IPFS
I'm really glad things like nerdctl and guix support it, but I wish more things too advantage of the p2p filesystem.
Petals.Dev and hivemind ml
P2P AI inference and training seem like the only true viable options to make foundational models that are owned soley by authoritarian government s and megacorps.
Matrix for federated general real time communication. (Not justs chat, video, images, but just data, with third room being on the cooler demos for what is possible)
Activity Pub for asynchronous communication between servers. The socialmedia aspect is obviously the focus and the most mature, but I'm also excited for things like Forgejo (Codeberg.org) and Gitlab's support.
I am also excited for QUIC for increased privacy of metadata and reduction of network trips.
The problem with IPFS is that kubo sucks. I used it for a while and it is always burning CPU doing nothing, failing to publish anything to the DHT and fetching files is slow. GC collects files that are supposed to be "pinned" by existing in MFS and so many other bugs all of the time.
I would love to see a new server take off that was built with quality in mind.
I think the core protocol is pretty good (although I think they should have baked in encryption) but even the mid-level protocols like UnixFS and DAG whatever are mired in constant change for no real reason as the developers jump on new fads.
Slow and requires additional tooling to run normally. Just not a lot of development on the core pieces tbh. Wasm support for example could make deployments way simpler (implement an ipfs proxy in any browser or server that supports wasm) but the ticket for that kind of died off. There is a typescript implementation, helia, that I haven't checked out yet.
We are honestly kind of in a decentralization winter again, with ActivityPub being one of the few survivors gaining traction from what it seems. OpenSource luckily doesn't just up and die like that, so I still have hope for some next spring.