The CPU is back: Rethinking the CPU-GPU split for LLM inference

Why agentic AI is driving the shift back to CPU inference.