Nvidia's latest performance once again exceeded industry expectations.
On February 21, Eastern Time, NVIDIA released its fourth quarter results for the 2024 fiscal year, showing that its total revenue and data center revenue reached historic highs during the quarter, and its revenue for the 2024 fiscal year also achieved a record breaking performance of $60.9 billion. According to the financial report, Nvidia's fourth quarter revenue was $22.1 billion, a 22% month on month increase and a 265% year-on-year surge; Net profit was 12.3 billion US dollars, a year-on-year increase of 765%.
After the strong financial report was released, NVIDIA's stock rose more than 12% before trading, reaching a new historical high, with a total market value increasing by $200 billion to $1.69 trillion.
Behind the unexpected performance is the result of the extended development of generative AI towards multimodality, multi data, multi-dimensional, and other directions. At the performance exchange meeting held on the same day, NVIDIA founder and CEO Huang Renxun once again stated that the computing era is changing: transitioning from general computing to accelerated computing - GPUs are being favored by more CSPs (cloud service providers).
21st Century Business Herald reporters have reviewed Nvidia's financial reports over the past two years and found that its data center business only contributed 45% of its revenue in the early 2023 fiscal year, but by the fourth quarter of 2024, it had already accounted for 83%. In the fourth quarter of fiscal year 2024, Nvidia's data center business achieved revenue of $18.4 billion, a year-on-year surge of 409%.
The other side of rapid growth is tight supply. Nvidia's senior management has repeatedly expressed that they are addressing the issue and have made improvements, but it is estimated that the tight supply situation will continue this year. Because Nvidia's product includes multiple supply chain links and multiple components, it is difficult to quickly fill the supply gap.
Depth Layout
The development prospects of NVIDIA are closely related to the future direction of data centers during the performance meeting.
Huang Renxun provided a detailed description of this: the current growth trend is still in a relatively early stage. General computing power represented by CPUs is gradually losing its appeal, and CSP tends to only use GPUs when conducting business. "We have some general-purpose computing data centers, and the depreciation time of infrastructure has been extended from 4 years to 6 years." He gave an example, which means that when CPUs cannot significantly meet the demand for accelerated computing, GPUs have become an inevitable choice.
Nvidia stated that in the past year, approximately 40% of data center revenue has been generated through artificial intelligence reasoning, and building and deploying artificial intelligence solutions has involved almost every industry. The vertical industries led by automobiles, financial services, and healthcare currently have a scale of billions of dollars.
In addition, Nvidia is also laying out GPUs in the dedicated CSP market. At present, companies such as Adobe, SAP and the consumer Internet all want to provide super personalized services to achieve early value, which is the driving force for laying out the dedicated market.
In addition to existing GPU products, there have been recent market reports that Nvidia is forming an ASIC dedicated chip team with the aim of expanding its influence in the AI chip field. Although Nvidia has not officially confirmed this news, the market generally recognizes this trend.
ASIC is one of the AI chips, but unlike GPUs, ASIC is a product with high customization properties and is more versatile.
Lu Jing, Partner and Managing Director of Sullivan Greater China, told 21st Century Business Herald that Nvidia, as a globally dominant artificial intelligence chip designer and supplier, is preparing to enter the ASIC field as part of its efforts to capture the explosive market for customized artificial intelligence chips and protect itself from the increasing number of companies seeking alternatives to its products. "Nvidia's entry into the ASIC field is likely to erode Broadcom and Meiman's market share in the ASIC field."
Zhang Yubin, a senior analyst in the IT department of Sigmantell, also analyzed to reporters that ASIC and GPU designs are similar in certain aspects, such as parallel processing and low-power design. By laying out ASICs, Nvidia can apply its technological strength in the GPU field to ASIC design, achieving technological collaboration and resource sharing.
In addition to hardware, Nvidia has recently launched Chat With RTX, demonstrating its positive attitude in promoting application implementation. This is a personalized AI chatbot suitable for Windows, with local file and online video retrieval and analysis capabilities. The industry believes that this move will play a role in enriching the ecosystem for the implementation of AI PC.
"This is a competition in the computing power center on the PC side." Zhang Yubin analyzed that Nvidia has a significant technological advantage in the GPU field, and Intel is also trying to integrate AI functionality into its CPU products, thus competing in the AI ecosystem. "Nvidia's move will further drive the end side, especially the landing speed of AI functions on PCs, and accelerate the transfer and tilt of computing power towards the end side."
In his view, this move will also enhance the AI experience of end users and accelerate the penetration of AI PCs in mid to high end models, especially laptop devices equipped with independent graphics cards. "However, we still need to maintain a cautious and optimistic attitude, as the development of AI in the PC field still requires continuous observation and verification. We predict that AI's development in the PC field will make significant progress after 2025."
The position of the head is difficult to shake
Under this wave of generative AI, Nvidia has become the most profitable chip giant. Especially with the current tight supply chain capacity, technology giants are actively cooperating with it. But precisely because of this, the high cost issue of Nvidia GPUs has always been a concern. From an application perspective, self-developed chips, driven by cost constraints, have become another solution.
The current industry consensus is that Nvidia's absolute leading position is difficult to easily shake due to its deep GPU hardware and software ecosystem. But the market is also willing to cultivate second suppliers or consider other supply routes.
This is the background of Groq's recent explosive launch of LPU chips. Although the industry generally believes that LPUs currently only appear to have strong computing power and cost issues cannot be ignored, it is undoubtedly one aspect of the current surging competition in the AI chip market.
Semianalysis predicts that Nvidia will remain the main force in GPU capacity allocation, accounting for approximately half of the capacity, based on Cowos (advanced packaging led by TSMC) capacity allocation; Next is Broadcom, which is a collaboration giant between Google TPU and Meta's first generation chip design chip, meaning that cloud computing giants will increase their self-developed chips; In addition, the proportion of Marvell (a partner company of Amazon Traineum2 streaming) and AMD is gradually increasing.
As latecomers, even though they are already giants in fields such as cloud computing, manufacturers such as Google and Amazon find it difficult to quickly break away from their dependence on Nvidia through self-developed chips. This is also the foundation of Nvidia's years of deep accumulation: software ecology and supply chain support.
Zhang Yubin believes that the dominant position of GPUs is difficult to shake, and the main limiting factors include technical difficulty, market acceptance, ecosystem construction, etc. Self developed AI chips require a large amount of research and development resources and time, and the technical difficulty is relatively high. And new chip products need to undergo market inspection and recognition in order to be widely applied. In addition, ecosystem construction needs to attract support and promotion from developers, partners, and other parties.
Lu Jing also stated that an increasing number of large model manufacturers choosing to develop their own AI chips will not shake Nvidia's dominant position in the short term.
"Firstly, Nvidia has an indisputable first mover advantage and rich industry experience. Secondly, Nvidia has built a complete CUDA ecosystem, and hastily replacing the ecosystem means that the cost of learning, trial and error, and debugging for manufacturers will increase. In addition, Nvidia has further consolidated its moat through investment. Since 2023, Nvidia has invested in more than 20 companies, including large new artificial intelligence platforms and small startups applying artificial intelligence to industries such as healthcare or energy." He further pointed out that the most important bottleneck for large model companies' self-developed chips is that supply chain issues cannot be solved in the short term. ". "The production capacity of foundries and wafer fabs has reached saturation, and considering comprehensive costs, self-developed AI chips may not necessarily have more advantages than external procurement. In the short term, Nvidia's CUDA ecosystem remains stable, and most users who need to train chips will still choose Nvidia."