Question Help multiplayer scale?

Hello,
I have currently a pair of games online (social card games) and they are both authoritative dedicated servers (one in c# with udp and one in java using smartfox).
Currently I have almost 10000ccu on c# and almost 5000ccu on java on a dedicated windows servers (unfortunately somtimes there are hiccups of disconnections but logs are clear of slow queries and any real slow function).
I'm not a networking expert tough and I think this setup has flaws and I wonder how to optimize. Is the only and next step going multi-server? I do not like the idea of clients hosting their game sessions since I think it would be a full rewrite of all the client and server code and could have huge issues of troubleshooting players' problems. Does anyone have success stories about this road?

So... what could I do? I was thinking about splitting the games in areas.
Autenthication, friendlist/club management, personal messages on a server (maybe in nodejs? request->response->close connection?).
Maybe Matchmaking on a dedicated server (this server shouldn't have many connection live at the same time) that spawns game instances on a list games servers.
I should need a shared DB server to synch all persistent data too.

In the current implementation there are a lot of functions that know if an user is actually connected but with a setup with many server I wonder if this is possible anymore and how I can change features such as, "send a message to someone and update its 'message notification' if he's not connected"? Should every player poll a server for updates? How could be this scalable to 10000ccu and more?

So, where I could find any tips about implementing a better architecture? Any tip?
I have found presentations about very large architectures (such as instagram) but the number of users I have to deal with is much lower and I also have to think in terms of budget.Seems there isn't much documentation around about average CCU and the step from one server to many seems a bit of a mistery to me.