The CPU Comeback Is Upon Us
TL;DR
Leaders at AWS told their engineers this year to conserve CPU cycles at all costs, after wait times for CPU server capacity exploded under AI workloads. CPUs had largely been left out of the AI story, since their limited parallelisation made them a poor fit for model inference while GPUs and memory dominated. The rise of agentic systems that run autonomously and call sub-agents is changing that. Analyst Matt Kimball reports a spike in CPU demand through 2026, much of it driven by agents.
Nauti's Take
This is an opportunity for infrastructure teams: planning CPU capacity and efficient agent architectures now buys a cost advantage while everyone else fights over GPUs. The risk is predictability, because if even AWS gets caught out by demand, wait times and price jumps are likely at smaller providers too.
Anyone running many agent loops should reserve capacity early and cap sub-agent calls hard.
Sources
