I honestly have no Idea what AI does to a processor
Parallel processing capability. CPUs historically worked with mostly-non-massively-parallelizable tasks; maybe you’d use a GPU if you wanted that.
I mean, that’s not necessarily “AI” as such, but LLMs are a neat application that uses them.
On-CPU video acceleration does parallel processing too.
Software’s going to have to parallelize if it wants to get much by way of performance improvements, anyway. We haven’t been seeing rapid exponential growth in serial computation speed since the early 2000s. But we can get more parallel compute capacity.