That is a very good point. Why buy big iron "pets" when you can buy x86 "cattle"? It works for a lot of stateless apps and services. A node dies, and k8s quickly moves the workloads to new nodes.

It does take some work to accomplish this on more complex apps, though. Things like SQL databases and rabbitmq are very often single points of failure in practice. At smaller companies, it's often easier to stick them on more resilient hardware than to architect an active/active or active/failover system. I agree that this isn't the best way to do it, and IMHO any important service should have HA of some sort.

That said, I use x86 based machines for all my personal projects, and I wouldn't buy IBM systems if I owned a tech startup just because they're like 10x the price of x86.