At the moment: because Intel predominantly makes CPUs with a less optimized and flexible instruction set and physical architecture vis a vis the demands of the ML circlejerk, and Nvidia predominantly makes GPUs that effectively define state-of-the-art for ML training and inference (read: they are absolutely printing money, and are pretty much the only company that is actually making money hand over fist, even taking into account the circular financing deals they have with some other megacorps).

Put it this way: when you pop a bubble, it doesn’t actually pop all at once. Intel is closer to the site of the initial bubble breach. Nvidia is basically on the side exactly opposite of the breach.

Patience, grasshopper.