NVIDIA: Q4 B revenue exceeded expectations

Performance: Q3 35B was still a bit above expectations; it was previously mentioned that the Q3 guidance was relatively conservative with some buffer left. However, the Q4 revenue guidance of 37.5Bn is slightly below the buy-side expectation of 38Bn, which is not a big difference. Since Q3 beat by 2.5, and raised by 2.5, if Q4 beats by 2-3, it will still be around 40Bn. However, it is true that NVIDIA has "barely" met expectations for two consecutive quarters, and the rising stock price has naturally raised expectations, but the "reality" of a host of supply chain issues has constrained supply. It should be noted that during this process, demand is not an issue.

The gross margin met expectations at 75%, and the CFO reiterated that the initial mass production of B will be in the low 70s, but will return to the mid 70s, around 75%, in the second half of 2025. When pressed by Bernstein analysts on the specific low 70s figure, the response was 71-72.5%.

However, Network saw a quarter-on-quarter decline, with the past three quarters being 31.7, 36.7, and 31.3B respectively. The CFO explained at the outset that Q4 will still see a quarter-on-quarter increase. (But it can be clearly stated that IB's growth has slowed down, and the future is uncertain.)

The key point is, the highlights from the conference call Q&A. Ranked by importance and impact on stock price, I was listening while watching the after-hours reaction. Personally, I felt that Jensen Huang's tone seemed more resolute this time compared to Q2 (sensing the feedback from the MS roadshow...)

1. B card's Q4 revenue is expected to be higher than previously anticipated, and the supply chain team is doing well, supply is on track, and visibility has improved. However, this huge integration task has many engineering issues, with several chips and many different custom solutions integrated into the IDC, it won't be a short-term miracle, and with so many components needing to ramp up, almost all the world's most important suppliers are on our list, TSMC, Amphenol, Vertiv, SK Hynix, Micron, Anker, KYEC, Foxconn, Quanta, Wistron, DELL, HP, SMCI (directly named, in the order mentioned by Huang, with Amphenol being second and no mention of Samsung).

This information from Huang is substantial, which will lead the market to believe that B card revenue is a bit more than "several billion", previously mentioned as 3-5Bn, now possibly 5-7? But the guidance remains at 37.5, so there will be concerns about whether Hopper will decline quarter-on-quarter.

It seems that the numerous "small issues" heard from the Taiwanese supply chain regarding rack system integration are indeed real, but note that they are all "small issues."

There are difficulties in ramping up component supply, especially since Amphenol's goods are lacking at the ODM... (everyone think about it).

2. H card revenue may still see quarter-on-quarter growth, depending on the situation.

After the CFO spoke, the stock price began to rebound, and it sounds like H card will at least not decline much quarter-on-quarter. If Hopper remains flat, and B card is "more than expected," then Q4 revenue is basically stable 3. The Scaling Law has not slowed down. Pre-training will continue intact; this is an empirical law, not a physical law. In addition, there are now Post-training scaling law and inference scaling law, three scaling laws are running simultaneously (still bold of Huang to say that...)

4. 2026 will not only grow but will grow to 2030. Two things: one is the "modernization" of the trillion-dollar traditional IDC, and the second is that new demands from GenAI will bring an additional trillion. This replacement will continue for 4-5 years, “we will grow to 2030!” (still bold of Huang to say that...)

1. When pressed by UBS's Tim, “Will Q1 B card revenue exceed H next year, can you confirm?” Huang did not answer directly and talked about a bunch of other things.

(Here it can only be understood for now that Q4 B revenue ramp-up exceeded expectations but is only about 5-10 billion. However, if Q1 B exceeds H, it needs 20 billion. It’s easy to sample a few billion in shipments, but 20 billion requires tens of thousands of cabinets, which is truly formal mass production. However, the supply chain shows that there are indeed many small issues with rack integration, so it is reasonable to assume that true mass production shipments should be between February and April, and Q1 will not be fully utilized. Therefore, Huang did not give a direct answer.)

1. Long-term outlook: Every company is running inference, 7 days a week, 24 hours a day. Inference demand is rising, with countless AI natives and enterprises, as well as Agent AI, robotics/Physical AI (Huang seems to have said this is underestimated). “Next year there will be many Agents,” “When one day there are tons of inference in the world, AI will have succeeded.” All companies are running inference, 7*24.

2. Continuing to maintain an annual iteration rhythm, everything is on track.

3. Sovereign revenue expectation maintained at 10-15 billion USD.

4. Software business AI enterprise annual revenue to reach 2 billion USD.

5. Demand and compliance issues in China: Compliance with relevant regulations is the top priority, and the rest is to “do our best.”

Source: Information Equality