The financial gap is something AAPL could possibly have worked out with INTC.

On the innovation side, there's a HW block called the neural engine where the use cases are limitless on the generative side (super resolution, speeding next word predictions, speeding up dumb graphics compute, and many others). Also speech and gesture use-cases on the discriminative side could be on the list.

They would have had to work w/INTC to extend the x86 ISA. That IP would then leak into other laptop customers (eventually server customers) ... or build a discrete part which would slow things down and make the interaction clumsy.

AAPL may have server ambitions as well.

Why bother sharing IP with a laptop chip maker? Build you own if you know the ISA + accelerator you want.