Yeah, and regarding using just raw IPFS hashes there's this comment further down the thread indicating not even that is enough (I know too little of IPFS to verify but seems legit): https://twitter.com/jonty/status/1372169695277760519
I'm two years out of date on what's going on with IPFS, but assuming the core ideas didn't change: that Twitter comment is technically right, but with caveats.
IPFS, as a P2P system, only contains the files individual users are hosting. There's no fundamental obligation to rehost other people's content - but there are various incentive schemes[0]. The address itself is a hash of file contents, so if someone were to reupload the exact same (bit-level identical) file to IPFS again, it would appear on the network with the same hash. The limit to long-term availability of any given file is whether anyone who has it is willing to keep it in the network.
--
[0] - For one, IIRC, there was some automated caching/mirroring of the files you pull using an IPFS client (not IPFS->HTTP gateway), to ensure popular files will get mirrored around the network for a while. Secondly, the team also started developing FileCoin, whose purpose was to create a scheme in which people could get paid for putting their spare storage on the network, which would be used to replicate files on IPFS. I don't know if they managed to get it to work.
My interest in IPFS started to diminish after I saw the team getting very involved in cryptocurrency world. I only care about IPFS for the P2P, content-addressable storage layer - and they kept talking about things related to Ethereum. So I stopped paying attention.
Moving IPFS to the crypto world has made it more of a success. I need distributed storage for an app I'm developing and I'm far more comfortable paying people to host the content instead of relying on goodwill.
IPFS, as a P2P system, only contains the files individual users are hosting. There's no fundamental obligation to rehost other people's content - but there are various incentive schemes[0]. The address itself is a hash of file contents, so if someone were to reupload the exact same (bit-level identical) file to IPFS again, it would appear on the network with the same hash. The limit to long-term availability of any given file is whether anyone who has it is willing to keep it in the network.
--
[0] - For one, IIRC, there was some automated caching/mirroring of the files you pull using an IPFS client (not IPFS->HTTP gateway), to ensure popular files will get mirrored around the network for a while. Secondly, the team also started developing FileCoin, whose purpose was to create a scheme in which people could get paid for putting their spare storage on the network, which would be used to replicate files on IPFS. I don't know if they managed to get it to work.
My interest in IPFS started to diminish after I saw the team getting very involved in cryptocurrency world. I only care about IPFS for the P2P, content-addressable storage layer - and they kept talking about things related to Ethereum. So I stopped paying attention.