Nvidia's customer changed

Colette Kress, Nvidia’s finance chief, has a gift for saying the uncomfortable thing in the voice of a woman reading a grocery list. On August 26th, forty minutes into a call in which she reported $96 billion of quarterly revenue, she said this about the frontier labs that buy her company’s most expensive products: they “are growing faster than what their balance sheets and credit profiles can support.” She was not warning anyone. She was explaining why Nvidia has agreed to backstop them.

The quarter itself was the one everybody expected and then some. Revenue doubled on the year, data-center revenue reached $89 billion, and the company guided to $108 billion next quarter and roughly 70% growth for the following fiscal year, a number it described as supply-constrained, with demand “much higher.” What the market priced was the growth. What the segment split shows is the mix, and the mix is where the credit story lives.

Nvidia now reports data-center sales in two halves. “Hyperscale,” the public clouds and consumer-internet giants, was $49 billion and grew 13% sequentially. “ACIE,” which the company defines as AI clouds, industrial and enterprise (neoclouds, corporate AI factories, sovereigns, and “hyperscalers purchasing capacity to supplement their own build-outs”), was $40 billion and grew 25% sequentially and 138% on the year. Sovereign alone, routed mostly through regional neoclouds, grew 35% on the quarter and more than tripled. Kress said the next quarter’s growth would come “primarily” from ACIE. On that trajectory the two lines cross within two quarters, and the customer who has historically written Nvidia’s checks out of operating cash flow stops being the majority of the book.