Agentic AI Demands a New Infrastructure Stack: AMD and Red Hat Deliver
Red Hat Blog, Thursday, May 14th, 2026
AMD and Red Hat deliver new infrastructure stack optimized for agentic AI workloads using MI355X and MI350P GPUs.
Agentic AI systems that reason, plan, and execute multistep tasks autonomously require a fundamentally different infrastructure than traditional inference. AMD and Red Hat are collaborating to deliver an open, high-performance infrastructure foundation with AMD Instinct MI355X GPUs featuring 288 GB of HBM3E memory and 8 TB/s bandwidth for efficient large model serving.
Red Hat AI 3.4 now supports inference on the MI355X with ROCm 7 integration, enabling streamlined deployment on Red Hat OpenShift, while the MI350P PCIe card preview brings AMD CDNA 4 architecture to standard datacenters with passive air cooling. These solutions help enterprises run more agents per GPU with lower power consumption and improved cost efficiency for inference-intensive agentic workloads.