AI Crash Monitor

Ranking NVDA against Apple by market cap says nothing about the AI trade – the comparisons that do are inside the stack. Hyperscalers (MSFT, GOOGL, AMZN, META, ORCL) are NVDA’s customers: if the arms dealer persistently outruns the armies, the market is paying for GPU shipments it doubts will earn a return. Neoclouds (CRWV, NBIS, IREN) are the levered GPU landlords – the financing leg of the buildout, and historically the first link to crack. DRAM/HBM (MU) is the supply-chain confirmation: HBM demand is a direct derivative of GPU demand, so Micron diverging from NVDA is a tell in either direction. Semis & networking (AVGO, ANET, SMCI) are fellow suppliers on the same order books; power (VST, CEG) is the physical constraint. Two lenses on the same peers: size – live market capitalization, every name’s cap as a share of NVDA’s (is the arms dealer really worth 50 landlords and 5 Microns?) – and performance – returns from 6-month daily closes (Yahoo), groups equal-weight; “vs NVDA” is the 3-month spread in points; ρ is the 60-session correlation of daily returns with NVDA – how much of a name’s tape is just NVDA beta.

The market’s live estimate of how fast silicon decays – the counterpart to the depreciation panel above, which argues the same case from accounting alone. Each chip’s reference is the Ornn print that Kalshi’s weekly “price to beat” strike is set at; the month-end figure is the implied median of the Kalshi ladder, shown only where the book supports one. Last-gen silicon walking toward the bubble market’s $1.00 H100 trigger is the obsolescence thesis priced, not asserted.

Where the same theme trades on several venues, disagreement is a confidence signal.

Five falsifiable claims from the economic-singularity thesis, each read against live prediction markets or official data. The interesting motion is convergence: odds repricing up toward the narrative means euphoria is broadening beyond talk – and a claim failing in the labor data is the backlash channel arming.