "Enhanced connectivity" uses task-oriented AI services to ensure task completion. "Beyond connectivity" extends the network to computing services and agent collaboration.
Huawei says this requires restructuring the network's interaction mechanisms, data architecture, and computing architecture. It adds that the transition depends on collaboration across the ICT ecosystem, including terminals, chips, and networks, through a terminal-network synergy mechanism.