3 ms·

I don't know? For some reason, I'm thinking most of the money went to either inferencing costs or NVidia.