Nvidia FQ2 26 – Almost Spotless.

China takes off the shine.

• Nvidia reported another mighty set of results, but the ongoing uncertainty around selling chips in China took the gloss off what would otherwise have been a flawless performance.
• FQ2 26 revenues / EPS were $46.7bn / $1.08, slightly ahead of consensus of $46.1bn / $1.01, which is pretty much what I was expecting.
• This is because Nvidia’s revenues are a function of the capacity that it has reserved at TSMC, meaning that Nvidia has very good visibility on what it is going to sell at least 3 months in advance.
• This allows it to very carefully manage expectations such that the company’s financial performance can slightly beat expectations quarter after quarter.
• China is the one area where there is no visibility, as its revenues are subject to the whims of both the White House and the Chinese state, which is precisely what we have seen this quarter.
• Compute revenues declined 1% QoQ as a result of not being able to sell the H20 in China, and guidance was also reduced slightly.
• Here, FQ3 26 revenues / gross margins are expected to be $54bn (+/- 2%) / 73.3% (+/- 0.5%), which does not include any shipments of H20 into China as a result of the uncertainty.
• This is broadly in line with the consensus revenue estimate of $53.3bn, but not including H20 breaks the trend of slightly beating expectations, and so the shares fell 3% in after-hours trading.
• Most of the street had factored in about $2bn of revenues from the H20 for FQ3 26, which is why the now usual slight increase in guidance over expectations has not occurred.
• However, the underlying message is clear, which, to quote Jensen, is that “The AI race is now on” and that one can expect that spending on data centres will continue to grow.
• On the call, Jensen refused to be drawn on whether the implied growth of 50% YoY in 2026 in data centre expenditure was realistic, but he did reiterate that he thought that not selling in China was giving up on a $50bn opportunity.
• I calculate the China opportunity at $27bn this year, but given how keen China is to run in the AI race, $50bn in 2026 is a believable estimate.
• However, I continue to think that Nvidia is flogging a dead horse when it comes to the Chinese market, as Mr Xi has made it quite clear that he intends to rid the local market of foreign technology wherever he can.
• Therefore, making long-term investments in the Chinese market makes no sense, and to Nvidia’s credit, I have seen no signs of the company doing this other than Jensen spending time lobbying governments to allow him to continue making sales.
• However, it is not all bad news as when Nvidia loses China, it will be to inferior homegrown AI chips that are more expensive to manufacture and with which it is more expensive to train and inference AI models.
• This means when it comes to the rest of the world, which will be making choices between Western and Chinese technology, the Western version will be a better and cheaper option for the first time.
• Hence, non-affiliated countries will be more likely to pick Western technologies, which will allow Nvidia to take more market share than otherwise might have been expected.
• We have already seen some signs of this with Saudi Arabia, UAE and potentially, Qatar turning towards the West and away from China.
• For example, Humain will be launching a data centre in Saudi Arabia in 2026, which will be built using Nvidia chips, which could easily have been Chinese chips.
• The net result is that I think that whatever Nvidia loses in China, it will more than make up across the rest of the world, which is why when Jensen worries about China, I don’t lose very much sleep.
• Nvidia is now trading on Jan 2026 PER of 41.1x, which is starting to get somewhat stretched, especially when one can have Google at around 20x and Qualcomm at 13x.
• If I had to own a direct AI investment, I would much rather have Nvidia over OpenAI or Anthropic, but the gap to other options is now so large, it is impossible to ignore.