On August 29th Beijing time, NVIDIA finally released its financial data for the second quarter of the fiscal year 2025. The financial report showed that both revenue and profit continued to grow exponentially. NVIDIA's revenue for the second quarter was $30 billion, a year-on-year increase of 122%. Under the Generally Accepted Accounting Principles (GAAP), the net profit was $16.6 billion, a year-on-year increase of 168%, and the non-GAAP net profit was $16.95 billion, a year-on-year increase of 152%, with a gross margin as high as 75.7%.

According to the company's performance guidance, the revenue for the third quarter is expected to reach $32.5 billion, with a fluctuation of 2%. At the same time, NVIDIA also announced the approval of an additional $50 billion stock repurchase plan and maintained the quarterly dividend per share at 1 cent unchanged.

Looking at the performance of the second quarter, NVIDIA's revenue continues to grow at a triple-digit rate. However, simple incremental growth can no longer meet the high expectations of investors. Optimists firmly believe in AI and expect NVIDIA's acceleration to continue to speed up; pessimists believe that AI is overestimated with a bubble, and NVIDIA's stock price will also face fluctuations.

Advertisement

After the release of the financial report, as of the closing of the day, NVIDIA's stock price fell by 2.1% to $125.61 per share, and fell nearly 7% after the market. However, NVIDIA's market value is currently fluctuating around $3 trillion, and the stock price has soared by more than 150% since the beginning of the year. Morgan Stanley's recent target price under normal circumstances is $144.

At present, NVIDIA has become a symbol of the AI civilization era. For "believers," a 2% drop is not considered a fluctuation. Although compared with the 262% revenue growth in the first quarter, the growth rate has slowed down, many institutions believe that NVIDIA will still develop beyond expectations.

Morgan Stanley stated in its research report: "NVIDIA is still our top choice. Regardless of expectations, the fact is that various headwinds, including last year's U.S. export control to China, the challenge of launching data centers, the potential suspension of Blackwell, and the current strategic delay of new products, have not affected the company's strong growth momentum. This is the reason for our view, and we expect that visibility will increase as new products are gradually launched."

The report believes: "The only thing to pay attention to is that investors' expectations are rising." Now, what investors are most concerned about is the situation of the delay of Blackwell. Morgan Stanley pointed out that the monthly data of the industry chain continues to show that production is very strong.

NVIDIA founder and CEO Jen-Hsun Huang responded in the financial report conference call that the demand for Hopper architecture chips is still strong, and the industry's expectations for the Blackwell architecture are incredible.

Blackwell mass production in the fourth quarter

Looking at the business, NVIDIA's data center revenue for the second quarter was $26.3 billion, a year-on-year increase of 154%, and a sequential increase of 16%; the revenue of gaming and AI PC business was $2.9 billion, a year-on-year increase of 16%, and a sequential increase of 9%.Professional visual business revenue was $454 million, a year-over-year increase of 20%, and a sequential increase of 6%; automotive electronics and robotics business revenue was $346 million, a year-over-year increase of 37%, and a sequential increase of 5%.

Among them, the data center business is still growing rapidly, but the increase in this business in the latest fiscal quarter is lower than the 427% increase in the first fiscal quarter. In terms of revenue share, the data center business accounted for 88% of total revenue in the second fiscal quarter. NVIDIA's data center revenue share surpassed gaming for the first time in the fiscal year 2023, and two fiscal years later, the share has further increased. NVIDIA's data center solutions based on liquid cooling technology are also increasing in market share.

After the release of the financial report, NVIDIA's management discussed computing trends in the performance call, stating that the next generation of models will require a 10 to 20 times increase in computing power to provide sufficient training for data, and it is expected that this growth trend will continue for four quarters. It is expected that more than 40% of data center revenue will continue to come from cloud service providers, consumer internet companies, and enterprise customers.

Blackwell was the focus of attention at the financial report meeting. Recently, there have been reports that the next generation of Blackwell series AI chips have been forced to postpone the release due to design defects. NVIDIA previously responded to reporters from the 21st Century Economic Report, saying: "The demand for Hopper architecture is strong, and the sample trial of Blackwell has already begun widely, and the output is expected to increase in the second half of the year. In addition, we do not comment on rumors."

Huang Renxun stated that the Blackwell chip does not need to undergo a "functional change" and is expected to start mass production in the fourth quarter, and there will be billions of dollars in sales revenue from the end of the year. NVIDIA CFO Colette Kress said that the company will continue to send Blackwell samples to partners and customers, but the company needs to "make changes to the mask of the Blackwell GPU to improve the production yield of Blackwell."

