Still, GKE is a managed environment for Kubernetes and I was wondering how difficult it would be to take your workloads out of GKE after 2-3 years and run them on your own hardware.
My personal opinion is that, how closely you're integrating with storage systems (e.g. databases and blob stores) and APIs (e.g. machine learning, pub/sub, load balancing) of the cloud platform you are running on.
Kubernetes already does a great job abstracting the compute infrastructure and networking topology you have on the cloud provider away. There's a lot of work going on in the Kubernetes Service Catalog[1] to make it even more platform-agnostic. With Service Catalog, you will be able to say "I want a Redis instance" and your underlying infrastructure will be able to provide you a Redis instance, you will no longer worry about integrating to GCP, Azure, AWS separately. Service Catalog sounds a lot promising as a step towards eliminating vendor lock-in. That said, it is still a work-in-progress.
Note that none of this offer a migration path for your existing databases etc. But I am seeing that many businesses are able to migrate their compute workloads between platforms and on-prem to the cloud a lot easier compared to the pre-Kubernetes era.