The tape and the filings are diverging
Two signals we track moved together for a year. This quarter they split — the widest gap in the series.
For a year, two signals we track moved together. One reads the tape — semiconductor momentum, how far price has stretched from its trend, how unstable the move is. The other reads the ground truth — the share of layoffs employers attribute to AI, insider selling at the companies doing the spending, and the gap between compute committed and revenue earned.
When both rise together, a boom is at least paying for itself. When they split — price climbing while the fundamentals weaken — that gap is the signal.
This quarter they split. Our divergence measure, D(t) = M(t) − G(t), swung from −1.80 to +4.06: the market signal pushing toward its highs while the ground-truth signal rolled over. It is the strongest single-quarter divergence in the series. See it on Capex Watch →
A second number says why it matters. Across the core AI labs, the same disclosed dollar of equity is supporting roughly 26× its value in committed compute on a funded-cash basis — and even present-valued, about 21×. Admit every reported secondary and the multiple compresses toward 5×. That range itself is the point: there is no arm's-length benchmark where these numbers sit comfortably.
None of this is a forecast. It is a measurement, and a falsifiable one — the inputs are public, the method is in the brief, and we will publish the number every quarter whether it confirms us or not. If G(t) turns back up, the divergence closes and the boom earns its price. If it does not, the gap is telling you something the tape is not.
That distance — between what the market is pricing and what the math allows — is the whole reason this desk exists. We will track it here, in the open.
Read the full fragility brief (PDF) ↓ · Browse the 68-company board →