That’s the very goal of IPFS: hosting is distributed. Every node taking part in the IPFS network may or may not have pieces of this particular website.
Eth.link seems to be a centralized proxy enabling you to access that without an IPFS/EthDNS browser.
It should also be noted that what is hosted is a simple list of hashes with a name (and a size). The search engine is trivially looking through the name and displaying matching hashes.
Then, your bittorrent client try to find any torrent matching the hash (this is the magnet protocol).
So, yeah, you download something from nowhere. This is the real "cloud" we have been expecting for the last 20 years ;-)
IPFS only serves files, there is no server side compute. You can’t just scan ipfs like a database so there needs to be some other tech used for the engine itself, as OP indicates.
Eth.link seems to be a centralized proxy enabling you to access that without an IPFS/EthDNS browser.
It should also be noted that what is hosted is a simple list of hashes with a name (and a size). The search engine is trivially looking through the name and displaying matching hashes.
Then, your bittorrent client try to find any torrent matching the hash (this is the magnet protocol).
So, yeah, you download something from nowhere. This is the real "cloud" we have been expecting for the last 20 years ;-)