Nvidia shares fall on a strong but not impenetrable quarter. There’s still so much to like
Nvidia delivered a better-than-expected quarterly results Wednesday night, but a few blemishes in the report — at least in the eyes of the market — has shares of the leading AI chipmaker falling in extended trading. That’s all some investors need to book profits in a stock that’s nearly doubled off its April lows. Revenue grew 56% year over year to $46.7 billion, outpacing the $46.06 billion the Street was looking for, according to estimates compiled by data provider LSEG. Adjusted earnings per share (EPS) in the three months ended July 27 increased to $1.05, exceeding the consensus estimate of $1.01, LSEG data showed. Nvidia also announced its board approved a $6 billion increase to its share repurchase authorization. NVDA YTD mountain Nvidia’s year-to-date stock performance. Bottom line Nvidia delivered a strong but not impenetrable quarter. And with the weight of the broader AI trade on its shoulders, it’s hardly surprising to see the roughly 3% after-hours drop in the stock Wednesday night. Sure, the magnitude of the quarterly beat and upside to consensus guidance expectations may not be quite what Wall Street was looking for. And the revenues in Nvidia’s crucial data center segment were a tad light. But it’s clear that demand for Nvidia’s chips was strong in the fiscal 2026 second quarter and only stands to strengthen in the coming months and production of its newest chip — dubbed the GB300, part of the Blackwell generation — accelerates ahead of an all-new Rubin family of chips set to debut next year. Nvidia’s adoption of an annual product launch cycle makes it hard to keep up, but that’s a high-quality problem for investors. The GB300, which was announced in March, is an upgraded version of the GB200, which began to roll out last year before ramping up in 2025. Remarkably, Rubin is already in the fabrication phase, Nvidia executives said on the call. Why we own it Nvidia’s high-performance graphic processing units (GPUs) are the key driver behind the AI revolution, powering the accelerated data centers being rapidly built around the world. But Nvidia is more than just a hardware story. Through its Nvidia AI Enterprise service, Nvidia is building out its software business. Competitors : Advanced Micro Devices , Broadcom and Intel Most recent buy : Aug 31, 2022 Initiation : March 2019 “This year is obviously a record breaking year. I expect next year to be a record breaking year,” Nvidia CEO Jensen Huang said. Supporting this view, Huang called out that that capital expenditures from the top four U.S. cloud service providers alone is tracking to be about $600 billion this year. Huang doesn’t think it’s unreasonable that Nvidia will grow to be the beneficiary of a “significant part of that.” Again, that’s only on the top four CSPs, not to mention the enterprise companies building on premises data centers. Longer-term, Nvidia thinks AI infrastructure spend will likely to hit $3 to $4 trillion by the end of the decade. Huang believes that Blackwell and Rubin will provide a strong opportunity to scale into that massive spend. CFO Colette Kress called out a few main drivers of the growth in annual AI spending, including the evolution in the very nature of AI computing less than three years since the launch of ChatGPT. The advent of reasoning agentic AI — in simple terms, these are AI systems that take time to chew on a task before taking an action — requires “orders of magnitude more training and inference compute,” Kress said. Additionally, Kress cited the global buildout of sovereign AI infrastructure outside the U.S., enterprise AI adoption and “the arrival of physical AI and robotics.” The profit-taking Wednesday night does not change the fact Nvidia still has a massive opportunity to keep growing as it supports the worldwide effort to build AI data centers, which Nvidia likes to call AI factories because they manufacture intelligence. We’re also fully aware that with Nvidia’s guidance, better than expected has simply become the default expectation. So while we once again saw Nvidia’s revenue forecast for the current period come in north of the consensus estimate, some investors were likely looking for an even rosier guide — the so-called whisper number on Wall Street. Plus, Nvidia’s guidance does not include any contribution from sales of its H20 chip to Chinese customers given the geopolitical uncertainty that still remains despite the Trump administration’s decision to walk back its April export ban. Management said on the call that there’s customer interest in the chips and the supply is there. If the China situation gets more favorable for Nvidia, its guidance could prove conservative. Given the results and what we heard on the call, we continue to view Nvidia as an “own it, don’t trade it” name. For the moment, we are putting our price target under review and maintaining our hold-equivalent rating, however. Commentary Coming into earnings, we wanted to get Huang and Co.’s thoughts on four key topics : Blackwell demand; the profit-margin impact of the transition from the GB200 to the GB300; total cost of ownership and lastly, the outlook on China. Here’s a closer look at what we ended up hearing. 1. Blackwell demand : This is clearly showing no sings of letting up, with finance chief Kress calling out that cloud customers are expected to invest roughly $600 billion in data center architecture alone, almost double the spend that we saw just two years ago. On the call, Kress noted that a $3 million investment in the GB200 can lead to $30 million in token revenue. “The more you buy, the more you grow,” as Huang likes to say. He repeated that line on Wednesday night’s call. Clearly these chips pay for themselves and it’s only getting better. In fact, thanks to the strength of Nvidia’s developer ecosystem, Kress noted that Blackwell performance has more than doubled since launch as the software that runs on the hardware takes advantage of its capabilities. 