Agentic workloads push CPUs into the AI infrastructure race

AI infrastructure spending is shifting from a GPU-led build-out toward a CPU-GPU model as inference and agentic workloads grow. Agentic systems split goals into iterative steps such as querying databases, calling APIs, executing code, invoking GPU clusters, checking outputs and looping, with most orchestration running on CPUs. Company announcements expect the server CPU market to grow from $30-35bn in 2025 to more than $200bn in 2030, implying a CAGR of more than 45%.

Demand is already straining supply. Intel reported 59% YoY growth in its data center and AI segment in Q2-26, driven by a 9% rise in server volume and a 48% increase in ASPs. AMD expects 70% YoY CPU revenue growth in 2027 and is targeting more than 50% server CPU market share by 2030, supported by its EPYC Venice ramp at TSMC on 2nm.

Competition is widening beyond Intel and AMD. Nvidia is moving from bundled Grace CPUs into merchant sales with Vera, while Arm has entered with its own AGI CPU and Qualcomm plans Dragonfly C1000 production in H2-28. Hyperscaler chips from Amazon, Microsoft and Google add pressure by serving internal cloud workloads. By the end of 2028, the merchant server CPU market is expected to expand from two vendors to at least five.