AI WAN：从智算互联到智能体互联，开启IP广域网创新之路

Core Insights - The rapid development of artificial intelligence (AI) technology is driving the emergence of innovative applications, which pose new challenges for computing power and network infrastructure [1][2] - The transition from "connection empowerment" to "computing power empowerment" and now to "AI empowerment" in IP wide area networks (WAN) is being facilitated by national policies supporting the development of computing power and AI technologies [1][2] Group 1: AI WAN Development - AI WAN is emerging as a high-intelligence, high-reliability, and high-security IP wide area network, integrating computing power networks with AI [2] - The need for a new type of AI network infrastructure is driven by the demand for intelligent services and applications, such as AI model training and inference [2] Group 2: Network Requirements for Intelligent Computing - Intelligent computing networks require lossless, ultra-high throughput, and ultra-low latency networks to address the challenges of large-scale GPU cluster interconnection [3] - The construction of a lossless network connection is essential, as even a 2% packet loss can significantly reduce throughput [3] - High throughput transmission mechanisms are necessary to handle the petabyte-scale data transmission in intelligent computing networks [3] Group 3: Transformation of Internet Paradigms - The advent of AI agents represents a paradigm shift in the development of wide area networks, enabling proactive demand discovery and resource collaboration [4] - The concept of an "intelligent agent internet" is proposed, which focuses on efficient connections and interactions among users, agents, and tools [4] Group 4: New Interaction Paradigms - The intelligent agent internet will give rise to a new interaction paradigm, termed multi-modal user interface (MUI), which integrates various sensory modalities for more natural human-machine interaction [5] - The collaboration model will shift from human operation to human decision-making, with intelligent agents executing tasks autonomously [5] Group 5: Protocol Innovations - Innovations in the transport layer are necessary to enhance throughput performance in wide area intelligent computing networks, as traditional TCP protocols face limitations [6] - The introduction of computing power routing technology aims to optimize network and computing resource states for low-latency scheduling [7] - G-SRv6 technology addresses the limitations of native SRv6, enhancing transmission efficiency and scalability for intelligent computing networks [8][9] Group 6: Physical Layer Innovations - Hollow-core fiber technology is identified as a core technology for AI wide area interconnection, offering significant advantages in reducing transmission latency and increasing capacity [10] Group 7: Application Layer Protocol Innovations - The intelligent agent internet will require new application layer protocols to manage diverse traffic types and enhance operational efficiency [11] - Key components include identity resolution, authentication, and routing mechanisms to ensure secure and efficient interactions among agents [13][14][15] Group 8: Network Management Enhancements - The integration of intelligent agents into network management can significantly improve efficiency and reliability by automating the translation of user intentions into network actions [16] Conclusion - The AI technology landscape is entering a phase of continuous business model evolution, necessitating further development of wide area networks to fully realize their potential [17] - The focus should be on integrating new productivity factors, enhancing performance metrics, and fostering new business models centered around AI training and intelligent agent interconnection [17] - The global landscape for intelligent computing networks and intelligent agent internet technology is evolving, with a need for continued development of autonomous AI WAN technologies [18]