In a groundbreaking advancement poised to redefine future wireless communications, researchers from China Mobile Communications Group Corporation and China Mobile Research Institute have unveiled an innovative design framework for a 6G AI-native architecture. This pioneering work goes beyond incremental upgrades by embedding artificial intelligence (AI) as a core architectural pillar rather than a peripheral feature, setting the stage for a truly intelligent mobile network ecosystem that seamlessly integrates AI within its fabric. Published in the prestigious journal Engineering, this study addresses the critical challenges and proposes an overarching blueprint intended to harmonize AI-driven network operations with AI-enabled service capabilities, heralding a new era of communications technology.
The journey toward 6G represents the third monumental paradigm shift in mobile communications history, tracing back from 1G’s analog voice systems through 2G’s digital cellular technology, and further to 3G and 4G’s packet-switched internet capabilities. However, unlike 5G, which first flirted with AI integration via limited mechanisms such as the Network Data Analytics Function (NWDAF), 6G is conceived with AI as an equal partner to traditional communication functions. Under the guidance of the International Telecommunication Union Radiocommunication Sector (ITU-R) IMT-2030 framework, 6G systems are expected to natively implement AI-driven processes and communication paradigms, with the 3rd Generation Partnership Project (3GPP) identifying deep AI-network symbiosis as an essential requirement for 6G deployment.
This research confronts existing limitations in current network architectures, noting that prior approaches have treated AI as add-on components designed only to enhance specific network functions or analytics. Instead, the newly proposed architecture lays out a foundational integration strategy, where AI capabilities are distributed and managed across multiple network layers, enabling the network itself to act as an intelligent entity capable of self-optimization and support for AI-centric applications and services. This shift is critical to meet the rising demands for ubiquitous AI services, real-time decision-making, and automation, which characterize forthcoming technological landscapes.
The study identifies four pivotal design challenges that any 6G AI-native architecture must address. First is capability generalization, ensuring that the architecture can handle a broad spectrum of AI workloads—from low-latency, high-reliability tasks to more processing-intensive applications—with diverse performance requirements. Second, quality assurance addresses the inherently probabilistic nature of AI output, necessitating robust mechanisms to mitigate uncertainty, guaranteeing the rigorous reliability standards expected in mobile communications. Third, efficiency optimization is crucial, balancing AI’s substantial computational and energy demands with sustainable network operation. Lastly, global optimization requires an integrated approach to harmonize these competing factors, achieving an equilibrium between capability, quality, and efficiency that holistically benefits all network stakeholders.
To contend with these challenges, the researchers propose a triad of core principles guiding the architecture’s development: practicality in deployment and operation; simplicity in design to reduce complexity and overhead; and flexibility to adapt dynamically to changing demands and emerging AI technologies. Complementing these principles is a systematic, task-driven design methodology. This methodology unfolds in four rigorous steps rooted in system theory: defining clear AI-task objectives; specifying architectural elements to support these tasks; establishing hierarchical relationships for scalable deployment; and delineating connectivity frameworks to ensure seamless coordination across the network. This process is iterative, incorporating continuous refinements to optimize trade-offs among the key metrics.
The resultant 6G AI-native architecture is characterized by the integration of distributed AI components for data processing and computing, while maintaining layered centralized control to orchestrate network-wide resources and decisions. The architecture’s primary constituents include three foundational elements—connectivity, computing, and data—and three core functional modules—service, control, and execution—each fulfilling specialized roles to sustain AI-native operations. Notably, enhancements to both the core network (CN) and the radio access network (RAN) are fundamental. The CN acts as a centralized orchestrator, dynamically managing resources to foster efficient AI tasks, whereas the RAN supports distributed AI execution at the edge, critical for applications demanding ultra-low latency.
Experimental verification of this architecture was conducted using the Free5GC platform, an open-source 5G core network implementation. Testing validated the feasibility of delivering network-native AI computing services, demonstrating successful convergence of connectivity and computing management. Moreover, it showcased dynamic orchestration of AI tasks—a key capability for future 6G networks expected to handle diverse, concurrent AI workloads with varying service-level agreements (SLAs). These empirical findings underscore the architecture’s potential to meet the demanding requirements envisioned for post-5G networks.
This research also surveys the progression of 5G standardization around network-AI integration. It highlights the evolutionary path of NWDAF and the incorporation of intelligent RAN specifications as foundational milestones. Building on this foundation, the paper maps out essential standardization trajectories for 6G encompassing a paradigm shift from patchwork AI implementations to native AI design embedded from inception. Furthermore, it stresses the necessity of cross-domain AI consistency, enabling interoperable AI functions across heterogeneous network segments and multi-vendor environments. The establishment of service architectures supporting AI agent ecosystems and frameworks for cross-domain AI inference coordination are identified as strategic priorities to ensure scalable and efficient AI deployment.
In its concluding remarks, the study emphasizes that the proposed task-driven, principle-based design approach is not merely a theoretical exercise but a practical baseline fostering synergy among industry stakeholders. Realizing the full promise of 6G AI-native networks demands concerted efforts to forge consensus on functional specifications, procedural workflows, and interoperability standards. Such a unified vision will underpin the development and deployment of next-generation mobile networks that are intrinsically intelligent, adaptive, and seamless.
This paradigm shift to AI-native networking promises transformative benefits: network operators can leverage automated, context-aware system management that dynamically optimizes performance and resource allocation. Simultaneously, users and industries will gain access to robust AI-as-a-Service (AIaaS) ecosystems supported natively by the communication infrastructure, unlocking unprecedented possibilities in sectors ranging from healthcare to autonomous vehicles and immersive digital experiences. As the wireless industry prepares for this new frontier, the insights derived from this research illuminate a clear path forward toward the realization of truly intelligent, AI-empowered 6G networks.
As this foundational research is disseminated and debated within academic and industrial circles, the coming years will likely see accelerated innovation cycles, prototype deployments, and the formulation of global standards embodying the principles outlined. The 6G vision articulated here is not merely an extension of past wireless generations but a fundamental reimagining, aligning communication technology with the exponential advances in artificial intelligence. This synergy is destined to reshape how humans and machines connect, communicate, and collaborate in the digital age.
Subject of Research:
Integration of Artificial Intelligence as a foundational component in 6G mobile network architecture.
Article Title:
A Task-Driven Design Approach for 6G AI-Native Architecture
News Publication Date:
29-Jan-2026
Web References:
Full paper: https://doi.org/10.1016/j.eng.2025.09.005
Journal website: https://www.sciencedirect.com/journal/engineering
References:
Wang, X., Lu, L., Li, Q., Sun, Q., Shi, N., Chen, Z., & Sun, T. (2026). A Task-Driven Design Approach for 6G AI-Native Architecture. Engineering. https://doi.org/10.1016/j.eng.2025.09.005
Image Credits:
Xiaoyun Wang, Lu Lu, Qin Li, Qi Sun, Nanxiang Shi, Ziqi Chen, Tao Sun
Keywords
6G, AI-native architecture, mobile networks, artificial intelligence, network design, wireless communications, 3GPP, ITU-R IMT-2030, NWDAF, network orchestration, edge computing, AI-as-a-Service
Tags: 6G AI-native architectureAI as core network pillarAI integration in 6GAI-driven wireless communicationsAI-enabled service capabilitiesAI-powered 6G network operationsChina Mobile 6G researchfuture mobile network ecosystemsintelligent communication networksITU-R IMT-2030 6G frameworknext-generation wireless technologytask-oriented network design