According to data provided by TrendForce to the 21st Century Economic Report, the Blackwell series is still in the early stage of shipment in 2024, and by 2025, Blackwell will become the main force of shipment, meeting the demand of CSPs (cloud vendors) and OEMs (original equipment manufacturers) for high-end AI servers with high-performance B200 and GB200 Rack.

The B100 is a transitional product, mainly focusing on low power consumption. After NVIDIA ships the existing cloud vendor orders, the B100 will gradually be replaced by B200 and GB200 Rack. TrendForce estimates that in 2025, the Blackwell platform will account for more than 80% of NVIDIA's high-end GPUs, and will promote the annual growth rate of NVIDIA's high-end GPU series to 55%.

At the same time, TrendForce pointed out that due to the tight capacity of CoWoS-L packaging, NVIDIA will provide the B100 and B200 capacity to cloud vendors with greater demand, and plans to supply in batches after the third quarter of 2024. While the yield and mass production of CoWoS-L are still to be prepared, NVIDIA is also planning to downgrade the B200A for other enterprise customers (aiming at edge AI) and switch to using the CoWoS-S packaging technology. According to a well-known artificial intelligence entrepreneur, NVIDIA's new generation of chips has adopted the latest integrated packaging and testing technology.

The demand for H200 and H20 is still strong.

Morgan Stanley's latest research report points out that NVIDIA's stock has largely eliminated concerns about possible delays in Blackwell, as the short-term business performance is strong, and it is expected that this year will see the expected growth of Blackwell. The report also stated: "Despite some temporary Blackwell delays, we expect the management to still indicate mass production and accelerate at the end of the fiscal year."From the earnings call perspective, Nvidia's management is indeed conveying confidence in production and demand. In addition to Blackwell, the shipment situation of Nvidia's last-generation Hopper architecture product series is also closely watched by the market. Nvidia's management stated that demand for Hopper remains strong in the second half of this year, with both Hopper and Blackwell experiencing robust demand, allowing for coexistence.

Jensen Huang indicated that Blackwell still requires time for capacity establishment, and generative AI by the end of the year will bring more vitality to the market. The choice between Hopper and Blackwell is up to the market and customers.

"Demand signals show no signs of slowing down, and the demand for H200 and H20 remains strong, which is expected to be sufficient to offset the potential revenue shifts from Blackwell's delay from January and April to later quarters. Supply and customer demand have quickly shifted towards H200," said Morgan Stanley.

TrendForce's survey of the supply chain also shows the same trend, pointing out: "In 2024, Nvidia's high-end GPU shipments will be primarily based on the Hopper platform, with H100, H200, and other models shipped to North American CSPs and OEMs, while for Chinese customers, AI servers equipped with H20 will be the main focus. It is estimated that H200 will only start to ramp up in the third quarter of 2024, becoming Nvidia's mainstream model, and continue into 2025."

In Morgan Stanley's view, customer enthusiasm for Blackwell is very high, but in the initial growth phase, the practicality of a small number of Blackwells is relatively low. For hyperscale computing companies, a large number of Blackwell clusters are needed to achieve relevant results. Therefore, for some customers, they prefer to get more Hoppers.

Additionally, the capacity increase of H20 is also an important contributing factor. H20 is a downgraded version of Hopper, and the research report points out: "Based on a survey in Asia a few weeks ago, we believe that the annual production of H20 has increased to about 1.2 million units, with an average selling price of around $130,000 to $150,000. We are not sure if all these products have been sold, but according to our understanding of Chinese hyperscale computing companies, the demand for H20 is indeed strong, so we expect that H20 will generate more than $10 billion in revenue in the next three quarters."

Overall, the demand for Nvidia chips in the industry chain is still on the rise. However, due to the third-quarter revenue guidance increase not meeting the analysts' expected triple-digit growth, coupled with concerns about profit margin compression, there has been some fluctuation in the capital market. Nevertheless, the aforementioned research report points out that the gross margin can still reach 74%, far exceeding some pessimistic expectations in the market.

Regarding future accelerated computing demand, Jensen Huang stated that computing is shifting from general computing to accelerated computing, with data centers worth $1 trillion being built globally, and the total value of future data centers will increase. Global data centers are accelerating their development, and accelerated computing has reached a tipping point, with developers doing everything possible to deploy accelerated computing, expecting a significant growth in data center business next year.