Building a Native AI Core to Support Next-Gen Agent Services

In an era where traditional mobile networks predominantly facilitated voice calls, text messaging, and mobile internet access, a new wave of innovation is on the horizon. As we transition into an age dominated by artificial intelligence (AI), the challenges of managing a vast army of AI agents have emerged, necessitating a fundamental shift in network architecture. To navigate this evolution, a core network that comprehends intent, performs intelligent decision-making, and ensures the completion of AI-driven tasks across the board is essential.

Huawei has posited that an AI-native core network is critical in this transformative process. By integrating AI into network interfaces, functions, and procedures, network operators can evolve from mere connectivity providers to intelligent networks that give rise to next-generation digital services. According to Huawei's "Intelligent World 2035" report, the global number of AI agents could skyrocket to approximately 900 billion by the year 2035—about 100 times the current global population. This phenomenal growth calls for a core network architecture that can accommodate new connections and services over the long term.

As AI-enabled devices and applications grow in prevalence, the concept of agent-based services will likely become a standard feature of mobile networks. This transition reflects a broader shift from simple data transport to an ecosystem where task-oriented tokens fuel interactions. To facilitate this, Huawei has identified three typical service scenarios essential for guaranteeing a seamless agent experience:

1. Agent over New Radio (AoNR): This approach is designed to ensure low latency for tasks. Considering the computation limits of wearable AI devices, factors like battery capacity and cost play crucial roles. AoNR must consistently provide outstanding experiences while maintaining strict end-to-end latency thresholds.

2. Wide Area Computing Network (WCN): By connecting computing resources across mobile networks, WCN enables standardized scheduling of tasks and computing services. This boosts the overall efficiency in how network resources are utilized and tasks are executed.

3. Agent Computing Network (ACN): When complex tasks demand the collaboration of multiple AI agents across different domains and ecosystems, an ACN becomes imperative. Unlike single agents constrained by model limitations or contextual conflicts, ACNs provide a safer, dynamic collaborative environment among specialized agents.

Three core capabilities for ACNs enable effective cooperation:
1. Reliable digital identity across domains: Essential for maintaining trust and security in the collaborative ecosystem.
2. Agent discovery and skill-based routing: Facilitates efficient connections and task execution among diverse agents.
3. Task-oriented dynamic networks: Aids in adapting to the demands and complexities inherent in agent interactions.

To bridge the gap between devices and networks, next-gen core network architecture should embody three key characteristics:
1. Intent-based Interfaces: These enable terminals to directly communicate their customized task requirements, integrating wiring needs, computing, and user experience.
2. C/U+DAP Functionality: The introduction of a dedicated AI-enabled data plane (DAP) works alongside traditional control and user planes to dynamically align resources in real-time based on task intentions.
3. Deep AI Integration in Procedures: Seamlessly incorporates AI decision-making within core network procedures, facilitating smooth coordination of connectivity and computation.

Looking ahead, Huawei plans a two-phase evolution for its AI-powered core network. The initial phase focuses on enhancing customer experience via AI capabilities, while the subsequent phase introduces an AI-native core network which will provide "enhanced connectivity" where task-oriented AI services ensure completion, extending "beyond connectivity" to include computing and multi-agent collaborations. This necessitates the restructuring of network architecture particularly concerning interaction mechanisms, data architecture, and computation frameworks.

Moreover, the transition to AI-native networks hinges on holistic collaboration across the ICT ecosystem encompassing terminals, chips, and networks. Through synergistic mechanisms connecting terminals and networks, a next-gen core network will empower unprecedented opportunities for new connections and services in the AI era, redefining business landscapes and laying the groundwork for smarter communication frameworks.