All news

Nvidia's forecast shows who earns first in generative AI

A graphite pickaxe and shovel leaning on a pale wall beside a single rough cobalt stone on the floor

Nvidia guided to $11 billion of quarterly revenue, up more than half in three months. The first profits of generative AI land with the supplier of compute.

Nvidia expects its revenue to jump by more than 50% in three months, to about $11.00 billion for the current quarter from $7.19 billion in the quarter to April. The guidance, given on 24 May alongside a record $4.28 billion from data centers, rests on demand for the chips that train and run generative AI models and comes with an expected gross margin of about 70% on an adjusted basis.

This is the first hard number on how much money generative AI is moving, and so far most of it lands with the supplier of compute. With margins near 70%, a large share of every dollar that labs, clouds and startups raise for training flows to one vendor. A good part of the capital now going into AI startups is, in effect, being passed on to the chip layer.

The market priced the shift within days. The shares rose by nearly a quarter the next day, and on 30 May Nvidia briefly passed $1 trillion in market value, the first chipmaker to do so. Its chief executive put the installed base of data center infrastructure at a trillion dollars and said it will move to accelerated computing. In this first phase value concentrates in hardware and the infrastructure around it, while every application built on top inherits a cost of compute set by a dominant supplier with long delivery times. The margins investors expect from AI software companies have to reflect that.

Compute is the largest line in a generative AI cost structure, and it will stay scarce for some time. Scarcity leaves room in three places: efficiency, through smaller and distilled models, better utilisation and faster inference; software that lets workloads run across different hardware and so reduces dependence on a single stack; and products where the value of each answer is high enough to absorb the cost of producing it. A founder raising money now should show a plan for compute and unit economics per task, not just growth in users.

A 70% margin is a scarcity rent, and rents invite competition. Nvidia's August report will show whether it delivers the guided quarter and what it says about the supply of its Hopper-generation H100 chips. The largest clouds and labs bringing their own chips or rival accelerators into volume use would erode the rent from the buyer's side, and the price of renting GPU capacity will show whether the shortage is easing.

The first profits of a technology cycle often go to the suppliers of its scarcest input and later move to those who own customers and workflows. Fanaura looks for AI-native companies positioned for that second phase, whose margins improve as compute gets cheaper, the bet at the centre of our thesis. The chip layer collects now; the application layer collects when the shortage ends.

For founders

Building AI-native products with unit economics that improve as compute gets cheaper? Send us the project.

Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.

Apply in five minutes
Subscribe

Get the news by email

One short note each weekday. No marketing, and you can unsubscribe at any time.