Nvidia: Blackwell shipments to start in the fourth quarter (FY25Q2 conference call)

NVIDIA (NVDA.O) announced its financial results for the second quarter of the 2025 fiscal year (ending in July 2024) after the U.S. stock market on the morning of August 29th Beijing time:

Below is the summary of the NVIDIA 2025 Q2 earnings conference call. For an interpretation of the financial report, please refer to " NVIDIA: AI Faith Crumbling, Sweet Poison Turning Bitter? 》

I. $NVIDIA(NVDA.US) Financial Highlights Review:

II. Detailed Content of NVIDIA Earnings Conference Call

2.1. Key Points from Executive Statements:

1) Business Progress

① Gaming and AI PC

a. Gaming and AI PC revenue reached $28.8 billion, up 9% quarter-on-quarter and 16% year-on-year.

b. Strong demand for gaming consoles, laptops, and desktops drove revenue growth.

c. Every PC equipped with RTX is an AI PC, with RTX PCs providing up to 1300 AI tops performance. Currently, there are over 200 RTX AI laptops on the market, with an installed base of 100 million devices.

d. NVIDIA ACE (Generative AI Technology Suite) is applicable to RTX AI PCs, supporting over 600 AI-driven applications and games.

e. NVIDIA's gaming ecosystem continues to expand, with the RTX and DLSS game libraries growing, and the GeForce NOW library expanding to over 2000 games.

② Data Center

a. Data center revenue reached $26.3 billion, up 16% quarter-on-quarter and 154% year-on-year, reaching a new high.

b. Key drivers of data center growth include NVIDIA Hopper, Blackwell GPU platform, generative AI model training and inference, AI-driven recommendation systems, and SQL and vector database processing.

c. Strong demand for Hopper, Blackwell platform will begin production in the fourth quarter and continue into the 2026 fiscal year. It is expected to bring in billions of dollars in revenue in the fourth quarter.

d. 45% of data center revenue comes from cloud service providers, with over 50% from consumer internet and enterprise companies e. The NVIDIA H200 platform began mass production in the second quarter, providing over 40% more memory bandwidth than the H100.

f. NVIDIA's networking business also performed well, with AI Ethernet revenue doubling compared to the previous period, a significant increase in customer numbers, and the widespread application of the Spectrum-X end-to-end Ethernet platform in AI processing acceleration.

③ Automotive Industry

a. The revenue from the automotive industry was $346 million, a 5% increase compared to the previous period and a 37% increase year-on-year.

b. Automotive manufacturers widely use NVIDIA's data center and computing capabilities to develop autonomous driving technologies, driving industry revenue growth.

c. The application of AI in the automotive industry is expected to continue growing, covering local and cloud-based consumption, and is expected to expand with the significant increase in computational demand from the next generation of autonomous driving models.

d. The gradual increase in AI applications in the healthcare field covers areas such as medical imaging, surgical robots, patient care, electronic health record processing, and drug discovery, with revenue in this field expected to reach billions of dollars.

④ Professional Visualization

a. The revenue from professional visualization was $454 million, a 6% increase compared to the previous period and a 20% increase year-on-year.

b. Demand is being driven by AI and graphics use cases, including model fine-tuning, Omniverse-related workloads, and virtual world construction.

c. Key growth industries include automotive and manufacturing, with the world's largest electronics manufacturer Foxconn and companies like Mercedes-Benz using NVIDIA Omniverse to create industrial digital twins for their factories.

d. The application scope of NVIDIA Omniverse is expanding, integrating generative AI assistants and agents into USD workloads through new USD NIM and connectors, driving enterprise digital transformation.

2) Financial Performance

① Revenue and Gross Margin

a. Revenue in the second quarter reached $30 billion, a 15% increase compared to the previous period and a 122% increase year-on-year.

b. GAAP gross margin was 75.1%, and non-GAAP gross margin was 75.7%.

② Expenses

a. Operating expenses increased by 12% compared to the previous period, mainly due to increased compensation-related expenses.

b. Operating expenses for the third quarter are estimated to be $4.3 billion (GAAP) and $3 billion (non-GAAP).

③ Cash Flow and Shareholder Returns

a. Operating activities generated $14.5 billion in cash flow.

b. Funds used for stock repurchases and cash dividends in the second quarter amounted to $7.4 billion, with an additional $50 billion stock repurchase authorization.

