more practically, how much will it cost to run e.g. a Matrix synapse server to support a community of a few thousands (the size of a typical small website community). Since chat is really deadweight and typically unmonetizable, if this cost is too high it will not get traction
I "add" you to my friends. I send you something like a v-card: A blob of metadata saying who I am and what my devices are (with certificates). You get a notification, you see this information and you get to decide there and then what access (if any) I get. This information gets sent to all your devices and you send me back a list of devices and their certificates. You post an update. It gets sent to all your devices. My device does a round-robin attempt on your devices for updates. It sees new stuff, downloads and spawns a notification (if applicable). I'd imagine there'd be ways to make push notifications happen too (ie for chat).
That scales well with a sensibly sized pool of people.
"But I'm an Insta-model with 2 million subs and a business!" ... then make one of your devices a real server in a real datacentre. That's the good thing here, with multiple devices, you can throw as much or little as you like at this. People at the low end still get first-class service up to a point, for just their data.
So people like Facebook and Google can stay in business but only as federated endpoints to this system, charging to provide delegated access.