Get the target deployment settled first
Method availability depends on the exact deployment. Settle it before comparing anything.
Method availability depends on the exact Autonomous AI Lakehouse deployment. Serverless, Dedicated, Exadata Cloud@Customer and the multicloud deployments do not offer the same provisioning routes, network paths or migration options, and support documented for one should never be assumed for another.
This sounds like a detail. It is not. I have seen a method selected, designed and half-built on the assumption of a deployment that the organisation’s network or sovereignty position then ruled out. Settle the deployment, the region, the network boundary and the provisioning model before you compare methods, and put the answer in writing with a named owner.
If the deployment is genuinely still open, treat every method choice as provisional and say so.