Nvidia stopped telling investors how big its networking business is, in the quarter it mattered most
The line that let anyone benchmark Broadcom and Arista against Nvidia has vanished from the 10-Q. The company has a defensible reason. It is also a convenient one.
If you want to know how much of Nvidia's data-centre business is networking, you now have to ask an analyst, because Nvidia has stopped saying.
A year ago the answer was in the filing. The quarterly report for July 2025 gave compute and networking as two lines, and networking was $7.3 billion, growing faster than compute. The report Nvidia filed on 26 August for the quarter just ended gives one line for the whole data centre, $89 billion, and a new split by customer instead: hyperscalers, then a bucket for AI clouds, industry and enterprise, then edge. A footnote says the presentation changed and the old periods were recast.
Recast, not restored. The product split did not move somewhere else in the document. It left.
The number everyone quotes is a guess
The figure now circulating, that networking is a fifth of Nvidia's data centre, comes from The Next Platform, which built it from something the chief financial officer said on a call: roughly $15 billion of networking against $60 billion of compute in the April quarter, grown forward a quarter. The author says plainly it is a model.
We ran the one comparison the filings still support. In the July 2025 quarter networking was 17.8 percent of the data centre. If the model is right, the mix has moved about two points in a year while the whole business more than doubled. Plausible. Unaudited.
Here is the paragraph that used to settle the question, from the July 2025 filing:
Data Center compute revenue was $33.8 billion, up 50% from a year ago. Sequentially, compute revenue declined 1%, driven by a $4.0 billion reduction in H20 sales. Networking revenue was $7.3 billion, up 98% from a year ago and up 46% sequentially, driven by the growth of NVLink compute fabric for GB200 and GB300 systems, the ramp of XDR InfiniBand products, and adoption of Ethernet for AI solutions at cloud service providers and consumer internet companies.
Who this hurts
Does anyone need this line? Not if you are buying Nvidia shares. The people who need it price Broadcom and Arista, the competitors who report their networking revenue every quarter and whose investors used to be able to hold it against Nvidia's. Networking is the one part of the data centre with named rivals, and it is now the part Nvidia describes least.
The timing sharpens the point. This is the quarter Nvidia's commitments to its suppliers rose from $119 billion to $279 billion. Companies rarely reorganise a disclosure in a direction that costs them.
The defence, taken seriously
The accounting rules say segment reporting should follow how the person running the company actually looks at it, and Nvidia can argue that is exactly what happened. The customer split is genuinely new information: the bucket of neoclouds, sovereigns and enterprises grew 138 percent and is now nearly as large as the hyperscalers. Nvidia's Compute & Networking segment still exists and still reports its revenue. It simply contains both things and separates neither.
So the change is defensible and convenient at the same time, and we do not think those cancel out. A company can follow the rules and still choose, among the presentations the rules allow, the one that tells its competitors' shareholders the least.
What would prove us wrong
We expect the compute-and-networking split to stay out of the next filing in late November, and to keep surfacing only as round numbers on earnings calls, where nobody signs anything. If Nvidia puts it back, we read the motive wrong and will say so.
One networking figure did survive the edit. The filing still notes that Nvidia's networking supply chain, including its Israel operations, employs about 6,200 people. Divide the modelled revenue by that and you get roughly $2.8 million a quarter per employee. Take the model away, and even that arithmetic stops.
