Silicon Valley hard currency Nvidia, which soared 500 billion overnight

The ongoing generative AI race has driven NVIDIA's stock price all the way. Under the global rush for NVIDIA chips, NVIDIA finally ended the year-on-year decline in revenue for many consecutive quarters.

In the early morning of August 24, Beijing time, NVIDIA released its financial results for the second fiscal quarter of fiscal year 2024 as of July 30. In the second fiscal quarter, NVIDIA's revenue was US$13.507 billion, up 101% year-over-year and 88% sequentially, setting a new historical record; Net income was US$6,188 million, up 843% year-over-year and 203% sequentially.

In addition to the continued rapid growth of the data center business (revenue of $10.32 billion in the fiscal second quarter, up 171% year-over-year and 141% sequentially), the gaming business also began to recover, achieving revenue of $2.49 billion in the fiscal second quarter, reversing the decline to 22% year-over-year and 11% sequentially.

Huang Jenxun: Image source NVIDIA official website

During the earnings call, NVIDIA CEO Jensen Huang and NVIDIA CFO Colette Kress exclaimed "tremendous (amazing, excellent)" from time to time. Benefiting from the expected continued rapid development of the data center business driven by AI chips, NVIDIA forecasts that revenue in the fiscal third quarter will hit another record, reaching $16 billion, up or down 2%, well above Wall Street's expectation of $12.59 billion.

Affected by the above positive factors, NVIDIA's stock price rose more than 10% after hours and is now back down to 6.58%. According to this calculation, NVIDIA's market value rose by about 75 billion US dollars (about 545.3 billion yuan) overnight.

Akshara Bassi, senior analyst at Counterpoint, pointed out that despite macroeconomic instability, the huge investment in data centers shows that NVIDIA is strategically re-shifting to AI infrastructure investments as enterprises and cloud providers prioritize data center transformation to accommodate AI workloads. This strategic shift will drive NVIDIA's data center growth and solidify its position as a key player in AI technology.

In the face of external concerns about the supply of chip production capacity, NVIDIA CFO Cress responded that the company is actively working with major suppliers to seek to reduce the time required for the production cycle, and is expected to continue to increase production capacity in the next fiscal quarter and the next fiscal year.

But at present, NVIDIA, which has unlimited development scenery, also hides hidden worries behind it.

The automotive business, which originally accounted for nearly 5%, fell to only 1.9% in the second fiscal quarter, and compared with the more than 100% growth momentum in the previous two fiscal quarters, the year-on-year growth rate in the fiscal quarter was only 15%, Kress said that it was mainly affected by the decline in demand in the Chinese market.

In addition to the automotive business, which is not improving as a new growth curve, the data center business, which is acting as the backbone of revenue, is also expected to suffer a new crisis from the development of the Chinese market.

In response to the impact of possible U.S. upgrades on China's AI regulatory measures, Nvidia said that given the strong global demand for the product, the company does not expect this additional export restriction to have an immediate and material impact on data center GPU financial results, "but if the long-term ban on selling data center GPUs to China is implemented, it will cause NVIDIA industries to lose the opportunity to compete and lead in one of the world's largest markets." ”

A

The data center that surpassed the gaming business for six consecutive quarters and became the backbone of NVIDIA's revenue, once again setting a new revenue record.

In the second quarter, NVIDIA's data center revenue was $10.32 billion, up 171% year-over-year and 141% year-over-quarter, a record increase.

In the face of the huge demand for AI chips from the outside world, Huang Jensen compared NVIDIA to a global AI engine and said that he was "working hard to serve all customers."

Bank of America Securities semiconductor analyst Vivek Aria even likened Nvidia's AI chips to "the hard currency of Silicon Valley."

The typical manifestation of the above hard currency is happening with CoreWeave, a cloud startup led by Nvidia. Recently, the company announced a $2.3 billion financing that it gave to investors as collateral for its own NVIDIA GPUs.

In addition to meeting all customer needs as much as possible, Huang is also accelerating product iteration.

In 1999, Huang Jenxun proposed the "Huang's Law" in the field of graphics card chips called Moore's Law, that is, the performance of display chips doubled every 6 months, which was 3 times faster than Moore's Law every 18 months. It is precisely based on the higher frequency update speed that Huang Jenxun uses a stronger GPU to better meet the operation experience of gamers, laying a good foundation for the advent of the AI chip era.

This rapid iteration capability continues to be practiced by Huang: in March, Huang launched the H100 GPU to meet the growing demand for computing power by enterprises; In May, Jensen Huang released the DGX GH200, a supercomputer consisting of 256 NVIDIA GH200 Grace Hopper superchips; In August, Huang brought a new version of the DGX GH200 with HBM3e memory, which has 1.7 times more memory and 1.5 times more bandwidth than the H100.

In addition to record data center revenue, NVIDIA's gaming business also began to pick up, with second-quarter revenue of $2.49 billion, up 22% year-over-year and 11% year-over-quarter.

A major contributor to the growth of NVIDIA's gaming business is the recovery of the PC market. Canalys second-quarter data showed that global PC shipments reached 62.1 million units, although still in a downward cycle, down 11.5% from the same period last year, but this is the second consecutive quarter of decline in the global PC market since shipments fell by more than 30% in the previous two quarters, indicating that the market is beginning to gradually return to stability.

Specific to the product line, in the second quarter, notebook computer shipments and desktop computer shipments both increased month-on-month, of which notebook computers increased by 5.7% and desktop computers increased by 18.6%. After a period of struggle, PC manufacturers are gradually returning to production to meet consumer demand.

