After all, practically everyone is behind NAT these days, and it's so rare for an application to ask you to open a port in your firewall, most users probably don't know how to do it.
And early videoconferencing was famous for its unreliability - even simple things like people being able to join calls. Particularly as users might be behind restrictive firewalls on corporate networks or cell phone connections. Video calling that only works 95% of the time isn't good enough for things like job interviews :)
And users don't only expect one-to-one calls - everyone offers multi-party calls. If eight people want to watch Bob while he's talking, you've got to find eight videos' worth of upload bandwidth from somewhere.
And there are far more people trying to video call from battery-powered devices and metered connections than there were 15 years ago. If your competitors' apps uploads one video stream and yours uploads eight at once, users are going to notice the battery and data consumption.
And if you only have to test compatibility between your server and every version of your software, that's a much simpler task than testing compatibility of every possible n-way call between different versions of your software. Especially as you'll have to support Windows, OS X, Android and iPhone at the very least.