Good analogy, but there's a key difference: mainframes were an institutional dependency, whereas the world's reliance on LLMs is consumer driven, ubiquitous, an uncapped (e.g. spend more on the same "loops"). Completely agree on the second point, though powerful local models are the inevitable next step, and they are arriving fast.