AMD executives said Zen 6 was not designed specifically for agentic AI, but the architecture can still be deployed for those workloads. Their distinction is about design intent, not whether the CPUs can run agentic software.
What AMD says about Zen 6’s design
On November 10, 2026, Tom’s Hardware reported separate remarks from AMD chief Zen architect Mike Clark and corporate vice president Madhu Rangarajan. Both said Zen 6, codenamed Venice, was not designed specifically for agentic AI. Rangarajan said AMD had not anticipated the rise of agentic AI four years earlier; Clark said AMD had expected CPUs to remain part of AI software workloads.
Clark also said AMD could deploy Zen 6 for agentic computing. The distinction matters: a processor can serve a workload without having been designed around that workload from the start.
A processor’s original design target and the workloads it can serve are different questions.
Why agentic workloads can favor different CPU strengths
Agentic AI systems carry out multistep tasks, often by calling tools along the way. Those tasks do not all place the same demands on a server. Coding agents may make more CPU-intensive tool calls than web-browsing agents, while agents that query databases may benefit from strong memory performance.
That variation underpins AMD’s case for a broad CPU portfolio rather than one processor design centered on a single agent workload. It also means that a claim about one task or server role does not automatically settle how a CPU will perform across other agent workflows.
What the Venice comparisons say
Tom’s Hardware characterized AMD’s 256-core, 512-thread EPYC 9996 as a high-throughput option. It described the frequency-optimized EPYC 9686F as going toe-to-toe with NVIDIA Vera in an AI host-node role. That is a qualitative comparison for that role, not a universal ranking of processors for agentic AI.
NVIDIA takes a different design-positioning approach: its Vera announcement calls the CPU purpose-built for agentic AI and reinforcement learning. NVIDIA also claims Vera is twice as efficient and 50% faster than traditional rack-scale CPUs. Those figures are NVIDIA’s comparisons, while AMD’s statements address Zen 6’s original design intent and its use in agentic workloads.
