FWIW in our use case of non-GPU instances they launched way faster and more consistently on GCP than AWS. So I guess it is complicated and may depend on exactly what instance you are launching.