3 ms·until its cheaper to train and infer than 100k gpu data centers...i doubt it will ever compete.by buffer_overlord 4mo agountil its cheaper to train and infer than 100k gpu data centers...i doubt it will ever compete.