[pdkl95]

Journals used to have several important roles: curation of articles, maintaining a reputation of quality (peer review, etc), and the actual physical publication and distribution of the papers. Cheap personal computers capable of "desktop publishing" and the internet made publication and distribution really cheap and easy. Those tasks no longer require a lot of expensive specialized skills and expensive typesetting/printing tools. This means journals need to stop treating those tasks as if they were still a scarce resource, and rework their business model around the tasks people still value highly: high quality curation and a reliable and trustworthy reputation.

The actual hosting of the PDFs (and TeX, and hopefully even the raw data) is something that universities or whomever the researchers are working for could host cheaply and easily. When I was attending UC Davis in the late 90s, the university hosted a huge archive that not only included their own publications, it also mirrored the publications of the other UCs and many important public archives like kernel.org.

Compared to huge archives of Linux distros and pre-GIT source code histories, hosting a bunch of PDF/TeX is effectively free. Reliable curation that saves a lot of people from wasting their own time and effort trying to find useful/interesting papers is extremely valuable.