We are currently not or barely covering our costs with Patreon AI subscription (depending on the traffic), as hosting an AI which is fast enough for real time conversation costs at least in the range of 500-1000$ / month, and subscribers for the AI tear are less than 100.
For physic issues: These are fundamental issues we constantly work on, but current SOTA physic engines have some limits of what you can do. And changing engine does not help with this, as there are no real alternatives on the market. Unreal Engine is slowly getting usable, as they are implementing their own engine, but we are using Unity like most Indie devs, which is using PhysX from NVIDIA.