A dataset at the scale of Scihub would have to pay someone to keep all the various PDFs online (which IIRC is around 70TB by now), which would mean a DMCA or N&T or similar would take down paid IPFS hosters.
I'm not sure how many people would willingly play IPFS hoster for 70TB datasets for free and while not giving a shit about authorities knocking on the door.
Plus sci-hub storage requirements are pretty big, >75 Tb...