④ Future Outlook

a. Revenue for the third quarter is expected to be $32.5 billion, with a variance of 2%.

b. Revenue growth in the fourth quarter will be driven by Blackwell product sampling, with the full-year gross margin expected to be within the 70% range.

3) Key Points Summary:

① Accelerated Computing Reaches a Critical Point: Due to the slowdown in CPU expansion, developers must turn to accelerated computing to meet the growing demand for computation. NVIDIA has opened up new markets and driven the transformation of data centers from general computing to accelerated computing by releasing the new CUDA-X library.

② Blackwell's Major Leap: Blackwell is not just a GPU, but a complete AI infrastructure platform. It integrates a variety of innovative chips and systems, providing 3 to 5 times higher AI throughput than Hopper, making it the core of the next-generation AI factory.

③ The Importance of NVLink: NVLink's full GPU switch can connect 144 GPUs in one domain, achieving a bandwidth of 259 TB/s. This is crucial for the training and inference of large-scale language models, significantly enhancing AI performance.

④ Accelerated Development of Generative AI: Generative AI is rapidly expanding in multiple fields, from text and images to 3D physics and biology. Internet services and AI startups are deploying generative AI on a large scale, driving the demand for cloud services and sovereign AI infrastructure.

⑤ The Dawn of Enterprise AI Wave: The NVIDIA AI Enterprise platform helps enterprises customize AI models and applications, driving digital transformation. With the increasing installation base of CUDA-compatible GPUs, NVIDIA's software market has tremendous potential and is expected to grow significantly.

2.2 Analyst Q&A

Q: You mentioned that there have been changes in the mask of the Blackwell GPU. I am curious if there are any other incremental changes in backend packaging or other aspects? You still plan to ship billions of dollars worth of Blackwell in the fourth quarter. Does this mean all issues will be resolved? What impact do these changes have on the timeline, revenue composition, and customer response?

A: The mask change has been completed and does not involve any functional changes. Currently, we are conducting functional sample testing of Blackwell, Grace Blackwell, and various system configurations, and have showcased around 100 Blackwell-based systems at Computex, starting ecosystem testing. The functionality of Blackwell remains unchanged, and production is expected to begin in the fourth quarter.

Q: There is intense debate in the market about customers and their return on investment. What does this mean for the sustainability of future capital expenditures? How does NVIDIA internally assess customer returns and their impact on capital expenditures?

A: First, when I mentioned production starting in the fourth quarter, I meant shipping, not starting production. Regarding long-term issues, we are undergoing two important computing platform transformations.

The first transformation is the transition from traditional general computing to accelerated computing. This is because the speed of CPU expansion has significantly slowed down, while computational demand is growing exponentially. Without new methods, computing costs and data center energy consumption will increase significantly Accelerated computing not only significantly improves the running speed of applications, but also greatly reduces costs and energy consumption. In fact, accelerated computing can increase the running speed of applications by up to 50 times, making computing more economically efficient.

The second transformation is the rise of generative AI. With the development of accelerated computing, training large-scale language models has become more economical and feasible, driving the widespread application of generative AI. Large-scale generative AI models have shown tremendous potential in many areas, especially in generating content, automating tasks, and data analysis.

Q: Can you talk about the driving factors behind the annual growth of sovereign AI revenue, and how should we view the 2026 fiscal year?

A: Regarding the growth of sovereign AI revenue, we see this as a rapidly growing opportunity. The demand for generative AI is increasing globally, with countries looking to develop localized AI systems to adapt to their specific language, culture, and data. This demand brings us significant growth opportunities, and as these sovereign AI systems further develop, we expect this growth to continue in the coming years.

Q: The press release mentions high expectations for Blackwell, but there is also strong demand for Hopper. You anticipate a strong quarter in October without Blackwell. How do you view the coexistence demand between Hopper and Blackwell? Can you talk about the transition to Blackwell? Do you think they will be used together, or will Blackwell's activities mainly focus on new clusters?

A: The demand for Hopper is very strong, and there is also high demand for Blackwell. There are several reasons for this. First, if you look at global cloud service providers, the available GPU capacity is basically zero because these GPUs are either used to accelerate their own workloads, such as data processing, or for other applications.

