have opened this in Desktop but it straddles both Desktop & Service
real world: if someone begins with file samples of their data - let's say they are a set of csv files;
interested to get it data modeled and an initial set of visuals established in Desktop, then a Service Dashboard sampling..... kind of a proof of concept scenario...
then at this point - if they think it is viable - they want to connect to their actual data sources; which could be onprem ODBC or cloud based (let's assume the data source is always one recognized/supported in PBI specs)
The question is: Can the data source always be transitioned without rebuilding the data model / reports / visuals?
The same question asked from the other direction: Which PBI Service / Data Source connection combination can ONLY be built in Service ?
The pragmatic aspects of this is understanding that while there is a linear progression from Desktop Data to Desktop Report/Visuals published to Service Report and pinned Service Dashboard; however if the data source cannot transition then upfront this linear progression should not be the approach to use - - and instead - - identify what must be built initially & only in Service because of the connection to the final data source.