“Server orders are quoting ~6 months (it used to be a week or two), and prices are up ~10-20% since March.”
The AI buildout changed the ratio to roughly one CPU per four GPUs, and ordinary compute got scarce as a side effect. Server lead times went from two weeks to six months. Everyone who built on the assumption that capacity arrives on demand now has to forecast it. Average Kubernetes CPU utilization sits near 10%, so the headroom everyone bought is about to get expensive.