I'm two years out of date on what's going on with IPFS, but assuming the core ideas didn't change: that Twitter comment is technically right, but with caveats.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.