On GCBD's servers in China. Possibly on Apple's servers in the US if they are running a wiretap. Due to the way key distribution works for iMessage, it is trivial for Apple and GCBD to do so.
Your message, through several layers of indirection, relies on a security conference paper from 7 years ago[0] + the assumption that Apple haven't updated the protocol in 7 those years.
No, my message relies on the fact that people have been looking at iMessage for years, and nobody, least of all Apple, has said that the implementation changed in any way to prevent Apple from viewing the messages.
Here is another article from 2016, which shows that Apple patched iMessage to prevent attackers who don't have access to Apple's servers from reading the messages but still kept the ability to read the messages themselves. https://blog.cryptographyengineering.com/category/imessage/
Apple was aware that people knew it could decrypt iMessage messages this entire time, but Apple made no changes that would fix that. That should give you some idea of whether Apple intends to ever fix that.
Apple can, of course, do whatever it likes, up to simply recording the screen and sending that to weird & wonderful government agencies. Like almost everything in mainstream security, it comes down to who you trust. It doesn't mean it isn't E2E though.
E2E encryption simply means that messages are only decrypted at the endpoints. That certainly isn't true of iMessage in China, and it might not even be true for some users in the US — we have no way of knowing because the protocol makes no guarantee against it.
So basically the first and second parties themselves need to do all encryption and decryption without any help from the third party running the service. Which is the age old usability issue famously holding back the casual adoption of PGP. Hard enough with text... To do it with video conferencing would be quite the feat. Someday, though.
[0] https://blog.quarkslab.com/imessage-privacy.html