Nvidia's FY2028 guide reprices analysts' revenue models by about 18%

Twenty-six points separate Nvidia's own forecast from the sell side's ^[1], which is another way of saying the company guided to roughly 1.6 times the growth rate analysts had in their fiscal 2028 models ^[2]. Hold the prior-year base unchanged, and moving a model from 1.44 to 1.70 lifts the FY2028 revenue line by about 18% ^[4], against a stock that added a bit more than 6% before the bell ^[2], or roughly a third of the revision ^[5]. A market that pays a third of a revision is either discounting the guide or had already paid for part of it in advance, and I lean toward the second reading.

The quarter that produced the guide more than doubled year over year ^[1], so 70% ^[3] embeds a deceleration of at least thirty points ^[3] before anyone argues about whether it is achievable, and Kress attached the supply-constraint caveat herself ^[3]. A growth number bounded by what a company can build describes the packaging line first and the buyer second, which is why the AWS disclosure carries more information than the percentage does: two million GPUs ^[4] is a unit count, and the report carries neither a price nor a delivery window ^[7], so nobody outside the room can convert it into a revenue figure for any particular year.

Huang's answer to the criticism of Nvidia financing its own customers was that the critics are "missing a very big point," and that startups' heavy funding needs are "really the nature of AI" ^[5]. That is a claim about how capital-hungry this industry is at birth, or rather about who is expected to carry that cost during the hungry part, and it leaves the accounting question untouched. What I would want is the split between orders backed by a buyer's own cash generation and orders backed by capital that Nvidia or its peers supplied, and the material offers no such split ^[5].

This is probably wrong, but the runway reads as real to me and the composition reads as the risk. The corroborating evidence sits a tier down from the chips: Okta jumped nearly 20% with demand for its identity security products spiking on agentic AI ^[8], and Salesforce rose 11% on its own beat ^[9], both of them lines that get paid by end customers rather than by investors. The thesis fails if the 70% is later trimmed on supply rather than on orders ^[3], or if the incremental beat starts arriving from accounts Nvidia also finances.

One allocation note, because cash of this size has alternatives: the reported $12.9 billion for Hugging Face ^[6], which neither party confirmed, is money pointed at open-source software distribution rather than at capacity or at more customer paper.