that would be cool, the only problem is size of data, about half a petabyte of pdfs uncompressed (77TB compressed). OCR and indexing at this scale would be pretty costly.

Also papers do not follow any formatting standard, as a human that does not bother us too much, but OCR will have a hard time making sense of multiple columns with random text above and below, figure captions that are separated by the main text with only a little bit of vertical space, etc...

There are some projects that try to automate that, but so far all I have used needed some manual intervention.

Computing power might be donated by volunteers. If there are volunteer software engineers, there must be volunteers to run a container.