Nvidia split its customer base in two, and the halves are still moving together

The reporting change handed investors two growth rates measured two different ways, and the choice of basis decides which story gets told. On the May call Nvidia leaned on sequential change, where ACIE grew 31% against 12% for hyperscalers ^[7]. The same restated history measured year over year says the reverse: hyperscaler revenue up 115%, ACIE up 74% ^[13].

Put StreetAccount's estimates for the quarter being reported back on the sequential basis and the diversification argument thins. Hyperscalers at $43.6 billion against $37.9 billion is about 15%; ACIE at $43 billion against just under $37.5 billion is about 14.7% ^[5]. The gap that made the case in May returns as a rounding difference. The mix sits still with it: hyperscalers were 50.3% of the two segments combined last quarter, and consensus has them at 50.3% this one ^[2][3].

Worth knowing before the numbers land, the two segment estimates add to $86.6 billion while the data center estimate for the same period is $86.3 billion, a $300 million overshoot ^[4]. Whichever bucket comes out larger on Wednesday, the ordering is inside the noise of the forecasts themselves.

Here is the concentration figure the split does not print. Data center is expected to be 94% of a $92.2 billion quarter ^[16][15], and the nine restated quarters put hyperscalers at roughly 55% of data center revenue over the past year ^[12]. That puts about 52% of everything Nvidia sells inside the group Jensen Huang described as five or six companies, against a remaining industry he counted at 250,000 ^[1][5]. Two of the largest buyers, Meta and SpaceX, are not cloud vendors by trade and have begun renting out Nvidia capacity themselves ^[3], which blurs the line the disclosure is meant to draw.

The financing arithmetic points the same direction. The credit channel Nvidia assembled this month is about 26% larger than the revenue analysts expect the company to book for the full year ^[8]. That is the scale required because the tail cannot pay cash. And the tail has to carry real weight: 83% growth to $396 billion this year followed by 44% ^[17] implies roughly $570 billion next year ^[6], or about $174 billion of incremental demand ^[7]. If the hyperscalers are near the limit of what they can fund from operations, most of that increment arrives on borrowed money, which makes the loan book, not the order book, the thing to read.