Nvidia's market capitalization has come down from its peak as investors reassess the timeline and magnitude of AI infrastructure spending — a repricing that tells you less about the company's quarter than about the market's patience for the story it attached to those quarters.

Start with what has not changed. Nvidia remains the indispensable supplier of AI compute, with demand for its accelerators still exceeding supply and its free cash flow margins — the paper counted 42% in the most recent quarter (per the chipmaker analysis) — the envy of an industry financing its own customers. Nothing in the valuation adjustment reflects lost orders, failed products, or competitive displacement. The chips are selling. The question that moved is the one underneath the sales: how long the buyers can keep buying at this intensity, given that the buyers' own combined cash burn just turned negative for the first time in two decades (see that report).

That dependency chain is the actual story, and the market finally priced it. Nvidia's revenue is hyperscaler capex; hyperscaler capex is now partially financed by balance-sheet tolerance rather than operating surplus. A supplier whose customers fund purchases from future expectations carries future expectations inside its own multiple, whatever its current earnings say. The correction is the market discovering that transmission mechanism and discounting it — not doubting the demand, but doubting the funding.

X framed the move as vindication, with the usual camps splitting between bubble-confirmed and dip-buying. Both miss the healthier reading: this is what pricing for perfection looks like when perfection arrives on schedule but profitability arrives later than assumed. Nvidia was never valued on current earnings; it was valued on the assumption that today's spending rate compounds indefinitely. Any assumption carrying "indefinitely" will meet a quarter where it gets tested, and the test costs twenty percent of a multiple regardless of fundamentals. Companies priced for inevitability are repriced by mere probability.

The scrutiny now attaching to AI spending generally lands on three questions, none answerable from Nvidia's income statement alone. Whether enterprise AI revenue grows into the infrastructure before depreciation catches the hardware; whether model efficiency gains reduce compute demand per task faster than usage expands total tasks (the Jevons paradox, running live); and whether the concentration risk — a handful of customers accounting for the majority of accelerator purchases — converts from theoretical to contractual fragility if any single buyer blinks (see the rotation coverage). Nvidia cannot control any of these variables. Its multiple nonetheless encodes all of them.

The implications radiate beyond one ticker because Nvidia functions as the market's AI barometer. Index funds hold it as their largest single position; passive flows amplify both directions of its moves; and every AI-adjacent valuation — from neocloud lenders to power developers — borrows credibility from its chart. When the barometer adjusts, everything calibrated against it recalibrates too. That is why a fundamentally intact company shedding hundreds of billions in market value constitutes information rather than noise.

The honest close is that nothing here contradicts the bull case's substance. The chips remain scarce, the software moat widens with each cycle, and demand growth keeps outrunning every forecast the bears cite. What died this month was narrower and more important: the assumption that the trade had no timing risk. The market has re-admitted time to its models. Companies whose customers spend more than they earn should be valued knowing that.

-- THEO KAPLAN, San Francisco