The short version is lockin. Unless you get very careful ops engineers making it truly cloud-agnostic, you can't even move from AWS to GCP without a huge amount of overhead.
I'm a contractor and several of the companies I've worked with recently have almost hard-coded their services to expect certain things available to them on GCP or AWS that aren't as easily available on the other.
You can argue that they should use 12facter and real DevOps and portability and so on but that's just not what people are actually doing so they've locked themselves in and for companies that are struggling to pay their hosting bill already, saying "We need 3 months of 6 developers' time to move" isn't going to go down well.
This is why we choose to sit everything on top of OpenShift, we get Red Hat support 24/7 to run it and if we needed to migrate to a different provider its far less painful as OpenShift abtracts the cloud providers API's away from us. The only real changes we need to make our admin based. Technical changes are very minimal.
OpenShift costs quite a bit. Have you compared, costs- and ROI-wise, the OpenShift approach with an alternative of having a small team of consultants (or internal team, if available) architect your platform in a cloud-agnostic way, based on vanilla K8s, plus required additional OSS packages?
No matter how careful you are to avoid lock-in, infrastructure has weight. You have data, network infrastructure, security audits, IAM permissions, Infrastructure as code (no Terraform is not a magic bullet), training, DNS entries etc.
I wonder how many people who think they have a large “cloud agnostic” infrastructure have actually sat down with project managers and come up with a project plan and cost analysis of a large migration.
I have witnessed how much it costs to do just a Workday migration.
If you've invested the effort and personnel in being able to move from some cloud provider to some other cluster/cloud solution, it probably already makes sense to just get your own servers and move yourself there, since you're basically capable of doing that (and your costs will be cut by a large factor).
