DW for Amazon may be far different than we expect. See http://mashable.com/2010/07/27/amazon-facebook-recommendations/

My speculation is they have generalized the interaction between data streams and little DW type views. So someone likes Pink Floyd it would follow down a neural network of all things related to Pink Floyd, and tell you to buy Ummagumma for their birthday. Or maybe an axe if the friend is named Gene. Or http://news.cnet.com/2100-1023-976435.html

For example, they give recommendations based on items that I looked at, but didn’t buy – that means they run analytics of sorts on this data, which means they have it in some kind of DW. How did it get there? They don’t tell.

Even in social media, how are they doing business? The need to be able to tap into this super valuable information of connected human networks to be able to make decisions about ads targeting and etc.

That is possible with linked lists but very data heavy. “Fred” has a linked list of all the updates they expect to see. They query them by getting the most recent update, which links to the previous on, and so on.
But that means that when one of “Fred’s” friends posts an update it has to add that update to “Fred’s” linked list. So if “Wilma” has ten friends, any update means ten new entries in different linked lists. Workable for tens/hundreds (and indeed, it was the way in pre-relational days).

At twitter scale, where some people have hundreds of thousands of followers, that would get scary.