It looks like it's an issue with a missing cloud provider interface. The capi-controller-manager was looping on "Waiting for infrastructure provider to report spec.providerID", and it seems there's no cloud-provider deployment in the cluster. Is that something that's removed during an upgrade?