I’m not saying I agree with the comment above you but Kubernetes upgrades and keeping all your addons/vpc stuff up to date can be a never ending slog of one-way upgrades that, when they go wrong, can cause big issues.
It's been a while since I spun up a k8s instance on AWS, Azure, or the like, but when I did I was astounded at how many implementation decisions and toil I had to do myself. Hosted k8s should be plug-and-play unless you have a very specialized use-case.
Last I checked, managed k8s clusters weren't much more expensive than the compute they ran on.