The sheer size of IPv6?

[gdt]

A full /64 of IPv6 address space has been traded for the autoconfiguration feature. IMHO this is well worthwhile. But it does mean that analysis of the available allocatable size of IPv6 is best done using /64 subnets rather than single IPv6 addresses (unless you're expecting every MAC address that ever was to be skiving off your unsecured wireless there will be vast underutilisation of the least significant /64).

A rough back-of-the-envelope calculation involving regional registry allocation policies suggests to me that there are about one billion allocatable /48 IPv6 networks (ie, business scale network allocations). A lot. But also not enough. The implication being that ISPs shouldn't give a /48 to every customer. Rather bridging a /64 to smaller customers (small business, homes) looks sustainable.

The top /64 of IPv6 address space won't be particularly heavily used either. Of the non-reserved regions about a third of the space goes to heirarchical routing. That is in fact one of the concerns with the endgame of IPv4: that the increased utilisation of IPv4 addresses will see a huge spike in the IPv4 forwarding table size as we try to track every small fragment of IPv4 address space.

In short, graphics which attempt to illustrate 1 in 2¹²⁸ should be treated with caution. Network addresses have internal structure and this leads to much underutilisation. That's not waste, but trading address space for features: namely, nicer routing and autoconfiguration. In that sense the IPv6 address space is huge: with only the 32 bits of addressing in IPv4 it wasn't really possible to trade address space for any features beyond the essential.