Annualise the print and the ratio does the work: $96 billion times four is $384 billion of Nvidia revenue against the roughly $725 billion of combined capital spending Amazon, Microsoft, Alphabet and Meta have disclosed for this year [1][3], which puts Nvidia's run rate at about 53% of what its four largest customers are spending on land, buildings, power, networking and everything else that is not a GPU [10]. The number is imprecise in both directions (not all of that $725 billion buys Nvidia silicon, and not all of Nvidia's revenue comes from those four), but it is the right order of magnitude for the point Fool analyst Emily Flippen makes about the company being "a lagging indicator of the AI buildout, not a leading one" [4].
Thirteen consecutive quarters of clearing its own guidance is about three and a quarter years of the same ritual [5][15], and the next instalment is already set at roughly $108 billion, some 12.5% above this one [2][14]. That is a well-run forecasting process, not new information about demand.
The financing is new information. Nvidia lined up $500 billion of outside capital with Apollo, BlackRock and Goldman Sachs this month to fund customer data centres, including a $100 billion commitment to OpenAI [8], which is 20% of the facility [12] and, at the current run rate, about 1.3 times a full year of Nvidia's own revenue [11]. The Motley Fool calls it circular financing that can flatter demand more than economics justify [9]. Broadcom, meanwhile, booked over $30 billion of AI orders last quarter for custom accelerators built for Nvidia's own hyperscale customers [7], a figure equal to roughly 31% of Nvidia's quarterly revenue [13] though measured on a different clock, since orders are a promise and revenue is a receipt.
Three ways this resolves. If the $500 billion turns out to be senior paper secured against contracted offtake at spreads lenders are happy to hold, this is project finance and Nvidia has simply widened its addressable market. If it is equity risk wearing a credit coupon, the first utilisation disappointment reprices the marginal buyer out of existence. If the hyperscalers lift capex above $725 billion next year straight out of operating cash flow, the vendor paper was optionality and wafer supply is still the binding constraint.
This is probably wrong, but the more interesting version of the quarter is that Nvidia has concluded its customers' cost of capital, not its own supply, is what caps orders, and has decided not to let customer balance sheets set that ceiling. What would falsify it: an undrawn facility twelve months from now, alongside a capex guide that goes up anyway.
