It is truly a tragedy that the story for scaling Kubernetes beyond 3k-5k VMs is "run multiple clusters'. It's 10% overhead for tiny scale, doesn't scale well to large scale- it works well in this happy midrange of 7-3000 VMs.

(If anyone is about to ask why you would need that many VMs - many companies who do extremely large scale infrastructure! The kind of software where when it breaks it can create a crisis for utilities, governments, healthcare systems, militaries, etc.)