For one thing there’s also a kind of “de-replication” with IPFS, because data sets (incl. websites) sometimes contain copies of the same files, and a node will only store these once. So there’s already a vast amount of data replication on the web, and IPFS can help in this respect. Large data sets can shrink immensely with this approach.
But I’ve had thoughts similar to your friend’s about the IPFS: one of the explicit goals of the IPFS is to create a web that archives itself. However, to have archival permanence, current content and all previous stages of that content need to be stored on a lot of nodes, not just on the originating node, so it will essentially be (in macOS terms) a global distributed Time Machine volume. This will gobble up an insane amount of storage over time, even with “de-replication”.
Every millisecond, stuff will be added to the IPFS, chat logs, emails, websites, scientific data sets, TV live streams, warez, photos etc., encrypted, unencrypted, as raw data, with filenames etc.
A possible solution lies within IPFS itself, indirectly. Currently, a content provider needs lots of server space, either his own, or rented somewhere. With IPFS, a user who accesses the content, will cache it on his local node (maybe even pin it), which reduces the provider’s need for big server solutions, so with IPFS, capacity previously used for “legacy storage” will be freed, which can in turn be used to strengthen the IPFS in general, used for archives etc., and also to earn some filecoin. Server space providers like Amazon and others will probably be the ones who will profit the most from this development, because they already have the necessary infrastructure. So the IPFS is (in my view) not really a disruptive technology like cryptocurrency, but an organic ®evolution. The players just need to get a taste of it, and then the thing will probably take off.