We get Nvidia's Q3 earnings tomorrow. And we've been watching this deceleration in data center growth for the past five consecutive quarters.
As a refresher, this data center revenue consists of "compute" and "networking equipment."
And revisiting Q2, as you might recall, Nvidia's compute revenue actually contracted – lower revenues than the prior quarter. This is Nvidia's golden goose, the GPU, with negative growth.
It was networking that carried the growth.
Here's the CFO snapshot again for context …
Now, that brings us to tomorrow.
Remember, a couple of weeks ago Jensen Huang (Nvidia's Founder/CEO) stood on a stage in DC and said: "we've already shipped six million of the Blackwells (chips) in the first several quarters (of calendar year 25)." He went on to clarify, that's over the first "three and a half quarters" of this year.
That's of particular interest, because we already know (roughly) what they sold in the first two quarters of the year — and six million is a much bigger number.
Looking back at Q1 and Q2 compute revenue, based on a selling price of $30k per unit, we can estimate that Nvidia sold about 2.25 million units (Blackwell chips) over the first half.
If we conservatively assume the six million to be recognized as revenue over the full calendar year 2025, then there is a lot of catch up to do in the reporting over the next two quarters — one of which (Q3) comes tomorrow.
How much catch up?
The math would suggest at least 3.75 million chips to be reported between tomorrow and the Q4 report.
Clearly, there should be a huge acceleration in compute revenue coming, starting tomorrow.
