Start with the service, not the location.
An application that works well as SaaS may have no reason to move. A high-volume video stream may benefit from local processing. A recovery environment may need separation from production. Treat these as different requirements.
Map the dependencies.
Identify users, data sources, identity, network links, external APIs and support providers. Establish what happens if each dependency is unavailable. Local hardware does not automatically make an application independent.
Measure what location changes.
Compare latency, data volumes, availability requirements, data boundaries and operating support. Calculate whole-of-life costs, including connectivity, migration, maintenance and recovery testing.
Choose a bounded next step.
A small assessment or pilot can test a specific assumption before a broader infrastructure decision. Agree the success measure and which services remain in their existing environments.