Has anyone considered the feasibility of using distributed technologies? Torrents can handle arbitrary file sizes and throughput, with no servers whatsoever.

People who download things simply "pay" for their use by providing identical service to others: sharing their brandwidth, their CPU share and their disk space (that one may be for free though, because you do need that library you've downloaded right?).

It's a "business" (resource sharing) model that works, proven by decades pf continuous operation.