IPFS seems perfect for the purpose. Files are content-addressed and a request for a part of a file can resolve to any accessible peer that has it. A peer has it

IPFS seems perfect for the purpose. Files are content-addressed and a request for a part of a file can resolve to any accessible peer that has it. A peer has it if they have pinned it or if they have accessed it themselves recently so that it is cached on their node. This way, users share the load for popular files, donating both bandwidth and storage. Torrents also seem like a perfect way to donate bandwidth and storage, and the IA already uses those. The only problem with these approaches or generally any distributed file store is that someone will have to share a copy, so you'll mostly be donating bandwidth by storing a redundant copy of data that they likely have to store themselves as well to guarantee accessibility.