The distributed network idea has been around for a long time, and ideally it's how the web will go.
But it's much more of a technical and social challenge than today's server-based web.
To win users it has to be significantly better than what's available today - not just another way to do the same things, but with a few extra complications and unreliabilities.
I think IPFS is already sufficient for that, and indeed someone has already built a chat application.
https://github.com/haadcode/orbit
The underlying IPFS event database implementation:
https://github.com/haadcode/orbit-db
I think that there's something related to subscribing to changes without any central point of failure that's not currently possible with the IPFS implementations, but I also think that this is planned.
But we need something like a Decentralized Firebase. And that is exactly what we're working on at https://github.com/amark/gun . I met Tim Berners-Lee last year at the Extensible Summit and chatted for a bit, and it is pretty incredible to know that he is rock solid in his principles and values. That takes a pretty level head.
Issues creep up when you need to support dynamic resources, which rely on :
a) user input b) stateful server
With b), a will feel more comfortable sharing data when there's trust built up in b). This seems to go directly against legacy-web as centralization was the solution to the trust problem.
With new technologies like bitcoin showing trust can be based in mathematics, and ethereum showing that interactions can be based around mathematical rules, we definitely have the technological raw power to built a scalable, trustable, non-censorable alternative to www.
I do hope these new technologies are not too-hampered by how ubiquitous and entrenched www is.