Actually you have a port on an OLT running at 622 Mbps downstream, 155 Mbps upstream, shared by up to 32 subscribers using optical splitters. At least that's how it is on my circuit which is wired for BPON. Some of them are getting upgraded to GPON which is 2.4 Gbps downstream, 1.2 Gbps upstream, shared by up to 64 subscribers using optical splitters, and the wire protocol is Ethernet instead of ATM. (Telcos took a long time to realize that no one cares about ATM anymore except for them.)
When IPv6 first started showing up, and was slow to be deployed, I often said that they should have just made IP addresses variable-length, and simply stick with 32-bit addresses until we got to a point where everyone was upgraded to new software. Then we could start using longer addresses. I envisioned an Internet where addresses were hierarchial, so if for example you had an address 169.254.0.1, you would automatically be routed addresses below it, such as 169.254.0.1.1, 169.254.0.1.2, 169.254.0.1.3, etc, and then the node at 169.254.0.1.3 would also automatically become the router for 169.254.0.1.3.1, 169.254.0.0.3.2, etc.
Yes, I am smarter than the entire IETF, but we're stuck with their design now.
Well, it turns out that I'm not the only one who thought this way. Here's an undated piece [ http://mercury.lcs.mit.edu/~jnc/tech/book3.html ] from none other than J. Noel Chiappa, who proposed pretty much exactly the same thing. I don't know when he wrote this, because very little of the material on his web site is dated, but it's a good bit of thinking. SNMP and LDAP use hierarchies like this, with absolutely smashing success: no one person or organization ever needs to be assigned more than one node in the tree, because they can just keep carving it up forever.
(For those who don't know who J. Noel Chiappa is -- he is one of the great unsung pioneers of the Internet. He invented the router.)