Does anyone have a similar guide for scaling AI systems? I am afraid this architecture cannot scale to 10M+ users serving LLMs for example.