Taiwan makes all those chips, so no matter if it’s running in the UAE or Arizona, they’re Taiwanese chips. But in Taiwanese jurisdiction, we’re not training Grok 4 or whatever; we’re doing horizontal diffusion. For each application, we prefer a distilled, narrow AI good for that sector—like law or healthcare—not distracted by unrelated contexts like Ghibli studio paintings. General-purpose reasoning models are useful for sparse data or synthetic data integration, but you end up with something runnable at the edge, on your phone or laptop—not dependent on a mainframe.