You would need to run the LLM on the system that has the GPU (your main PC). The front-end (typically a WebUI) could run in a docker container and make API calls to your LLM system. Unfortunately that requires the model to always be loaded in the VRAM on your main PC, severely reducing what you can do with that computer, GPU-wise.
Just a couple thoughts (I have a mix of 2.5Gb and 10Gb):
Mikrotik switches are a nice alternative to Unifi. Much less lipstick on the UI but reliable and fairly priced.
If possible, you’ll probably want to use your own router rather than the all-in-one provided by the ISP. In my case, the router provided to me (Eero brand) did not even have a port fast enough for my service, and would have been an instant bottleneck.
Options for 10Gb-capable PCIe adapters (what you might put in your server or desktops) are more limited (at least they were when I transitioned a couple of years ago). Intel-based network adapters seem to require less effort to get working (driver-wise) vs. some of the other 10Gb / SFP+ capable adapters.
Finally, you are correct: nobody needs an 8Gb internet connection. Aside from well-seeded torrent file transfers, you will never reach that limit (and probably still never). And, you’ll need an adequate storage backend to write that fast.