It's an old cliché, but car analogies rarely work for explaining computer phenomena. You won't be running a 1000 little kubernetes cars, and that's what k8s approach is all about running thousands of disposable little servers at Google scale.

With any car it's about making that single car more reliable or performant.

K8s doesn't care about reliability of any single instance, just the uptime of the whole service