It is a fundamental issue of volume. Unless we assume that backbone providers and Internet companies habitually over-provision, then it would take about the same size pipe, and the same size data centers, to "copy" the Internet in real time.

Consider how big the Internet is. Even if the NSA has 5 100,000 square foot data centers, think about adding up the aggregate data center footprint of Microsoft, Facebook, Apple, Google, Yahoo, Amazon, Rackspace, etc, etc. That comes to a lot more than 500,000 square feet. And this does not even get into the enormous data resources in Asia and Europe.

A much more likely scenario is that they are heavily filtering the data in real time and keeping just what looks useful or suspicious. That is still scary, but less fantastic than the idea that every packet is getting stored by the NSA.