Not if it's 5-10x slower than a remote inference server. Mac prefill latency is exhausting.