Here’s my take on NVDA after the Q2 FY25 earnings results; I’ll include this in my next portfolio update which I plan to publish this weekend.
NVDA reported Q2 FY25 results on 28Aug. NVDA continues to put up enormous numbers with revenue topping $30B for Q2, up 122% Y/Y. Data center revenue continues to explode: data center revenue was $26.3B in Q2, and, before AI-induced breakout occurred in 2023, data center revenue was only $4.3B in Q1 FY24. Data center revenue growth for Q2 was still 154% Y/Y in spite of lapping the first breakout quarter (Q2 FY24). This surge in growth has been accompanied by very high margins: 75.7% gross margin, 56.4% net margin, and 44.9% FCF margin! The fact that NVDA is able to keep delivering such huge margins is a symptom of a business that is not subject to price cutting pressures from competition. NVDA’s total solution for advanced computing is continually several steps ahead of the competition. Thus, looking backwards NVDA’s performance has been nothing short of amazing. The question for investors is how quickly will growth decelerate (it’s obviously decelerating and it will continue to do so). Investors who believe 1) that deceleration will come more quickly, and 2) that this boom phase of the typical semiconductor boom-bust cycle will soon enter a bust phase will not continue to hold onto their shares. On the other hand, investors who believe that AI is causing a multiyear boom cycle for computing and that this particular boom cycle will last much longer than typical in past cycles will continue to hold their shares. I fall into the latter camp. Let’s dive deeper into the reasons why I think this AI boom will last longer than other investors expect.
Several days ago, there was an excellent podcast, in which Patrick O’Shaughnessy interviewed another investor, Gavin Baker. Gavin followed up that podcast with a post on X. The podcast and follow-up post are excellent, and they’re aligned with my thinking about future demand. A side note: I find it important to pause the podcast or stop reading when the speaker or author uses a term that I do not understand or don’t know. Continuing to listen or read without fully understanding all the terms, concepts, or acronyms will lead to an incomplete understanding on my part. Furthermore, an incomplete understanding of a premise makes it impossible for me to judge whether or not I agree with the speakers’/authors’ conclusions. So, if you come across something that’s not clear to you, stop and spend a little time looking up a term or an acronym becfore continuing on. A key conclusion in Gavin’s podcast and post is that spending on CapEx AI hardware will continue until the scaling law breaks. The scaling law states that large language models (LLMs) will continue to show significant improvement with each 10x addition of parameters, and, to add more parameters, requires a big increase of compute, compute for which NVDA is the sole arms dealer. There are five contestants (MSFT, AMZN, x.ai, META, and GOOG with some investing via OpenAI and Anthropic) participating in the AI arms race, and they all believe that the first to develop a super-intelligent AI (i.e. like a digital god) will unlock a potentially $100T opportunity; furthermore, the losers of the arms race could see their existing businesses completely disrupted. All five contestants have access to hundreds of billions of dollars to spend on CapEx to make it all happen. The AI game that’s being played by the big technology companies puts NVDA in an enviable position.
For NVDA to continue to be a good investment, growth needs to stay higher for longer than the market expects. Revenue growth is the single most important metric as all other growth metrics are derived from or a symptom of revenue growth. In fiscal 2024, revenue grew 126% over FY23. The consensus of analysts’ forecasts predict revenue growth in FY25 and FY26 to come in at 100% and 39%, respectively. Beyond FY26, most analysts aren’t yet making predictions. For FY25, there are only two more quarters left, and the analysts’ collective prediction will likely be close to the actual number. However, for FY26, I think that the revenue estimates ($168.6B) will turn out to be too low. During the Q2 earnings call management, for the first time since the rush to build out AI hardware infrastructure began, did not remark on how demand far exceeds supply; supply of GPUs is now more balanced with demand. Part of the reason, I think, has to do with NVDA’s and its suppliers’ efforts to increase manufacturing capacities over the past year. Another reason is likely the transition from Hopper to the Blackwell platform, but the fact that Hopper sales are still increasing demonstrates not only a lack of an Osbourne effect due to the imminent release of Blackwell but also the strength of NVDA’s GPU franchise. I think Blackwell will usher in the next wave of growth for NVDA; demand for Blackwell should be massively constrained because 1) it will take time for the supply chain to ramp production, and 2) competition among the big tech companies will have them all scrambling to be the first to get access to Blackwell, which is the next big step up in compute with lower compute/watt, in their quest not to lose the race to develop the world’s first super-intelligent AI.
Regarding NVDA in the portfolio, it remains one of the largest allocation positions. When the shares dropped from the all-time high of around $140 to $106-ish and below on rumors that Blackwell was being delayed for three months, I sold a portfolio of my shares and bought Dec 2025 call options with strike prices of $105 and $100. I was unsure whether the Blackwell delay rumors were true, but I reasoned that even with a delay, NVDA’s value (given that the Hopper chips could pick up the slack until Blackwell was ready) would not be materially impacted. Thus, I leveraged my position to benefit when the shares rebounded. After the Q2 result, I reversed the shares to LEAPS trade back into shares because 1) NVDA shares are now less of a fat pitch than they were around $105 or below, and 2) I was expecting more comments from management around how NVDA GPUs are still supply constrained (now supply and demand are more in balance as was revealed on the earnings call), and 3) given number #2, I think that analyst revenue estimates of $121.6B in FY25 revenue are pretty close to what we can expect; therefore, I’d rather wait for another fat pitch (if we get one) rather than continue to hold LEAPS and a leveraged position.