A leading expert in the field has projected a significant transformation in global internet traffic patterns driven by artificial intelligence.

Wu Hequan, a member of the Chinese Academy of Engineering and former chairman of the Internet Society of China, shared these insights during a recent industry conference. He explained that large AI models have initiated a new technological era. These models are evolving into specialized agents, or "Agents," which integrate memory systems, tool utilization, and planning capabilities to tackle specific tasks. The convergence of multiple such agents, sharing tools and experiences across open networks, is giving rise to what is termed the "Internet of Agents." This shift is fundamentally reshaping the foundational architecture of the traditional internet, affecting everything from traffic flow to network design and service models.

Traffic Composition Shift

The analysis highlights a coming upheaval in the structure of both global and Chinese network traffic. It is estimated that by the year 2030, AI-related data will constitute more than 60 percent of all global IP traffic. Within this, north-south traffic flows are expected to be relatively balanced, while east-west traffic—which is highly sensitive to packet loss and latency—is forecast to far exceed north-south volumes. Focusing on China, the proportion of traditional conversational token traffic is predicted to plummet from around 50% in 2025 to just 12% by 2030. Conversely, traffic generated by AI agents and Agent-as-a-Service (AaaS) platforms is anticipated to surge to 75%.

Computing Power Evolution

Simultaneously, the primary application of AI computing power is shifting from model training to inference. By 2030, inference workloads are projected to account for over 60% of total AI computing demand. This new paradigm, characterized by massive data volumes, ultra-low latency requirements, and high bandwidth needs, is directly driving the evolution of the internet towards an AI-native infrastructure.