Nvidia's report was published on 26 August at 16:20 New York time. The investor call was held about an hour and forty minutes later.
And five things said on it did not appear in the press release - one of which changes what we knew about the margin.
Errors or inaccuracies are possible. Spotted something that looks wrong? Write to me and I will correct it.
Why a Call Is Worth Reading Separately
A press release is a legal document. It is reviewed by lawyers, worded carefully, and delivers what must be delivered - no more.
A call is a different place. There, executives answer questions they did not write, and often the most interesting number is said there rather than in the release.
And that is exactly what happened here.
One: Guidance for a Full Year Beyond the Guided Quarter
Colette Kress, the chief financial officer, gave guidance for revenue growth of about 70% in fiscal 2028.
Analysts expected 44%.
And the press release contained no reference at all to fiscal 2028. It stopped at the third quarter.
And on top of that she added a qualifier that enlarges the significance rather than reducing it: this is supply-constrained guidance. That is, the number does not describe demand - it describes how much Nvidia can supply. Unconstrained demand, in her words, is significantly higher.
That is a very strong claim, and it is also unverifiable from outside. There is no way to measure demand that was not supplied.
Two: The Margin - and Here the Picture Differs From the Release
The press release stated that the third-quarter gross margin is expected at 74.0%, plus or minus 50 basis points. No reason was given there, and nothing was said about what comes after.
On the call the full picture was given:
| The quarter reported | 75.0% |
| Third quarter | 74.0% |
| Fourth quarter - the trough | 71%-72% |
| Fiscal 2028 | Recovery to 72%-73% |
That is, not one percentage point - three to four.
And the cause was stated explicitly, in Kress's words: "extreme pricing conditions in memory" - at increases exceeding the company's prior expectations.
And that finally settles a question left open in the piece on the report. There I noted that the company had not given a reason for the margin decline, and raised Vera Rubin as my own hypothesis - marking it explicitly as an inference and not a fact. Now there is an answer, and it is memory.
And why this matters beyond Nvidia: memory is not a component unique to AI accelerators. It goes into every computer, every server and every phone. And on that very same evening, HP reported PC sales in which revenue rose 18% and units fell 16% - and it too mentioned memory supply.
Two companies that do not compete, on the same evening, pointing at the same bottleneck.
Three: The Number That Quantifies the Demand Thesis
Jensen Huang was asked about the shift to AI agents, and answered with a figure:
"The amount of compute necessary for an agent versus a human using it is probably 15 to 100 times" - depending on task complexity.
The logic behind it is simple, and worth understanding because it underpins the whole demand thesis.
A human who asks a question gets one answer. They read it, think, and ask another. The pace is limited by the human's reading speed.
An agent given a task is not limited that way. It decomposes the task, runs several paths in parallel, checks itself, backtracks when it errs, and calls external tools. Each of those steps is another call to the model.
So the shift from human use to agent use is not an incremental improvement in demand - it is a multiplication by an order of magnitude.
And that is a claim testable indirectly: if it is right, it will show up in inference growth at cloud providers, not only in model training.
Four: One Line That Bears Directly on the Circular Financing Question
Kress noted that days sales outstanding rose to 60 days - "reflecting extended payment terms for investment-grade customers".
That sounds like a technical working-capital detail, and it is not.
Days sales outstanding is the time between delivering the product and receiving the money. When it rises, the company is in effect financing the customer - it has handed over goods and is waiting for payment.
So this is a third form of customer financing at Nvidia, alongside two we already knew about:
| Equity investment | nearly $50 billion in AI labs, per the company |
| Arranged external financing | over $500 billion in partnership with six funds and banks |
| Working capital | extended payment terms - 60 days sales outstanding |
And I want to be precise here, because it is easy to overstate: extended payment terms for investment-grade customers are an entirely normal commercial practice, and the risk in them is low. This is not a sign of distress.
But it does add a layer to a picture in which part of the company's revenue is supported by money it itself provided, in one form or another. And analysts on the call asked about this explicitly.
Five: Vera Rubin, and the Inventory Built Ahead of It
Kress reported that inventory grew to $32 billion, and explained that it is being built ahead of the Vera Rubin launch.
And Huang added that it is expected to be "the fastest product ramp in Nvidia's history", and that purchase orders have already been received from hyperscalers and major cloud providers.
And that closes a loop I left open in the piece on the report. There I wrote that inventory growing 47% is a double-edged thing - a supply build ahead of shipments, or a first sign of a slowdown. The answer from the call: the first.
And What Analysts Pressed On
The list of questions on the call says a great deal about what the market is actually worried about:
| Sustainability of the 70% | If unconstrained demand is higher, what is the constraint |
| Bottlenecks | Memory, power, data-centre capacity and wafer availability |
| Custom chips | Frontier labs developing their own silicon |
| Open models | The effect on the value of closed-model companies |
| Revenue per gigawatt | The pricing trajectory beyond Vera Rubin |
| Circular financing | Nvidia's investments in its own ecosystem |
And Huang's answer on custom chips was his key claim: "We are the only company in the world that creates and builds, offers an entire AI factory platform, a full stack system" - as against whoever offers a purpose-built chip for one infrastructure.
הזווית שלי
דעה אישית של אילן אברמוב - לא ייעוץ ולא המלצה
What I take from this call is not a single number, but how far apart the release and the call were.
The release said 74% and gave no reason. The call said 71%-72% and named memory. Both are correct, and both were published the same evening - and the difference between them is the difference between one quarter and a trajectory.
And that is the practical lesson: whoever reads only the release gets what must be disclosed. The number that decides is often said on the call, in answer to a question.
And what drew me most was actually a small figure - 60 days sales outstanding. Not because it is worrying in itself; it is not. But because it is the third way in which Nvidia finances the other side of its own transaction - after nearly $50 billion of investment and after $500 billion of arranged financing.
And when a company finances its customers in three different ways, "revenue growth" stops being an independent gauge of external demand. That does not make the demand fake - it is real and measured. It does mean it has to be measured somewhere else: in the customers' cash flow, not in the supplier's revenue.
And the 70% fiscal 2028 guidance is, to my mind, the boldest claim made there. Not because of its size, but because of the qualifier: "supply-constrained". A company saying it is selling less than it could sell, in a market where everyone is hunting for signs of a slowdown, is choosing to price itself by its production limit rather than by demand.
If it is right, that number rises. If it is wrong, it was the peak of the cycle. And we will not learn the difference from the call - we will learn it from customers' purchase orders, and from the gross margin that appears in the fourth quarter.
(It is important to stress: this is my personal opinion only, and nothing herein constitutes a recommendation to take any action.)
