The 21st Century Economic Reporter Ni Yuqing reports from Shenzhen There is no suspense. NVIDIA once again produced a report card that exceeded expectations, and its revenue and data center business reached a record high.
On May 23, Nvidia released its first quarter financial report of fiscal year 2025. Under the U.S. General Accounting Standards (GAAP), its revenue reached $26.04 billion, up 262% year on year and 18% month on month; Net profit was 14.88 billion US dollars, up 628% year on year and 21% month on month.
At the same time, NVIDIA's gross profit margin in the first quarter was as high as 78.9%. It is expected that the revenue in the second quarter will be $28 billion, up or down 2%, which is also higher than analysts' expectations. Nvidia also predicted that the gross profit rate for the whole year is expected to be around 70%.
Huang Renxun, the founder and CEO of Nvidia, said frankly: "We are ready for the next wave of growth. The Blackwell platform has been fully put into production, laying the foundation for trillion parameter level generative AI."
It is worth noting that Nvidia has announced the plan of one split and ten share splits. After the close on June 7, shareholders holding Nvidia's common shares will receive nine additional shares, and from June 10, they will start trading on the basis of split adjustment. On this basis, quarterly cash dividends increased by 150% to $0.01 per share. The stock split makes it easier for investors to buy Nvidia shares, which is generally considered to be good news in the US stock market, and also reflects Nvidia's confidence in the company's growth.
Recently, major institutions have also raised the stock price expectation of Nvidia. According to the report released by Morgan Stanley, the target price of Nvidia's stock is expected to be 1000 dollars; HSBC reported raising the target price to US $1350 (previously US $1050).
As of the close of May 22, US Eastern Time, NVIDIA's share price was 949.50 US dollars/share, with a market value of 2335.6 billion US dollars, but after hours, its share price exceeded 1000 US dollars. Since the beginning of the year, Nvidia's stock price has risen by about 92%. Since March this year, its market value has fluctuated around 2.3 trillion dollars. Among the technology companies, only Microsoft and Apple are the first, and Huang Renxun is the second.
Data center revenue soared by 427%, accounting for nearly 90%
Specifically, on the core business line, in the data center sector, which accounts for the highest proportion, Nvidia's revenue in the first quarter reached a record $22.6 billion, an increase of 23% month on month and 427% year on year.
Based on this calculation, NVIDIA's data center has accounted for 87% of its revenue. Speaking of growth, Huang Renxun said: "The growth of data centers is due to the strong demand for generative AI training and reasoning on the Hopper platform. In addition to cloud service providers, generative AI has also expanded to consumer Internet companies, as well as enterprises, sovereign AI, automobile and medical care customers, creating multiple vertical markets worth billions of dollars."
At the same time, Blackwell's new chip products are being produced and shipped, which will also become Nvidia's new "money printer". The HSBC report points out that Nvidia's chip pricing has further increased. Among them, the ASP (average selling price) of GB200 is US $60000 to US $70000, while the ASP (average selling price) of independent B100 GPU is US $30000 to US $35000. In addition, the report also believes that Nvidia is expected to achieve revenue of US $196 billion in fiscal year 2026.
A report recently released by Morgan Stanley shows that the supply chain has started the cooperation of GB200 DGX/MGX server system project, and is carrying out design optimization and various tests. According to the situation of Asian semiconductor supply chain, based on CoWoS (packaging technology) capacity, it is estimated that about 420000 GB200 super chips will be supplied to downstream in the second half of 2024, and there may be a quarterly lag from chip shipment to downstream assembly. It is estimated that by 2025, according to the allocation of CoWoS capacity, there will be about 1.5 million to 2 million GB200 chips produced.
In addition to Blackwell, Huang Renxun also mentioned that "Spectrum X (switch) has opened a new market for us, which can introduce large-scale AI into Ethernet private data centers. NVIDIA NIM is a new software product. It provides enterprise level, optimized generative AI through an extensive ecosystem partner network, and can run on CUDA anywhere (from the cloud to the local data center and RTX AI PC). "
It can be seen that NVIDIA is building a new system to sell "data centers" with more comprehensive solutions, rather than just chips, from computing chips, transmission links to software systems.
