Leopold Aschenbrenner wrote an essay two years ago called “Situational Awareness,” arguing AI capability was compounding faster than almost anyone in finance had priced in. He was persuasive enough that he raised a fund and named it after the essay. This week Reuters reports JPMorgan curbed its lending to that fund after AI-related losses, and CNBC has it running options positions tied to AMD, Bloom Energy, and CoreWeave, the exact infrastructure names that were supposed to prove the thesis right.

Here’s the thing about being right on the direction: markets don’t pay out on direction, they pay out on timing and leverage, and those are the two variables the essay never had to survive contact with. Being confident AI changes everything by 2030 is a very different position from being solvent on the way there. I keep thinking about Thiel’s fund going from zero equity positions to 72% power and infrastructure back in August, essentially the same trade Aschenbrenner is reportedly in now, just without a margin call attached yet.

And then there’s Amodei, the same week, telling Bloomberg it might be time to slow AI model advances, after two years of an industry racing on exactly the premise Aschenbrenner monetized. Nobody in this story is wrong about where the plane is going. The disagreement, expensive for at least one of them, is about how fast you’re allowed to be right.

Sources read for this entry