I find it hard to imagine nvidia's moat not drying up - the hyperscalers already have more cost effective silicon and the AI labs already use a mixture of all the capacity they can get their hands on.