[bigiain]

I wonder?

Unreliable but redundant commodity grade hardware pretty much allowed "cloud computing" to eat 99.9% of "big iron" sales.

I strongly suspect what Google does internally with software defined networking looks a lot closer to what these people are doing than the traditional "buy a small number of powerful-but-expensive entrenched vendor network hardware" approach.

This mesh is kinda the network equivalent of autoscaling or autorepairing cloud computing strategies that are so common these days. Once this grows to 2 or 3 IXP connections it won't surprise me much to find they stomp the traditional ISP model reliability-wise, and if they can build that "virtually", where they don't need much more than a bunch of somewhat transient mesh participants with a few medium reliability "super nodes" which are mostly completely remotely configurable...