SciHub is costing around $50k a year to host 50 million papers for free. Elselvier makes a 37% profit margin on 2.07 billion pounds a year. Funny enough, SciHub shows Elselvier could host all scientific knowledge for under 1% of 1% of its revenue while making an operating profit. I'm sure we can find a better trade between widespread access and profitable management of that access than Elselvier. ;)
But no publisher is arguing that the hosting of PDFs is the expensive part of their business. If it was then you're right and researchers should just send their finished PDFs straight to sci-hub for hosting. But hosting a PDF is only a tiny, and as you've identified, an inconsequential part of the equation.
The other critical parts still don't result in a 37% net margin as it's basically labor. Most of it is low-skilled labor that should have plenty competition. The peer reviews of specific fields is high-skilled labor with decent to uncommon competition depending on size of field.
My claim still applies even if those costs are factored in. A yearly fee should be enough to cover everything. Charging $10-30 per article and such is just profiteering.
