Running a Character AI style LLM server involves significant costs across GPU hardware, memory, bandwidth, and oper