Porter, your enterprise-software post-mortem is excellent-especially the railroad/fiber analogies and your analysis of switching costs. But I think it explains why the trade broke, not fully why it became so crowded in the first place. The missing layer is reflexivity: Big Tech and chipmakers finance frontier labs → labs spend heavily on compute → neoclouds, miners, and power providers report surging demand → those results reinforce the “software is dead, compute is destiny” narrative → more capital enters the same trade. Leopold wasn’t merely analyzing that cycle from the outside. His public thesis and reported leveraged positioning helped amplify it. That doesn’t prove coordination or misconduct. But it does mean this was more than a conventional mistake about enterprise moats. The thesis depended on two mutually reinforcing claims: 1. AI would rapidly replace application software and labor. 2. Achieving that outcome required an extraordinary, sustained expansion in compute. When immediate capability gains fell short of the rhetoric, brute-force inference and “token maxing” could still sustain compute demand temporarily-but not necessarily the economics or timelines implied by the narrative. So your account explains why financial gravity eventually won. The complementary question is how the reservoir filled: This wasn’t only a bad fundamental trade. It was a leveraged position inside a capital cycle whose financing, spending, reported growth, and public narrative reinforced one another-until they didn’t.