AI BriefingAgentic AI Drives Up CPU Demand
Key point
Forecasts suggest that the spread of agentic AI could increase the CPU share in AI infrastructure.
Details
AMD, Arm, and Microsoft forecast that the spread of agentic AI could significantly increase CPU demand in AI infrastructure.
At OCP APAC 2026, the SVP of Tai Computing and Enterprise AI at AMD explained that agents increase CPU load by additionally executing orchestration, search and retrieval-augmented generation, and tool-calling alongside GPU tasks. Consequently, he stated that the existing CPU-to-GPU ratio of 1:4 could approach 1:2 or 1:1.
The regional representative for Taiwan and Southeast Asia at Arm stated that AI agents can generate up to 15 times more requests than humans. He explained that as agents run without interruption and call other agents, CPUs could become a bottleneck in current infrastructure.
Microsoft introduced the Cobalt 200, targeting agent workloads, as an ‘agent-native CPU’, highlighting that it reduces agent call latency by 33% and increases throughput by 23%.
If these forecasts materialize, the likelihood increases that CPU, memory, storage, and networking will need to be expanded alongside the growth of agent-based AI services, in addition to GPUs.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.