Here's the question, if the big labs go bust some how, can all of the excess capacity be deployed for full on inference profitably?