AI agents open a new growth opportunity for AMD server CPUs

AI agents are increasing demand for server CPUs as workloads expand beyond GPU-based inference. AMD enterprise executive Derek Dicker says processors must orchestrate tools, APIs, databases and execution environments across complex tasks, pushing the balance of CPUs and GPUs towards a near 1:1 ratio in some deployments. That includes additional CPU servers for agent execution, while GPU servers can still use multiple GPUs per CPU.

AMD is targeting that demand with Helion, a rack-scale infrastructure blueprint. Its CPU-only configuration uses sixth-generation EPYC processors and supports up to 32,768 CPU cores. Another configuration combines EPYC processors with Instinct MI350P GPUs to support agent workflows and inference within the same rack. The Venice processor family targets both high-performance AI infrastructure and general-purpose enterprise workloads.

Competition is intensifying as AMD gains ground against Intel and NVIDIA expands through Arm-based CPUs. Modal reported stronger agent-task throughput from NVIDIA Vera than from its existing Intel and AMD systems and plans to add Vera alongside its existing fleet. AMD claims its newer Venice processors outperform Vera in SPEC benchmarks, but independent testing with matched software and realistic agent workloads remains necessary.

AMD also sees server consolidation as a way to free constrained data centre capacity. The company estimates that refreshing older infrastructure with sixth-generation EPYC processors could deliver a 13-to-1 consolidation ratio, releasing rack space and power for AI while reducing cooling and infrastructure complexity.