Apple (AAPL.US), Microsoft (MSFT.US), Google (GOOG.US), AMD (AMD.US), Intel (INTC.US), Qualcomm (QCOM.US), TSMC (TSM.US) Technology giants that are deeply involved in the field of artificial intelligence have announced their June quarter results, so the hope of the whole village

Apple (aapl.us) , Microsoft (msft.us), Google (goog.us), Advanced Micro Devices (amd.us) , Intel (intc.us), Qualcomm (qcom.us), TSMC (tsm.us) and other technology giants that are deeply involved in the field of artificial intelligence have announced their June quarter results, so the hope of the whole village has fallen on the top artificial intelligence. Smart chip stocks Nvidia (nvda.us) .

In the last week of August 2024, Wall Street trading fell relatively, focusing on Nvidia’s July 2024 financial quarter results.

NVIDIA’s stock price fell sharply after the results

Judging from the performance data, NVIDIA’s overall quarterly performance was higher than the market consensus expectations:

As of the second fiscal quarter of fiscal year 2025 ending on July 28, 2024, NVIDIA’s revenue was year-on-year Increased 122.40% to US$30.040 billion, higher than the market consensus of US$28.42 billion (quoting Zacks data, the same below); non-accounting standards diluted earnings per share were US$0.68, higher than the market consensus of US$0.64.

Judging from the guidance provided by Nvidia for the third quarter of fiscal year 2025, the company expects quarterly revenue of US$32.5 billion (plus or minus 2% range), which is also higher than market expectations of US$30.97 billion.

In addition to performance exceeding expectations, NVIDIA's board of directors also approved an increase of US$50 billion for share repurchases with no deadline, doubling the US$25 billion repurchase limit increased a year ago. By August 26, 2024, Nvidia will have a total of US$53.9 billion available for repurchase.

Even so, NVIDIA, which announced its results after the market closed on August 28, 2024, saw its stock price drop by 6.89% during the extended trading period, as shown in the figure below.

Market participants generally believe that the slowdown in the quarterly growth rate of data center segment revenue, which reflects its AI chip revenue, is the main reason, and they do not know how to envision NVIDIA's prospects.

Caihua News Agency believes that the expected gross profit margin will decline in the next two quarters, and Nvidia's stock price has risen too high and profit-taking may be the main reason for the decline in its stock price.

All business indicators in the second fiscal quarter saw double-digit growth

In the fiscal quarter ending July 28, 2024, data center revenue increased by 154.50% year-on-year and 16.44% quarter-on-quarter to US$26.272 billion. The most important sources of revenue and growth will be described in detail below.

game quarterly revenue increased by 15.85% year-on-year and 8.80% quarter-on-quarter to US$2.88 billion. Management revealed that revenue from consoles, notebooks and desktop games continued to grow, demand was strong and growing, and channel inventory remained healthy. The

professional visualization business had quarterly revenue of $454 million, up 19.79% year-over-year and 6.32% quarter-on-quarter, primarily driven by demand for artificial intelligence and graphics use cases, including model fine-tuning and Metaverse-related workloads.

Automotive and manufacturing were the key industry verticals driving revenue growth. Companies are racing to digitize their workflows to improve operational efficiency. Foxconn, the world's largest electronics manufacturer, is using NVIDIA Omniverse to empower digital twins of its physical factories that produce NVIDIA Blackwell systems. Several large global enterprises, including Mercedes-Benz, have signed multi-year contracts with NVIDIA Omniverse Cloud to build industrial digital twin factories.

Automotive and Robot segment revenue was US$346 million, an increase of 36.76% year-on-year and 5.17% quarter-on-quarter. The year-on-year growth was mainly driven by the increase in new customers for autonomous driving platforms and the increase in demand for AI cockpit solutions.

On the consumer side, Boston Dynamics, BYD Electronics , Toge, Intrinsyc, Siemens and Teradyne Robotics are using the Nvidia Isaac robotics platform to develop autonomous robotic arms, humanoid robots and mobile robots.

As shown in the figure below, all Nvidia business segments have achieved double-digit growth year-on-year, with the most significant growth in data centers.

