M1 with 8GB runs 7B models. M4 Max with 128GB loads 70B+ that need multi-GPU on PC. Unified memory sizing, MLX vs Ollama speeds, and Mac Mini as an always-on AI server.