AMD says agentic AI is increasing CPUs’ role in enterprise infrastructure

AMD corporate vice president Derek Dicker said agentic AI workflows are increasing CPUs’ role in enterprise AI infrastructure.[1] He described a shift from CPUs’ use in general computing and as head nodes for GPU clusters toward agent orchestration.[1] That framing puts CPU orchestration alongside GPU acceleration in AMD’s account of how enterprise AI infrastructure may develop.[1] Dicker said agentic workloads could eventually require a one-to-one CPU-to-GPU ratio, describing a possible future requirement.[1]

AMD says inference workloads are expanding dramatically and may become the dominant share of compute utilization.[1] The company’s projection points to a growing role for inference, but the evidence does not specify when it might become dominant or how broadly that change will apply.[1] AMD says fifth-generation EPYC processors can replace as many as eight older servers with one EPYC-based server.[1] Separately, calculations suggest a 13-to-1 consolidation ratio for sixth-generation products.[1] AMD says server consolidation can free data-center space for new AI deployments.[1]

In a press release dated Oct. 12, 2026, AMD introduced Helion as an open blueprint for agentic AI infrastructure.[2] AMD says it is based on an Open Compute Project specification design and powered by AMD compute.[2] Its CPU-only rack, designed for hyperscale customers, can support up to 32,768 sixth-generation EPYC cores and connect agent execution to separate model-inference infrastructure.[2] An all-in-one configuration combines EPYC CPU nodes, Instinct MI350P GPU nodes and general-purpose compute.[2] AMD says Helion supports liquid- and air-cooled versions and can let customers use existing IT infrastructure for agentic AI.[2] The ET report separately describes Helios as a 72-GPU rack designed to work with ROCm and ecosystem partners.[1]

AMD and Tata Consultancy Services, through its HyperVault subsidiary, plan to co-develop Helios-based AI infrastructure in India.[1] The collaboration includes a data-center blueprint supporting up to 200 MW of capacity.[1] AMD says the Instinct Coder is the first in a series of such appliances.[1] For Helion, AMD plans to contribute the specification to the Open Compute Project for further development and adoption.[2] AMD says Helion configurations were under development with ecosystem partners and planned to be available in November 2026.[2]