Secondly, many companies now need accelerated computing and no longer want to build more general computing infrastructure. Hopper is the most advanced technology, and if you are now choosing infrastructure for your business, choosing Hopper is clearly the best choice. Therefore, people are eager to transform existing $1 trillion infrastructure into modern infrastructure, and Hopper is at the core of this transformation.

Q: Regarding the return on capital expenditure, some companies are driving the advancement of Artificial General Intelligence (AGI) and are willing to invest heavily to achieve new capabilities, while other customers are more focused on capital expenditure and return on investment. How do you prioritize these investments in new technologies and their investment time frame?

A: People investing in NVIDIA infrastructure can immediately see returns. It is currently the best investment return for computing infrastructure. For every $1 billion of general computing infrastructure built, you may only be able to rent out less than $1 billion. Building Hopper or Blackwell infrastructure can save costs and provide higher returns Therefore, choosing a modern accelerated computing infrastructure is undoubtedly the best choice.

Q: You have increased your annual operating expenses, and the increase in your procurement commitments and supply obligations also indicates optimism. However, on the other hand, some people believe that many customers may not be ready to use liquid cooling yet. How do you view the growth trajectory of Blackwell?

A: Remember, the world is transitioning from general computing to accelerated computing, and all data centers in the coming years will adopt accelerated computing. As for the growth trajectory of Blackwell, we offer multiple configurations, including air cooling and liquid cooling. Liquid cooling can provide higher AI throughput and is a more cost-effective solution. Therefore, I am confident in our growth.

Q: You expect to achieve billions of dollars in Blackwell revenue in the fourth quarter, is this additional? You mentioned that Hopper demand will strengthen in the second half of the year, does this mean that Hopper will also strengthen in the fourth quarter and Blackwell will add billions of dollars in revenue?

A: Regarding Hopper and Blackwell. We believe that Hopper's growth will continue in the second half of the year, and Blackwell will start to grow in the fourth quarter.

Q: Regarding gross margin, you expect a median range of 70% for the full year, if it is 75% for the full year, does this mean the gross margin in the fourth quarter will be around 71%-72%? How do you view the evolution of gross margin after Blackwell's growth next year?

A: As for the gross margin, we provided a non-GAAP gross margin of 75% for the third quarter, around 75% for the full year. We expect some fluctuations in the fourth quarter, but it will not be as low as you mentioned.

Q: The 10-Q filing mentioned a sequential decline in revenue in the United States, while revenue in some Asian regions increased significantly. Can you explain the dynamics behind this? China performed very well, as you mentioned in your remarks. Can you clarify whether with these favorable revenue dynamics, the overall revenue growth rate in the fourth quarter will accelerate?

A: Our geographical disclosures in the 10-Q can sometimes be challenging because we have to disclose based on sales and invoicing locations. What you see is the invoicing location, not the final destination of the product. These products may be transferred to other locations through OEM or ODM, so the data you see sometimes changes due to this transfer.

Regarding gross margin and the revenue dynamics of Hopper and Blackwell, Hopper's demand is expected to continue growing in the second half of the year, and the fourth quarter will be a growth opportunity. Additionally, we will have contributions from Blackwell.

Q: You have already started an amazing annual product rhythm, and the challenges will only get bigger. Can you talk about the potential impact on your vertical integration, supply chain partnerships, and the impact on your gross margin? A: The reason we can maintain such a high speed is because the complexity of the model is increasing, and we hope to continue reducing costs and increasing the scale of the model. We believe that by continuing to expand AI models, we will reach an extraordinary level of practicality and usher in the next industrial revolution. We will continue to work hard to achieve this goal.

We are able to design and build AI factories because we have all the components. Our supply chain is very extensive and can serve global cloud service providers and enterprise data centers. We do not intend to integrate this part, but rather act as a technology provider.

Q: Regarding the product cycle of Blackwell, how do you view the proportion of rack-level systems in the Blackwell cycle?

A: Blackwell rack systems are designed and architected based on racks, but sold in the form of system components. We design the entire rack, and the software will run perfectly on the entire rack. The integration of components is usually done near the data center because these racks are very heavy. Therefore, our ODM and OEM partners have a very extensive integration network globally.

Risk Disclosure and Statement of this Article: Dolphin Research Disclaimer and General Disclosure