AMD’s CTO: Agentic AI Needs More CPUs Than GPUs
AMD’s stock has roughly doubled in six months, and its market value is closing in on a trillion dollars. Its CTO, Mark Papermaster, attributes this shift to a lesser-noticed trend: agentic AI requires significantly more CPUs than just GPUs.
The Shift in AI Processing
On stage at the RAISE Summit in Paris, an interviewer joked that investors should have bought AMD shares six months ago. Papermaster explains that AMD’s leadership in server CPUs, dating back to 2017, has positioned it for this moment.
Papermaster argues that running agentic applications actually relies more on CPUs, not less. He cites a figure indicating that AI workloads require roughly four times as much CPU processing as traditional methods. The reason lies in the complexity of orchestration: managing multiple agents, deploying sub-agents for specialized tasks, and maintaining context across a growing number of interactions. This coordination layer, crucial for agentic AI, operates on the CPU.
AMD’s Experience and Innovations
Papermaster, with a career spanning PC development, IBM, Apple, and cloud computing, is well-positioned to observe this shift. He emphasizes that agentic systems represent a significant leap forward, enabling machines not just to look up answers but to complete actual tasks through the combination of steps.
AMD’s New Business Model
Consequently, AMD is evolving its business model beyond chip sales. It now offers optimized systems, focusing on efficiency across the entire stack. This includes the acquisition of ZT Systems, a hyperscale infrastructure builder, and the development of Helios, AMD’s rack-scale AI system, which integrates 72 CPUs into a single powerful unit.
Papermaster’s vision is clear: to design for the system as a whole, from application to hardware, marking a new era in AI processing.