I'm no accountant, but I imagine they can classify various Prime customers into buckets based on what services they use and see that the average usage of any one bucket of customers matches up with the cost of Amazon Prime. If a customer were to use ALL the features then Amazon loses, but if that became the new average then they'd raise the price.

The cleverness of this idea is that each Amazon Prime customer perceives the overall value of the service as higher for the potential to use those extra features, even if s/he doesn't use them much if at all.