Does he think that the demand for AI-accelerating hardware is just going to go away? That the requirement of fast, dedicated memory attached to a parallel processing/matrix multiplying unit (aka a discreet GPU) is just going to disappear in the next five years‽
Maybe the idea is to put it on the CPU/NPU instead? Hence them going so hard on AI processors in the CPU, even though basically nothing uses it.
That, and people don’t know how to adjust them, or are unwilling to. My parents’ cars have a dial to adjust the headlight angle for when carrying weight in the back of the car, or when towing, but they never touch the setting.