First, to the control cost and performance of the connection between Google large data centers. If (say) AT&T wanted to squeeze more money out of Google, and/or hobbled performance by “prioritizing” traffic, the Google folks have an option for routing traffic around AT&T. Even if you do not use the option, the fact that it exists gives you leverage … which may be crucial given AT&T recent notions.

Second, to reduce latency. In the long run this is a very big deal. Network latency can cripple the performance of interactive applications, and web applications are getting more interactive over time.

Do a ping and traceroute to mail.google.com or docs.google.com. What is the round-trip time (latency), and how many intermediate hops? ISPs and the big carriers are (at best!) weakly motivated to reduce latency. Google is strongly motivated to reduce latency. Low latency allows their web applications to become more compelling.

For interactive web applications, you want the fewest hops from the server to your desktop. If Google can run direct lines to the regional centers of big ISPs (Cox, Comcast, etc.) and many smaller ISPs, then Google web applications have a large advantage over their competitors for very large segments of the market.

To me at least it seems unlikely Google would want to get into the business of covering the “last mile” to end users. That is an entirely different business from Google’s core, and could foul up the relationship with the big ISPs.