The sustainability of the rapid growth of the data center business

As mentioned above, the data center’s second fiscal quarter revenue reached US$26.272 billion, mainly driven by strong demand for NVIDIA hopper, GPU computing and network platforms. Among them, computing revenue increased by more than 2.5 times, and network revenue increased by more than 2 times year-on-year. Cloud service providers account for approximately 45% of their data center revenue, with more than 50% of revenue coming from consumer Internet and enterprise customers.

Nvidia revealed at the performance conference that its customers continue to promote the purchase of hopper architecture and are preparing to adopt blackwell. Key workloads driving its data center growth include generative AI model training and inference; pre- and post-processing of video, image and text data using CUDA and AI workloads; synthetic data generation; AI recommendation systems; SQL and vector database processing.

management expects that next-generation models will require 10 to 20 times more computing power to train on more data, and this trend will continue.

Nvidia estimates that inference has contributed more than 40% to its data center revenue growth over the past four quarters. Consumer Internet companies (CSPs) and enterprise customers benefit from NVIDIA's inference platform. Demand for Nvidia comes primarily from front-end model creators, consumer internet services and tens of thousands of enterprises and startups creating generative AI applications for consumer, advertising, education, enterprise and healthcare, and robotics.

Driven by strong growth in CSP demand, the Nvidia h200 platform began to climb in the second quarter, mainly delivered to large CSP and enterprise customers. NVIDIA h200 is built on the hopper architecture and can provide 40% more memory bandwidth than the h100.

With the NVIDIA MGX modular reference architecture, its OEM and ODM partners are quickly and cost-effectively building more than 100 Blackwell-based systems. The nvidia blackwell platform brings together multiple GPUs, CPUs, DPUs, nvlinks and link switches as well as network chips, systems and nvidia cuda software to empower the next generation of artificial intelligence across use cases, industries and countries. The nvidia gb200 nvl72 system using fifth-generation nvlink enables 72 GPUs to work like one GPU and provides 30 times faster inference speed for large model workloads and unlocks the ability to run trillion-parameter models in real time.

What everyone is more concerned about is the demand and progress of NVIDIA's hopper and blackwell, because this will be the future growth driver of NVIDIA's data center.

hopper and blackwell

Nvidia said that hopper is in strong demand and blackwell is being widely trialled. It has made adjustments to the blackwell gpu mask (for blackwell gpu) to increase production.

In the second quarter, NVIDIA will provide customers with samples of the blackwell structure. Blackwell plans to increase production starting in the fourth quarter and continue until fiscal year 2026. Demand for blackwell platforms remains much higher than supply, and this trend is expected to continue into next year. NVIDIA expects Blackwell's revenue to reach billions of dollars by the fourth quarter.

Huang Renxun emphasized at the performance conference that when talking about blackwell’s production and delivery in the fourth quarter, it did not “just” start production, but has already been delivered: “We are transforming two platforms at the same time. One is from general computing to Turning to accelerated computing, this is because CPU expansion will gradually slow down. It is now as slow as a crawl, but the computing demand is growing significantly, at least doubling every year, so we have no other way. Computing expansion will push up every The cost of each company will also push up the energy consumption of global data centers. It can only turn to accelerated computing to increase the speed of applications, so that larger-scale calculations can be performed, such as scientific simulations or database processing, but this directly translates into more Low cost and lower energy consumption.The second is the empowerment of accelerated computing, because we have reduced the cost of training large language models or training deep learning, and now we can have larger-scale models - hundreds of millions of parameter models, and train on top of them - just Input world knowledge topics and let the model figure out how to understand human language expressions, how to program knowledge into neural networks, and how to learn to reason, triggering a generative AI revolution. "

Jensen Huang expects the growth of the data center business to be quite significant next year, and blackwell will completely change the rules of the industry.

hopper supply and production have improved, and hopper shipments are expected to increase in the second half of fiscal 2025. The chief financial officer said in the results release Said at the meeting: "We believe that hopper will continue to grow in the second half of the year, we have many new products for hopper, our existing hopper products will continue to climb from the next few quarters, including Q3, and new products will For growth in the fourth quarter, hopper will be a growth opportunity compared to the first half of the year. "

However, compared with the rapid growth in the previous few quarters, the data center's annual and quarterly growth rates have slowed down in these two quarters, as shown in the figure below. After the highly anticipated dual launch of unknown hopper and blackwell, Will they replace each other and lead to slower revenue growth?

