It makes very little sense. I mean, sure, maybe try to avoid vendor specific databases and use stuff like rds/managed cassandra/whatever vs Bigtable or dynamodb, but it's
MUCH cheaper to lean in as hard as you can to one cloud provider and just pay for a month or three of consultants to move you if you ever actually want to than it is to build that in from the start.
k8s is a nice way to get there, if your k8s is running on gcp, aws, or azure it doesn't really impact your pipelines or process at all.
Things are harder at the bleeding edge, I'm working atm on a site that's all in on aws lambda with server less framework and aurora, kinesis and such, to the point where a migration would mean a rewrite tbh.. And that's alright for them, they're ok with their cloud partner.. And it's cheap too..