In the early years, AI data center construction revolved almost entirely around the GPU. From large-model training to inference, the GPU became the most closely watched compute chip in the whole infrastructure stack, while the CPU mostly handled data preparation, task scheduling, and system management. Under this division of labor, server CPUs seemed to matter less for a time.
But as AI moves from large-model training and inference toward Agentic AI, things are changing. An agent no longer just receives a question and generates an answer. It has to query databases, call APIs, retrieve context, execute code, use external tools, and coordinate across different tasks. Much of this work doesn’t happen on the GPU at all, yet it requires continuous CPU involvement.
believes this will reshape data center compute demand once again.
At the recent AMD In-Depth event in Las Vegas, AMD positioned Agentic AI as an important driver of the next round of growth in the server CPU market. It expects the data center CPU market to grow at a compound annual growth rate of over 50%, reaching roughly $220 billion by 2030. Forecasts cited by AMD also show the number of active agents in production reaching 2.5 billion by 2030, nearly 80 times the current figure.