“While running local AI models can save on token costs, it introduces significant expenses for hardware, maintenance, updates, and specialized personnel.”