Sure, but even heavily discounted Nvidia chips won’t be competitive for inference if they’re worse on perf/W.