AI is a broad field. The same equipment runs ML, molecular combinatorics, factory robotics, symbolic processing, image analysis, and mega-multivariable processing. I don't even see LLMs as significant consumers of the available processing power. That's the foam on top of the ocean for the large buildouts. A use for the unused capacity leftover in between actual workloads. But it is the visible part. The fact that a digital twin of a bridge doesn't spark your imagination is an indicator of this.