2. GB300 transition : On the call, Kress called out that the GB300 has begun production shipments, adding that growth is aided by sales to so-called neoclouds (think companies like CoreWeave ), enterprises and sovereigns. “We are at the beginning of an industrial revolution that will transform every industry,” she said, adding that she sees “significant long-term growth opportunities for Nvidia.” While we didn’t get any quantitative thoughts the GB300 margin impact, Nvidia’s adjusted gross margin of 72.7% in the quarter was above Street expectations and Kress guided for a sequential improvement in the current period and reiterated its forecast of exiting the fiscal year in the mid-70s. No major curveballs there. Perhaps a key reason why: Kress described the manufacturing transition from the GB200 to the GB300 as “seamless” thanks to the shared rack architecture between the two products. Those comments were encouraging given some of the temporary hiccups Nvidia experienced when shifting from the Hopper architecture to the Blackwell generation. Production of the GB300 is now up to about 1,000 racks per week, and Kress said output is expected to accelerate throughout the third quarter as more capacity comes online. As a result, the team expects widespread availability in the back half of the fiscal year. Compared with Nvidia’s older generation H100, the GB300 is delivering a 10x increase in inference performance with reasoning models (the latest version of large language models), according to management. 3. Total cost of ownership: One factor driving demand for Blackwell is energy efficiency gains being realized with new chips versus Hopper — and this helps advance our understanding of the total cost of ownership of Nvidia’s technology. This was an especially big theme of Jim’s heading into the print. The crux of it all is whether Nvidia’s customers are actually seeing a return by spending billions on its chips and the server systems and networking technology paired with them. Looking to the GB300 versus Hopper, in particular, Kress noted that a GB300 NVL72 AI factory can deliver a 10x increase in tokens per watt — a massively important factor as energy is quickly proving to be a bottle neck in compute. Later in the call, Kress went on to say that Blackwell’s efficiency gains are enabling companies to monetize their compute at unprecedented scale.” That’s a huge win for Nvidia and its customers as efficiency gains reduce the total cost of ownership as a data center can pump out significantly more compute without increasing power consumption. Huang elaborated on this dynamic in response to a question about why some of its customers continue to shell out money for Nvidia chips at the same time they’re developing in-house chips. “Not only are we the most energy efficient, our [performance] per watt is the best of any computing platform. And in a world of power-limited data centers, [performance] per watt drives directly to revenues. And because the performance per dollar is so incredible, you also have extremely great margins.” 4. H20 and China : Kress did confirm that some China customers have already received licenses from the U.S. government to resume purchases of the H20, a throttled-back version of Hopper chips designed to comply with Washington’s previous export restrictions. However, despite the interest, nothing has yet been shipped based on those licenses as management is looking for more clarity from the U.S. and Chinese governments. As a result, Nvidia saw a $4 billion sequential decline in H20 sales — though $650 million worth of H20 chips were sold to customers outside of China — and opted to not include any H20 sales to China in its outlook. That’s hardly surprising to us. Nevertheless, the opportunity remains, with Kress saying: “If geopolitical issues reside, we should ship $2 billion to $5 billion in H20 revenue in Q3. And if we had more orders, we can bill more.” Nvidia continues to work toward the approval of a Blackwell-generation chip into China, which Huang reiterated has immense long-term potential given the technology ecosystem in the country. He called it the second largest computing market in the world behind the U.S. He said by his estimation, China would be a $50 billion opportunity this year alone for Nvidia were the company able to address it with competitive products, growing at a rate of about 50% a year. Equally important, Huang called out all the innovation taking place in China, noting that “the vast majority of the leading open source models are created in China,” and that “about 50% of the world’s AI researchers are in China.” As a result, he believes it important that American companies are allowed to address the Chinese market. Aside from the revenue opportunity, there is also the argument that it’s to the U.S.’s advantage to standardize the world’s AI research on American-developed hardware. Not only does it make others more reliant on the U.S., it also means that advances coming out of China — think open source models like Deep Seek — can be taken advantage of by U.S. developers also working within the Nvidia ecosystem. Time will tell if Blackwell receives approval from Washington to sell into the Chinese market, but Huang said he sees it as “a real possibility,” adding “we just have to keep advocating, the sensibility of, and the importance of, American tech companies to be able to lead and win the AI race and help make the American tech stack the global standard.” Guidance Looking to the third quarter, management’s outlook was better than expected, though some may have been looking for the guidance to be further ahead of consensus estimates than it ultimately was. As we noted in our preview note, a beat with strong guidance was the minimum we needed to see given the recent strength in the stock. Revenue of $54 billion, plus or minus 2%, ahead of the $53.14 billion LSEG consensus estimate. Importantly, the company is not assuming any H20 shipments to China in this outlook. Adjusted gross margins are expected to be 73.5%, plus or minus 50 basis points, better than the 72.8% estimate compiled by FactSet. Expectations for adjusted operating expenses in the fiscal third quarter of $4.2 billion are about in line with expectations. (Jim Cramer’s Charitable Trust is long NVDA. See here for a full list of the stocks.) As a subscriber to the CNBC Investing Club with Jim Cramer, you will receive a trade alert before Jim makes a trade. Jim waits 45 minutes after sending a trade alert before buying or selling a stock in his charitable trust’s portfolio. If Jim has talked about a stock on CNBC TV, he waits 72 hours after issuing the trade alert before executing the trade. THE ABOVE INVESTING CLUB INFORMATION IS SUBJECT TO OUR TERMS AND CONDITIONS AND PRIVACY POLICY , TOGETHER WITH OUR DISCLAIMER . NO FIDUCIARY OBLIGATION OR DUTY EXISTS, OR IS CREATED, BY VIRTUE OF YOUR RECEIPT OF ANY INFORMATION PROVIDED IN CONNECTION WITH THE INVESTING CLUB. NO SPECIFIC OUTCOME OR PROFIT IS GUARANTEED.
