In the truest sense of the word, if data for a given hash is on any computer, that computer can provide the data behind the hash.

If that data was found 1000 years from now, and entered into the system, the data would generate the same hash, and then be able to fulfill any hash requests.

Also, IPFS does data deduplication across blocks (chunks of 512KB - I believe), files, and directories. So if I share a source repo that uses someone else's repo, and they share it too, then my copy helps speed his files up as well.

There's an ongoing project to get OpenStreet Maps in IPFS, so that local areas have their own hash, all the way up to state, country, continent, and world. It's so the whol geographic repo could be shared amongst all, without the burden of storage on a single person or group.