AMD’s new open rack architecture rethinks the balance between CPU-driven agent execution and GPU inference, raising important questions about power, cooling, networking and the economics of next-generation AI data centers.