Since the beginning of the new year, Sora, a video application for cultural and educational purposes, has once again sparked a wave of enthusiasm in general artificial intelligence, and Nvidia, a global computing leader, has also shouldered higher market expectations. On February 22, Beijing time, Nvidia released its fourth quarter financial report, which doubled again. Its revenue for the 2024 fiscal year reached a historic high of $60.9 billion, an increase of 126%.
At the performance briefing, company executives stated that countries around the world are investing in AI to support the construction of large-scale language models in their own languages, domestic data, and support for local research and corporate ecosystems. However, due to the impact of US export controls, the proportion of data center revenue contribution in China has rapidly declined in the latest fiscal quarter, and the company has developed and delivered alternative products to Chinese customers.
In terms of the market, NVIDIA's stock price fell during the trading session, closing down 2.85% at $674.72 per share, with a total market value of $1.67 trillion. However, NVIDIA's stock price rose after the trading session, with an increase of over 8% as of the time of writing.
Performance growth exceeds guidance
NVIDIA's latest earnings report shows that as of January 28, 2024, its fourth quarter revenue was $22.1 billion, a 22% increase compared to the previous quarter and a 265% increase compared to the same period last year. Its net profit reached $12.285 billion, a 769% increase compared to the same period last year, and GAAP diluted earnings per share were $4.93.
In the fiscal year 2024, Nvidia's revenue increased by 126% to $60.9 billion, with a net profit of $29.76 billion, a year-on-year increase of 581%. GAAP diluted earnings per share were $11.93. The financial data exceeded the previous performance guidelines.
"Accelerated computing and generative AI have reached a critical point. The demand of companies, industries and countries around the world is surging." Jensen Huang, the founder and CEO of Nvidia, introduced the demand for data processing, training and reasoning of large cloud service providers and GPU professional providers, as well as enterprise software and consumer Internet companies, and the demand for cars The vertical industries led by financial services and healthcare are both forming diversified driving forces for data center business growth.
Nvidia expects to achieve revenue of $24 billion in the first quarter of the fiscal year 2025, with a fluctuation of 2%. The GAAP and non GAAP gross profit margins are expected to be 76.3% and 77.0%, respectively, with a fluctuation of 50 basis points.
China's regional income is experiencing a "rapid freeze"
The details of each business segment are as follows:
Data Center: Revenue reached a record breaking $18.4 billion in the fourth quarter, a 27% month on month increase and a 409% year-on-year increase. The annual revenue increased by 217%, reaching a record breaking $47.5 billion.
Game: Revenue in the fourth quarter was $2.9 billion, unchanged on a month on month basis and a year-on-year increase of 56%. The annual revenue increased by 15% to reach 10.4 billion US dollars.
Professional visualization: The fourth quarter revenue was $463 million, an increase of 11% month on month and 105% year-on-year. The annual revenue increased by 1% to reach 1.6 billion US dollars.
Automobile: The fourth quarter revenue was 281 million US dollars, an increase of 8% month on month and a decrease of 4% year-on-year. The annual revenue increased by 21% to 1.1 billion US dollars. In the automotive business, Great Wall Motors, ZEEKR and Xiaomi use DRIVE Orin to provide power for the intelligent auto drive system. Ideal cars choose DRIVE Thor as their central on-board computer.
From a business perspective, the main data center of revenue still maintains strong growth, with annual revenue reaching 47.5 billion US dollars, a year-on-year increase of twice. However, according to regional division, the proportion of income in China has rapidly declined.
At the performance briefing, Nvidia executives stated that China accounted for a single digit percentage of the company's data center revenue in the fourth quarter (approximately 5-9%), and it is expected to remain within a similar range in the next fiscal quarter. Except for China, the company's data center growth from all other regions is strong. Data shows that in the past quarter, Nvidia's data center business revenue from restricted countries and regions such as China contributed about 20-25%.
It is reported that after the US government established export control regulations in October last year, Nvidia has not yet obtained a license from the US government to ship restricted products to China. However, the company has started shipping alternative products to China.
"The US government hopes to restrict the Chinese market from gaining the latest capabilities in accelerated computing and AI. When the restriction measures were introduced, we immediately suspended the delivery of related restricted products." Huang Renxun said that for the Chinese market, the company has been redesigning products and is now providing samples to customers. The company will make every effort to win success in the Chinese market within the regulatory scope of the restrictions.
However, Nvidia's AI replacement products are not well received. According to the news, Nvidia has launched new products HGXH20, L20 PCIe, and L2 PCIe in accordance with its export policy, but the computing power is only 20% of H100. It plans to mass produce in the second quarter of this year, but Chinese companies are not paying for it. In addition, Internet manufacturers at home and abroad have developed AI chips by themselves or invested in related manufacturers. AI giant OpenAI is planning to raise $7 trillion to manufacture self-developed AI chips, so as to get rid of dependence on Nvidia AI chips and compete with Nvidia.
Supply chain improvement and increased competitive pressure
The strong demand for AI has extended the supply cycle, causing Nvidia to be criticized by customers. At this briefing, Huang Renxun stated that the supply chain of Nvidia supercomputers has been comprehensively improved in various aspects, including wafers, packaging, storage, and networks. It is reported that a single Hopper GPU has 35000 parts, not just a single new product.
According to media reports, Nvidia A100 and H100 GPUs are produced by TSMC's CoWoS packaging technology. However, with the surge in demand for generative AI, TSMC's CoWoS production capacity is severely constrained. Therefore, Nvidia has certified other CoWoS packaging suppliers for production capacity backup.
According to the latest report released by UBS analysts, the delivery speed of Nvidia AI GPUs is continuously accelerating, with 8-11 months left at the end of last year and now reduced to 3-4 months. The delivery time has been significantly shortened, and the market was once concerned about the slowing release of AI demand.
In addition, Nvidia is facing increasingly fierce market competition. In addition to traditional competitors such as AMD and Intel, the recently launched chip inference speed by startup Groq is 10 times faster than Nvidia GPUs, with a cost of only 1/10 of it. The large model generation speed is close to 500 tokens per second, crushing ChatGPT-3.5 at a speed of about 40 tokens per second. The AI chip adopts one of the fastest read and write storage devices currently available, SRAM, which has driven the rapid rise of A-share, Beijing Junzheng, and other related storage concept stocks in recent days.
Analysts have divergent views on the subsequent trend of Nvidia's stock price.
Morgan Stanley believes that although Nvidia's chip supply is still facing bottlenecks, it cannot be denied that the market demand for AI chips is still strong. Although cloud service providers are still searching for alternatives to Nvidia H100 and B100, it currently appears that Nvidia remains the undisputed king in this chip war. It is expected that B100 will become a game changer for artificial intelligence, even more powerful than the previous generation flagship AI chip H100. For Nvidia's long-term trend, Morgan Stanley maintained its target price of $750, believing that technology stocks were far from a foam.
However, some argue that the market's pursuit of Nvidia should return to rationality, and high expectations may mean that Nvidia cannot meet them.
Cathie Wood, a star fund manager on Wall Street and CEO of Ark Investment Management, admitted in a media interview that Nvidia is the undisputed leader in the AI field, but the market's expectations for it are too high, and Nvidia may not be able to meet them. With intensified competition and inventory adjustments posing challenges, she recently sold Nvidia stocks worth approximately $4.5 million. It is reported that large technology companies such as Meta, Tesla, Alphabet, and Amazon are joining the chip war, which may put pressure on Nvidia's future demand.