In fact these people distribute content via deep caching and multicast provided by the ISP's and the CDN's - all those people (apart from Apple) have distributed infrastructures.
It sort of depends on distance, in reality it depends more on circuit noise and attenuation. If you have a really quiet circuit and high gauge high quality copper you will get a good signal to noise ratio which will give the modem margin to train to.
The curves used to calculate speed over distance make bulk assumptions.