B

The demand in the short supply of market pushed NVIDIA to give a fiscal third quarter performance guidance that exceeded expectations, and predicted that the revenue in the third quarter will reach $16 billion, setting a new record again.

As early as the first fiscal quarter conference call, Huang Yanxun lamented that the data center business has "an incredible number of orders", and NVIDIA's core priority is to develop and ensure the production of as many advanced chips as possible that meet market demand.

But really in the fiscal second quarter, the explosion of external demand still exceeded Huang's expectations. According to CoreWeave co-founder and CTOBrian Venturo, in the first quarter of this year, it was easy to get NVIDIA GPUs, but from April, the market suddenly became extremely tight, and even if you order it now, the delivery time will have to wait until Q1 or even Q2 of 2024.

Alphabet asked Li Peng, a domestic supplier engaged in the development of large models, that the above delivery time was verified, according to him, after sending orders to NVIDIA, it will take six or seven months to get the goods.

To what extent is NVIDIA chip out of stock? The GPU Utils website has given a calculation that the gap is as high as 430,000 for H100 alone.

"Who will get how much H100 and when will get H100 is the hottest topic in Silicon Valley." OpenAI co-founder Andrej Karpathy lamented.

Computing power has become a strategic resource for cloud vendors to develop large models and compete for customers. Under this situation, domestic Baidu, Byte, Ali, Tencent and other large technology companies have also been exposed by the media to place a total of $5 billion in A800 chip orders to NVIDIA. Of these, only $1 billion will be delivered within the year, and another eighty percent of orders will not be delivered until 2024.

For the news of domestic manufacturers rushing to buy chips, NVIDIA said in response to the media: "Consumer Internet companies and cloud service providers invest billions of dollars in data center components every year, and usually place orders months in advance. ”

Just a day before the earnings report, a person familiar with the matter revealed to the media that NVIDIA plans to increase the production of H100 chips by at least twice as much, and the expected shipments in 2024 will be between 1.5 million and 2 million, compared with the expected shipment of 500,000 pieces this year, an increase of more than 2 times.

The production capacity increase plan has also been confirmed by NVIDIA. In the earnings call, when asked whether NVIDIA has plans to increase production capacity next, Kress mentioned that NVIDIA is actively working with major suppliers to reduce the time required for production cycles, and is expected to continue to increase production capacity in the next quarter and next fiscal year.

C

Although with the outbreak of the wave of AI large models, the current development of NVIDIA can be called a scenery, but looking forward to the future, there are some hidden worries.

As a new growth curve, the automotive business said goodbye to the high revenue growth momentum of more than 100% for many consecutive quarters, and in the second fiscal quarter, NVIDIA's automotive business revenue was $253 million, the year-on-year growth rate fell to 15%, down 15% sequentially, and the proportion of revenue also fell from nearly 5% in the past to 1.9%.

In contrast, in the first fiscal quarter of fiscal 2024, NVIDIA's automotive business achieved revenue of US$296 million, a year-on-year growth rate of 114.5%. At that time, NVIDIA looked forward in its earnings report that with the continuous penetration of applications such as autonomous driving, the automotive business is expected to bring new growth to the company.

As Nvidia tilts more resources into AI chips, the growth possibilities of its automotive business are being further eroded. As for the unfavorable situation in the automotive business, Kress said that the main reason was the overall decline in automotive demand, especially in the Chinese market.

Perhaps in order to revive the development of the automotive business, recently, Wu Xinzhu, vice president of autonomous driving at Xpeng Motors, was revealed to join NVIDIA as a global vice president and report directly to Huang Jenxun.

The data center business, which supports NVIDIA's revenue pillar, may also suffer a new round of market hits.

After it was reported that the United States would upgrade its AI regulatory measures in China, the impact of the Chinese market on Nvidia is a topic that cannot be bypassed.

When large domestic manufacturers concentrated on placing orders for $5 billion chips, some insiders analyzed that one of the reasons may be that they are worried that the United States will introduce more stringent sales restrictions in the future. Prior to this, due to the impact of US regulatory measures, NVIDIA could only sell customized versions of A800 and H800 chips with lower prices and castrated performance to the Chinese market.

According to Li Peng, the domestic A800 chip price is more than 100,000 yuan, while the price of A100 chip reaches nearly 300,000 yuan, almost twice the former.

Commenting on the Biden administration's possible new export restrictions on semiconductor technology to further restrict the Chinese market's access to A800 and H800, Huang said in an interview, "If we are deprived of the Chinese market, we have no emergency measures." Because there is no other China in the world, only one China. ”

According to the Semiconductor Industry Association (SIA), in 2022, China's semiconductor purchases amounted to $180 billion, accounting for more than one-third of the total global semiconductor procurement of $555.9 billion, making it the largest global single market.

After the A100 chip ban last year, Morgan Stanley assessed that the potential impact on Nvidia could cost Nvidia about $400 million per quarter. If A800 and H800 data center chip sales require additional licensing requirements, Bank of America analysts predict that this will affect 7% of Nvidia's total revenue and 10% of data center sales. According to the current revenue performance of NVIDIA, its damage scale may be $5 billion.

Commenting on NVIDIA's dominance in AI chips, Inge Heydorn, partner at investment research firm GP Bullhound, commented: "I've been working in the technology sector since 1994, but I've never seen the current environment – too dependent on a single company... AI is really the last pillar of growth, and everyone is relying on it. If NVIDIA performs weakly, the market could see a considerable correction. ”

After taking LinkedIn Weida from a small sampan to a giant ship on the sea, Huang Jenxun will face more and more storms.