oblio 4 hours ago

I'm not sure I understand how B (H100s costs 3x what they did a year ago) follows from A (exponential decrease in GPU inference costs).