The future for data warehousing is column-store and in-memory tables. Most organisations with a BI store of under 2TB will not need to bother with complexities such as SSAS because in-memory can do the aggregations on the fly within acceptable response times.

All of the big 3 vendors now have variations of column-store and in-memory tables, and there are a number of enterprise-quality BI niche databases also with these functions.

This adds up to saying that any data warehouse based on a tabular database is no longer fit for purpose. This does not mean that it has suddenly broken. It means that tabular is to steam trucks what column-store is to diesel trucks. Both can carry stuff. But diesel can carry an order of magnitude more than steam for an order of magnitude less overhead cost.

One of the impacts we will see over (say) the next 4 years to 2018 is a drastic simplification of BI stacks. Most of the work we have traditionally done to give end users the performance they need is no longer necessary. In-memory can take the load and give the aggregates on demand. Another impact is likely to be puncturing the big-data bubble, as organisations realise that 2TB is too small to count as big data, and the shift to these techniques is not really needed.

Relating all this to the OP, as SQL2014 goes into RTM look out for what Microsoft announces for Azure. I have no inside track to Microsoft plans, but anyone who has had the time has seen what Hekaton can do within SQL2014, and can work out what capabilities this will add to Azure. Regarding whether you should host your data in your normal DC or in the cloud, treat both options as equally worthy from a technical viewpoint and look at what is most cost effective for you.