Real-time notification scaling: Fan-out versus Fan-in

Hey everyone, I've been chewing on how to scale up real-time notification systems. We're getting tons more users, and the way we're doing things now, sending individual messages to every single subscriber, is starting to really max out. Has anyone here managed to move away from that all-out fan-out method to something smarter, maybe a fan-in or a mix of both? What were the big hurdles you ran into, and what did you have to change about your system's design? I'm really interested to hear what worked for you when handling lots of real-time updates at once.