Lets suppose you are launching a new line of business. It's a bit of a punt. You hope it will be wildly successful but equally it might not.
Do you invest in new hardware, licencing, resilience and failover capability for everything needed to operate this new line of business?
Perhaps you should spin it up in the cloud, determine what the demand is. Run it long enough so that the natural ebbs and flows of business are determined and you have genuine sizing and capacity metrics then bring it back in house.
Where I work, we punt new databases occasionally, often times working from draft requirements and not knowing what the eventual scope and user base will be. Most of the time these would be a staging databases for injesting some new file format from a client or a datamart for a new executive reporting request. Right now, this type of database would get initially deployed on either a new virtual server instance or perhaps the database would be contained alongside another somewhat related database on an under-utilized physical instance. After a few months it would get scrapped or shuffled into a more permanent location. These arn't typically transactional database with an application front end, they're bulk loaded, heavily crunched on, and accumulate at least several GB of data per month. Also, they would typically contain PHI. So, a 3rd party cloud hosted database solution wouldn't be a good fit.
However, at some point in the future the organization could decide to host it's own cloud based database farm to serve all of the global offices. That could feasibly happen. But that would also be a lot of infrastructure and architecture investment.
It's good that SQL Server now has a cutting edge cloud based deployment option and services, but at this point it's not a paradigm shift. Just like NoSQL, it meets the demands of those with a niche need.
"The universe is complicated and for the most part beyond your control, but your life is only as complicated as you choose it to be."