> Content addressing: IPFS uniquely identifies immutable content stored
  > on the network. This approach simplifies storage and reference to the
  > underlying data, as a node can use a unique key to unambiguously fetch
  > content from its peers.

This is unfortunately not true -- IPFS uses content addressing for metadata, but file content isn't content-addressed. The result is a network similar to BitTorrent's DHT, where if you know the hash of the torrent file you can download the content, but having a hash of the content doesn't get you anything.

An example I posted previously (https://news.ycombinator.com/item?id=32426051): debian-10.7.0-amd64-netinst.iso has SHA256 checksum b317d87b0a3d5b568f48a92dcabfc4bc51fe58d9f67ca13b013f1b8329d1306d, but a very large number of CIDs could be used to identify that file. The following two CIDs are both valid identifiers for it:

https://cid.ipfs.tech/#bafybeihjy54iyvheotna2aeqmzhqnro6yot4...

https://cid.ipfs.tech/#bafybeihfqpypuhmtyzazrj3g4b4f4nqk2ziy...

You'll only be able to download that file from IPFS if you know which CID to use, and if there are peers hosting it with the same CID.