Tim, this has been explained to you many times, but there is something you're not getting here. What you are either unwilling or unable to comprehend is the concept of the virtual circuit (PVC, also described as a logical network within a physical network, or can also be described as the WAN analog to a VLAN).
Let's use an actual example to make it clearer:
Suppose AT&T has a single OC-48 between Los Angeles and Dallas. There are multiple virtual networks within this one physical connection. One of the virtual networks is the public Internet, another is the U-Verse private network, another is AT&T's private business network, another is the AT&T POTS phone system.
Even though all 4 of those types of traffic travel over the same physical OC-48, they are logically separated from each other. The switching devices on both ends of the OC-48 are tagging each packet with the PVC information. On the other end, the tag is removed, and the packet is switched to it's designated hardware.
This means that if I was a hacker, even though I could get an IP packet from my computer onto this OC-48, I cannot hack into the U-Verse logical network or the phone logical network because my Internet packets are tagged with the PVC for the public internet. The switching devices on the ends will not allow the traffic to mix together.
The protocol that is used on each network is independent of the network's function. The public internet uses the IP protocol to transfer data. It just so happens that the U-Verse network also uses the IP protocol, but that was a decision made that reduced the cost of the RG and STBs. Off-the-shelf chips are available that process IP. If you were to use a custom protocol, you would need custom chips, increasing the cost.
On the business network, you don't have to use IP. If I'm a business customer that has a ton of old Digital Equipment Corporation mainframes that I need to link together, AT&T can provide me private WAN service that travels over their business network, and carries the DECnet protocol. Guess what? Those packets also travel on that same physical OC-48. And they don't interfere or mix with the other networks, and they don't interfere or mix with other business customers' traffic because of the PVC tagging.
Now, you might ask, WHY does AT&T split the networks into logical sections like this? Beyond the simple answer of security and isolation, it's also because the desired service and features of each logical network is different. An excellent example is to compare IP on the public Internet with IP on the U-Verse network:
Even though they both use IP, there are some features of IP that are being used on the U-Verse network that the public internet does not support. For example, multicast. Multicast IP packets are not supported on the public internet because not all internet service providers have multicasting turned on at their routers. Another example is QOS. QOS is used and required to be used on the U-Verse network to ensure timely delivery of TV traffic.
Your traffic from U-Verse (Internet, Voice, and TV) is eventually split into those 3 separate logical networks somewhere in the AT&T system. I don't know exactly where it occurs, but it could be as soon as the traffic hits the VRAD. Your voice traffic (in IP packets, by the way) could be switched to the voice PVC, the internet traffic onto the Internet PVC, and the TV traffic from the U-Verse PVC, all travelling on the same fiber that supplies the VRAD.
"The Internet" is not an ethereal backbone hanging out there in limbo somewhere. The Internet begins in your house at your computer. Every router, switch, and gateway between the source and the destination is considered part of the public internet. This notion that you have that traffic must leave AT&T's network to be considered "on the internet" is not true. If you type "www.google.com" into your browser, the very first packet that leaves your computer with a destination IP address of Google's servers is most certainly "on the Internet" before it even reaches the RG.
Plus, there are strange routes that IP packets on the public internet taks all the time. Let's look at another example:
Let's say I'm a U-Verse customer in Los Angeles, and I need to pull up a Yahoo web site that's served out of San Francisco. Normally, that traffic would stay on physical links that are all owned by AT&T, and would travel over AT&T's fiber between the two cities. But let's suppose AT&T needs to do maintenance on one of the routers that handle that fiber. Thus, that router is down. AT&T has peering agreements with other providers in both cities. Let's say AT&T purchases a short-term traffic routing agreement from Level(3). AT&T will temporarily route Los Angeles traffic destined for San Francisco to Level(3), and Level(3) will carry the traffic to San Francisco, at which point it will hop back onto AT&T's network. Now, temporarily, even though I'm an AT&T customer pulling up a web site hosted by and on AT&T's network, my traffic was routed over a non-AT&T network.
But none of that matters when considering whether that traffic was "on the Internet" or not. It most certainly was, at all points, regardless of who owned the physical links.
I hope this gives you more insight as to how "the Internet" works, and why the Investor Presentations that you keep referring to cannot be used as any basis whatsoever to infer the technical workings of the network. When those things say "one IP-based network" that is a gross oversimplification that is targeted at non-technical corporate execs. It is not technically accurate in the slightest.
