I think Amazon has reached the limits of Seattle -- Seattle can't build housing or transportation fast enough to match Amazon's growth, so starting somewhere fresh makes sense.

Plus there's the redundancy benefit -- if the Pacific Northwest suffers the catastrophic earthquake they are predicting, Amazon's HQ2 will live on (and make a fortune shipping supplies to the area).

Building a second HQ is horizontal scaling unless they plan to consolidate at the new location and continue vertical scaling from there. It sounds like they're actually distributing the HQ across two locations, not just building a specialized satellite office. (All of the mentioned companies have those.)

If computing works as an analogy, decoupling a system enough to spread across two servers without one being the true master/head is already half the battle. But you'd rarely design a distributed system only to run it on two servers. The synchronization overhead wipes out most of the gains. Once you have the architecture, you might as well move to commodity servers that are collectively cheaper and more powerful than two top-of-the-line servers.

I wonder if there is an organizational equivalent of the CAP theorem.