Templates deployed when installing CNV on arm64 are bound to amd64 via the spec.template.spec.architecture field set to amd64.

Therefore, users cannot 'quickly' create VMs based on those templates on arm64 nodes, as expected.

When using containerdisks, we may use manifest-list images as support.

When using qcow2 image links, should we consider one annotation per architecture so that, once a VM is created and scheduled on a given node, KubeVirt's components can set the proper qcow2 link based on the architecture of the node?

Assuming the previous points are solved, can the architecture field of a template be set empty to let that template be arch-agnostic? 

See https://gist.github.com/aleskandro/48d5f2b52483271ebb4fe0276d51fa21

1.
2.
3.