My new article explores the economics of neoclouds through a microeconomics lens rather than the usual equity-research framing. I thoroughly enjoyed writing this one (possibly my favorite?), it really got my creative juices flowing.

The writeup breaks the business into two stacked markets — token sales upstream and GPU-hours downstream — and explains why the difference in market structure matters for who captures value in AI. It then walks through three shocks that shape neocloud outcomes, the specific ways a neocloud can earn economic profits despite heavy capex and depreciation risk, and what that means for names like CoreWeave and Nebius. There’s also a concrete trade angle at the end.

If that kind of semiconductor and AI market research interests you, please subscribe!