copied from a deleted reddit account, it's quite brilliant :
ELI5: Whole thing kind of works like a virus
So, the client doesn’t know where your file is. They call a primary server for a list of some people connected to the IPFS network, then ask around that list progressively spreading, like a virus transmitting between hosts
Once the “virus” finds the file, it brings it back to the client, stores a copy, and closes the connection. And since the client is now IN the network (knows where some other people are) then in order to retrieve further files they don’t need to use the primary server any more. Also, other clients can now fetch the file from the original client. So every copy makes finding those files faster — like auto scaling.
IPFS built an algorithm around that concept, to make file cleanup, lookup, minimization, and integrity checking possible. They also use a hash ID system to store data so like, if the hashes match, you could get a piece from file A or file B (whichever is closer) to complete the hash sequence needed to build file C.
It’s a pretty clever system, if you’re curious totally worth reading into the details.
Probably got a lot wrong, but that’s how I understand it.