The breakthrough isn’t things moving faster but more fibers per cable.
No, it's actually more cores per fiber, and using those very well for space division multiplexing on top of the normal wavelength division multiplexing. They are talking about 22.9 Pb/s per fiber, not cable, the Tom's Hardware article is just wrong.
Cables can already contain hundreds of fibers, for example 576 here or into the thousands if you use stacks of ribbon cables in the subunits, for example 3456 here
For really useless call centers this makes sense.
I have no doubt that a ML chatbot is perfectly capable of being as useless as an untrained human first level supporter with a language barrier.
And the dude in the article basically admits that's what his call center was like:
So evidently good support outcomes were never the goal.