Nvidia booked $40B of its $89B data center quarter outside the hyperscalers

Split the data center line the way Nvidia itself split it and the diversification argument reads as guidance rather than history: $89 billion less the $49 billion booked to hyperscalers leaves $40 billion, which is 44.9% of the quarter ^[7][11]. Kress's verb was "will represent" ^[13], and the two growth rates sitting on either side of that split are why the tense is the whole story.

Take the guidance midpoint, $107.95 billion, which is 12.2% above the $96.2 billion just reported ^[3][5][12]. Apply that rate to data center and you get about $99.9 billion; let hyperscalers repeat the 13% sequential gain they just posted and they reach $55.4 billion, leaving a residual of $44.5 billion, or 44.6% ^[15]. Half of $99.9 billion is $49.9 billion, so the word "roughly" is asking the non-hyperscaler book for something near 25% sequential growth while the hyperscaler book does 13% ^[16]. That, not the customer roster, is the claim a bear has to argue with.

The supporting number is the one worth deflating, or rather the one worth tracing. Kress put global AI venture funding above $400 billion in the first half of 2026 against $265 billion for all of 2025, with roughly 70% of it going to compute ^[14], which implies about $280 billion of compute spend in six months versus about $185 billion implied by the whole of last year ^[17][18]. Fortune's account of her point is that those dollars come back to Nvidia through GPU purchases or cloud rental ^[10], and the rental half is the leaky one, because a startup paying a NeoCloud for GPU hours is spending against silicon Nvidia already sold and already recognised. Set that against the revenue Nvidia can actually name: nearly 20 platform companies above $1 billion of annualised run rate, up from 13 in the fourth quarter ^[1], is at least $20 billion, or about 7% of the $280 billion ^[19].

This is probably wrong, but the diversification I would pay for is diversification of funding source rather than of customer name. A sovereign programme is a government budget line, a regional NeoCloud is a credit decision, and an air-gapped enterprise deployment is operating expense ^[13], and only the last of those is insulated from the yields-and-debt overhang that Dan Ives called out as not going away in the same breath as he called the quarter a masterpiece ^[8][9]. Notice also where Nvidia spent its disclosure: the mix got named, the sequential rate attached to it was the hyperscalers' 13% ^[7], and the trajectory of the segment carrying the argument is not quotable from outside. With fiscal 2028 sales guided up 70% from the prior year ^[6] and data center up 117% on the year already ^[4], the composition of that growth matters more than its size.

If the non-hyperscaler line prints 25% or better sequentially next quarter, I am wrong and Kress was reading an order book rather than a hope.