A study published in Engineering presents a novel task-driven design approach for a 6G AI-native architecture, developed by researchers from China Mobile Communications Group Corporation and China Mobile Research Institute. The research systematically analyzes the core challenges and design requirements for integrating artificial intelligence (AI) into 6G mobile networks, and outlines a comprehensive architecture that embeds AI as a foundational element rather than a supplementary feature, aiming to support the dual development of AI-empowered network operations and network-enabled AI services.

The paper first contextualizes 6G’s evolution as the third major leap in mobile service paradigms, following the shifts from circuit to packet switching and from individual to industry-focused services. Unlike 5G, which introduced limited AI integration via functions like the Network Data Analytics Function (NWDAF), 6G is defined by the ITU-R IMT-2030 framework to include AI and communication as a core scenario, with 3GPP identifying inherent AI-mobile network integration as a fundamental 6G requirement. The research notes industry consensus on 6G’s role as both an AI-enhanced communication platform and an efficient AI-as-a-Service (AIaaS) enabler, while highlighting current research gaps in holistic architectural design for deep AI-communication convergence.

The researchers identify four key design challenges for 6G AI-native architecture: capability generalization to support diverse production and service-oriented AI tasks with varying latency and performance demands, quality assurance to mitigate AI’s inherent uncertainty in high-reliability mobile networks, efficiency optimization to balance AI’s high resource consumption with network sustainability, and the global optimization of these three competing dimensions. To address these, the study proposes three core design principles—practicality, simplicity, and flexibility—and a task-driven four-step design methodology grounded in system theory, encompassing task definition, element definition, hierarchy definition, and connectivity definition, followed by iterative optimization to balance capability, quality and efficiency.

The proposed 6G AI-native architecture integrates distributed AI data and computing components with layered centralized collaborative control, structured around three fundamental elements (connectivity, computing, data) and three core functional modules (service, control, execution). It features enhanced core network (CN) and radio access network (RAN) functionalities, with the CN acting as a central hub for dynamic resource orchestration and the RAN enabling distributed AI execution for low-latency edge AI services. Experimental verification on the Free5GC platform validated the architecture’s feasibility in delivering network-native AI computing services, demonstrating converged connectivity-computing management and dynamic AI task orchestration.

The research also analyzes 5G standardization practices for network-AI integration, including the evolution of NWDAF and RAN intelligence specifications, and outlines key 6G standardization directions. These include a paradigm shift to native AI design at the architectural inception, cross-domain AI consistency alignment, and the development of service architectures for AI agent ecosystems and cross-domain AI inference coordination. The study concludes that the proposed design principles and task-driven methodology provide a baseline for 6G AI-native architecture design, calling for further industry consensus on functional and procedural details to advance unified standardization.

The paper “A Task-Driven Design Approach for 6G AI-Native Architecture,” is authored by Xiaoyun Wang, Lu Lu, Qin Li, Qi Sun, Nanxiang Shi, Ziqi Chen, Tao Sun. Full text of the open access paper: https://doi.org/10.1016/j.eng.2025.09.005. For more information about Engineering, visit the website at https://www.sciencedirect.com/journal/engineering.