AMD Says Agentic AI Demands Open Data Center Architecture

Agentic AI is shifting computing beyond large-model training toward workflows that plan tasks, invoke tools and coordinate repeated inference. AMD says that change will force data centers to be designed as integrated systems rather than collections of isolated accelerators. Open standards and collaboration across hardware and software suppliers are becoming essential as operators seek flexibility, scalability and an alternative to closed, single-vendor infrastructure.

At the OCP APAC Summit in Taipei on Aug. 11-12, 2026, AMD outlined four requirements for agentic AI infrastructure: composability, verifiability, interoperability and deployability. The chipmaker said CPUs should handle stages such as data preparation and agent orchestration, while GPUs execute dense parallel computation, with resources assigned according to each phase of a workload. AMD disclosed no investment, purchase value or new financial commitment alongside the architecture guidance.