AWS Curbs Internal EC2 Access as Agentic AI Pushes CPU-to-GPU Demand Toward 1-to-1
Updated
Updated · Tom's Hardware · Aug 7
AWS Curbs Internal EC2 Access as Agentic AI Pushes CPU-to-GPU Demand Toward 1-to-1
1 articles · Updated · Tom's Hardware · Aug 7
Summary
AWS told engineers in May to cut EC2 CPU waste, and some now wait days for instances that had typically been available within hours.
CPU demand has surged as agentic AI workloads shift data-center needs from traditional 8-to-1 or 4-to-1 GPU-to-CPU ratios closer to parity, with tool calls and orchestration running heavily on CPUs.
Spot capacity appears to be the main pinch point: a consultant told The Information contracted AWS capacity has not seen shortages even as internal and external demand rises.
Amazon disputed that the move reflects new constraints, saying EC2 efficiency drives are long-standing and that it still meets the overwhelming majority of internal and customer compute needs.
The squeeze highlights a broader industry turn toward CPUs in AI infrastructure, with AMD launching Zen 6 Venice for data centers first and Nvidia elevating its Vera CPU pitch.