The actual marginal cost for a digital publication is negligible, as demonstrated by arxiv.org, archive.org, and sci-hub.
The reason why ACM and IEEE charge so much for digital library access is that they use the money for unrelated purposes.
As for sci-hub? They’re just taking the finished work. It’s like saying software shouldn’t cost much because The Pirate Bay can deliver software for free.
> > That process costs money because it takes people and time to do it well. Journal subscriptions are often not enough to recover those costs.
Some of the items left to consider here are actually putting together the reviewed papers and publishing them. I would say the cost of these two items has gone down with the extended use of computers over the last 3 decades.
What other items are left? It would be interesting to know.