Lots and lots of parts of the networking stack (esp. when talking about both virtual and hardware interfaces) either don't support IPv6, or didn't until very recently. For example: OpenVPN didn't support IPv6 until last year, and even today, you're hard-pressed to find a commercial VPN provider that provides IPv6 support, despite the fact that IPv4-only means that there's a massive privacy hole when using it unless IPv6 is explicitly disabled[0].
Also, it's very expensive to move existing systems over without risking any potential downtime or performance impact for existing customers. This is one of the reasons that new and smaller providers (like Digital Ocean) are more likely to support it than the bigger names (and even Digital Ocean didn't support IPv6 until very recently).
I don't have direct experience with this myself, but this is what my contacts at AWS and other companies told me.
You can set up a MySQL database, which has an IPv6 address by default. The documentation[1] says
> Each instance has an IPv6 address that is free to use. To connect over IPv4, > you must explicitly assign an IPv4 addresss to your instance, which incurs a > charge per hour idle.
IPv4 sounded like a hassle, so I tried to use the IPv6 address from my Google Compute Engine instance. But that doesn't support IPv6. It's the oldest outstanding feature request [2].
(Perhaps other parts of Google's cloud offering support IPv6, I haven't used much else so far.)
[1] https://cloud.google.com/sql/docs/mysql-client#connect-ipv6
[2] https://code.google.com/p/google-compute-engine/issues/detai...
Cost/benefit. Your internal virtual machines don't need IPv6 addresses, only where you're terminating to the outside world (usually ELBs, but sometimes public addresses).
For the vast majority of AWS users, the status quo is sufficient.
$ dig AAAA www.netflix.com
;; QUESTION SECTION:
;www.netflix.com. IN AAAA
;; ANSWER SECTION:
www.netflix.com. 178 IN CNAME www.latency.prodaa.netflix.com
www.latency.prodaa.netflix.com. 47 IN AAAA 2620:108:700f::36d6:2a51IPv6 behaves differently than IPv4 (which is known and predictable) and that the network congestion tools need to catch up. You are in effect having to monitor and adjust core routers for double the number of packet types.
The network administrators are not against IPv6, it's just that it requires careful steps forward.