Enter the Network Supercycle: Preparing Data Center Networks for AI’s Next Wave

Data center networks are entering a "supercycle" driven by the rapid growth of AI workloads, which are expected to triple networking traffic within three years and increase 100 times by 2030. This matters because AI infrastructure planning can no longer focus only on computing power and energy, as networks are becoming equally critical to data center performance. The shift is driven by new types of AI work, particularly agentic AI agents that maintain high operational levels continuously and consume significantly more networking resources than traditional chatbot-style applications. Data centers are redesigning their networks into three distinct layers, called scale-up, scale-out, and scale-across fabrics, to handle the massive data movement required across thousands of processors that AI training and inference depend on.