That seems a bit silly to my eyes, self-hosting a server seems sufficient. But not included in this bill, so not an issue here
> Think if fortnite died tomorrow how many different servers it would take to host. Could an argument be made that an end user couldn't be expected to launch a dozen aws services? More dev time, more costs.
In this specific case, it's not so hard to imagine a single home computer handling the traffic of 100 connected users for a game of battle royale, the server compute for those kinds (baked-in world, low physics) games can be cheaper than running an instance of the game. Just some physics calculations, networking, and game state.
The main point would be if you start development from the premise that your server executable will be released to the users, the architecture/performance considerations are not that different at all.