[zinekeller]

I would want to say that IPv6 works, but as someone who manages an on-prem stack with workers that has dual-stack internet, surprisingly there are much more demand on IPv4 than IPv6 even though I've expected than reverse. Long-term connectivity are also preferred on IPv4 by a factor of around 7 times. I guess that the equipment that the ISP deployed handles IPv4 much better than IPv6 (which is telling of the state of IPv6).

[zamadatix]

I get why IPv6 only isn't an option I'm curious what about delivering conference services requires an additional 128k IPv4 addresses on top of the 128k+ they already had. Even ignoring terminating the HTTPS on a load balancing proxy via SNI I'm not sure what requires so many public IPs.

I feel like there is a service offering of there's I'm missing - usually you only need this many IPs when you're reselling hosted services.

[zinekeller]

Except for dedicated servers for certain enterprises (they started in enterprises so no surprise there), I think that STUN traversal in CGNATted areas may be the reason - Zoom has no control over how does the NAT machine on the other side works, which on some CGNAT equipments each outbound IP address are limited to a certain bandwidth (to evenly share the data throughput of the equipment). If Zoom has a limited range, then CGNAT users may suffer from limited bandwidth even though Zoom's network may handle this just fine. If they have additional IPs, they can give each participant an endpoint and circumvent whatever restriction there are on the ISPs' side.

TLDR: Except for companies requiring a dedicated proportion, probably to work-around CGNAT problems.