This makes no sense. This is just unsubstantiated speculation right now. There is no reason to believe that dumb AGI will hardware-scale to smart AGI for free.

This makes no sense. This is just unsubstantiated speculation right now. There is no reason to believe that dumb AGI will hardware-scale to smart AGI for free. In fact most machine learning algorithms have diminishing (logarithmic) returns with data and compute.