Latency will be higher. We are considering if we do go forward: a) Partnerships with smaller providers where there is an option to colocate equipment b) Colocating as close to as possible to targeted providers.

The main thing we are trying to figure out - if there is a need for service that has significantly higher tears then what Azure, AWS offers.

I would imagine people who need either more AWS can supply (or cheaper) but cannot pay for it, and also haven't figured out how to run it themselves, to be a very limited set. Happy to be proven wrong though and curious about what you will find out :) Is there a way to keep in loop?

The thing that is impractical to do on premise is "transparent scaling" say your load increases and you need a way bigger box that's the kind of problem the service can solve (take slave offline remount storage on a bigger box catchup on replication switch to slave repeat for former master). We are at the early stage just investigating if we should move forward so @andre1sk on twitter would prob the only place :)