Next-Generation Datacenter Networking: Architectures, AI-Driven Automation, Software-Defined Infrastructure, and Future Research Directions

Abstract

Modern hyperscale datacenters form the computational backbone of the global digital economy, supporting workloads ranging from cloud computing and artificial intelligence to real-time streaming and edge computing. As traffic volumes grow exponentially and application demands become increasingly heterogeneous, traditional datacenter network architectures face fundamental scalability, efficiency, and operational challenges. This paper provides a comprehensive survey of next-generation datacenter networking, spanning four interconnected domains. First, we examine emerging network architectures, including spine-leaf topologies, fat-tree derivatives, dragonfly interconnects, and optical circuit-switched fabrics, evaluating their scalability, fault tolerance, and cost profiles. Second, we analyze the role of artificial intelligence and machine learning in network automation, covering intent-based networking, autonomous traffic engineering, predictive failure detection, and AI-driven security orchestration. Third, we investigate software-defined infrastructure paradigms, encompassing Software-Defined Networking (SDN), Network Function Virtualization (NFV), programmable data planes via P4, and the convergence of network-compute-storage disaggregation. Fourth, we synthesize open research challenges and future directions, including terabit-scale optical switching, quantum networking integration, sustainable energy-aware networking, and the evolution toward self-driving datacenter networks. Through a systematic analysis of over 80 references spanning academic literature, industry whitepapers, and standards documents, this article provides researchers, architects, and practitioners with a unified reference framework for understanding and advancing the state of datacenter networking.