I've had this same question myself for a current project we are working on. We decided to go the Azure Analysis Services route for several reasons.
The most obvious to management is cost. PBI Premium is expensive and can be hard to justify depending on what you are doing. For example using an B2 AAS instance gives you 20GB of memory and is ~$630/mo. This could support many power bi pro users before it would make sense to switch to premium from a cost perspective.
However, as an architect, the primary reason we went the AAS route is the seperation it provides. In our situation we will have power users with admin priviliges in the PBI environment. This means that if we had PBI premium and therefore were storing our data in Premium that the power users could modify or delete the data either intentionally or accidentally. This would hose us big time and it's not something we are willing to risk. By putting the data in AAS it allows us to separate the power users from the data and only allow them read access to create reports to their hearts content.
There are also those who do not care to user power bi and prefer excel connections and so they can directly connect to the AAS instance without having to have any interaction with PBI.
Hope that helps a little in your decision making.