Huang Renxun particularly emphasized that blackwell is a leap forward from hopper. Blackwell is an artificial intelligence infrastructure platform, not just a GPU.

He said: "With the As we show more blackwell and sample systems to our partners and customers, the future of blackwell will become clear. Blackwell's vision took almost 5 years and 7 unique chips to realize, gray CPU, blackwell dual gpu and a color pack, connectx dpu and bluefield dpu, nvlink switch for gpu communication, and infiniband and ethernet Quantum and spectrum-x of the Internet can support the huge traffic of AI. The Blackwell Artificial Intelligence Factory is building scale computers. NVIDIA has designed and optimized the blackwell platform, from chips, systems, networks, and even structured cables, power and cooling, to a large amount of software, full stack end-to-end, enabling customers to quickly build artificial intelligence factories. These are very capital-intensive infrastructures.

Customers want to deploy devices as soon as they get them and deliver the best performance and total cost. Blackwell delivers three to five times the AI ​​throughput of Hopper in power-constrained data centers.

The overall gross profit margin is likely to decline in the future.

In the second fiscal quarter ending on July 28, 2024, NVIDIA's gross profit margin was 75.15%, and the non-accounting standards gross profit margin was 75.66%, a quarter-to-quarter decrease of 3.21 percentage points and 3.28 percentage points, Nvidia explained, mainly due to the increase in the proportion of new products in the data center and the allocation of inventory for low-efficiency blackwell materials.

Nvidia estimates that total revenue in the third quarter may be US$32.5 billion, with a fluctuation range of 2%. This revenue guidance combines considerations such as the continued growth of the hopper structure and the trial of blackwell products. Blackwell's production is expected to climb in the fourth quarter. GAAP and non-GAAP gross profit margins are expected to be 74.4% and 75%, with a range of 50 basis points. The data center portfolio will continue to shift toward new products, with this trend expected to continue into the fourth quarter of fiscal 2025. For the whole year, gross profit margin is expected to be in the mid-70% range.

NVIDIA’s non-accounting standards gross profit margin in the first and second quarters of fiscal year 2025 were 78.94% and 75.66% respectively, while it expects the gross profit margin in the third quarter to be 75%. This may be due to the increased contribution of new products. The latter The profit margin is relatively low. The expected gross profit margin for the whole year may be about 75%. Based on this calculation, the company's gross profit margin in the fourth quarter is likely to be lower than 75%.

Considering that blackwell will start delivery from the fourth quarter, profits may be further diluted. According to its management, the gross profit margin may decline in the future.

Caihua News Agency believes that this should be one of the reasons for the decline in Nvidia’s stock price.

NVIDIA’s pressure

“If you want to wear a crown, you must bear its weight.”

As the undisputed leader in artificial intelligence chips, Nvidia’s stock price has been chased very high because investors have high hopes for its future AI cores. Once the real situation is not as expected, investors will revise Nvidia's valuation.

AMD and Intel have both released their own AI chips and related products to challenge Nvidia's leadership. Although their delivery times may be several quarters behind, if these products bear fruit, they will inevitably become powerful competitors.

Since the beginning of this year, the stock prices of AMD and Intel have fallen by 0.71% and 60.45% respectively. Caihua News Agency estimates that after the announcement of the June financial quarter, the stock price of AMD has increased by 5.72%, while Intel has fallen by 32.22%. %.

Nvidia’s stock price has risen by 153.68% this year. This post-performance stock price adjustment may be due to investors’ profit-taking, or it may be due to concerns about its prospects – in addition to AMD and Intel, Nvidia’s End customers such as Tesla (tsla.us) are developing their own AI chips to reduce their dependence on NVIDIA, so there are still questions about whether NVIDIA's return on investment is as ideal as expected.

Author: Mao Ting