Looking at the game and AI PC business, NVIDIA's related revenue in the first quarter was 2.6 billion US dollars, down 8% month on month and up 18% year on year; Meanwhile, NVIDIA's professional visualization business revenue during the period was $427 million, down 8% month on month and up 45% year on year; The revenue of auto and robot business was $329 million, up 17% month on month and 11% year on year. Business outside the data center has also achieved double-digit growth. At present, NVIDIA has maintained its momentum in four core businesses.
In Huang Renxun's view, AI will bring significant productivity improvement to almost every industry
The demand for AI is still growing, Brady Wang, deputy director of semiconductor research at Counterpoint Research, told 21st Century Business Herald that the industry has shifted from training AI models only on LLM to deploying multimodal models. In addition, the growing importance of AI reasoning has prompted the large-scale data center to significantly expand its infrastructure, thus making Nvidia the largest beneficiary among AI hardware suppliers.
At the same time, Nvidia achieves growth by increasing GPU sales and its diversified AI software revenue. According to its enterprise AI software suite pricing model of $1 per GPU per hour, Nvidia will also significantly increase its software revenue.
Cloud manufacturers actively purchase AI servers, and chip demand is rising
The core driving force of Nvidia's rapid growth in performance is AI infrastructure such as data centers, and its key customers are the leading cloud manufacturers. In recent years, cloud manufacturers have invested a lot of money in AI infrastructure, and their capital expenditure has also become an important indicator of chip manufacturers' growth expectations.
According to the data estimated by various institutions, the head cloud manufacturers' spending on AI continues to increase this year. But the competition on the chip field is more fierce. Even Nvidia, the "strongest on the surface", is making more preparations to stabilize its position.
Counterpoint data shows that the capital expenditure of global cloud manufacturers will increase by 4% in 2023, but it is expected to surge by 42% in 2024, mainly driven by the massive investment of American cloud manufacturers in AI infrastructure.
According to Morgan Stanley's estimated data, the capital expenditure of global cloud computing in 2030 will reach 300 billion dollars, of which AI semiconductor related capital expenditure will be 230 billion dollars, and AI hardware 70 billion dollars.
Accordingly, super customers will be more active in AI server procurement in 2024. These super giants include Microsoft, Amazon, Google, Meta, Tesla, Byte Beat, Tencent, Alibaba, Baidu, etc. From the perspective of capital expenditure ratio, Microsoft has reduced its share, while Tesla has significantly increased.
Specifically, in terms of Nvidia products, Morgan Stanley reports that in 2024, Microsoft will account for 23% of Nvidia's demand for HGX/DGX servers, Coreweave will account for 11%, Dell will account for 10%, Amazon MataOracle will account for 9%, and Google will account for 8%.
There is no doubt that Nvidia's products are still popular in the world, but that does not mean Nvidia can rest easy. Intel and AMD are all eyeing for share, and a number of Internet manufacturers are also vigorously developing AI chips to complement each other.
For example, Intel recently released the latest generation of 5nm AI chip Gaudi3, which is expected to be shipped to OEM manufacturers in the second quarter of 2024, mainly targeting Nvidia H100 GPU. Gaudi3 especially emphasizes its higher energy efficiency and lower cost; AMD launched MI300 series products last year. MI300X chip has more than 150 billion transistors, and the benchmark is Nvidia H100, which has certain advantages in reasoning workload. Microsoft also said at the recent conference that it has adopted MI300X chip.
At present, the AI chip market is in a stage of rapid development and fierce competition. In this battlefield without gunpowder smoke, the strong players seek to occupy a favorable position in various core fields.
It should be pointed out that a more intelligent data center is naturally a must, but the competition of giants is not limited to this. AI PC is also driving new demand. New forces are surging in the PC market. Intel and AMD are sticking to the "Windows+x86" position, while Nvidia, MediaTek and Qualcomm have attacked the PC hinterland, and "Windows+Arm" has joined the battle.
Morgan Stanley's report pointed out that Nvidia is launching a highly integrated PC chip. By combining its game graphics technology with MediaTek's Arm Architecture SoC, it may be released at CES next year, making the game notebook thinner and with AI functions.
The report also said: "TSMC's CoWoS technology and Nvidia's NVLink interface are key technologies that combine Nvidia and MediaTek into one chip. Our PC supply chain research shows that Qualcomm will start shipping WoA PC chips at the beginning of 2024, with an initial shipment of about 2 million pieces. Later, the PC chips jointly designed by Nvidia and MediaTek will be released in 2025, with an annual shipment of 12 million pieces. "
From servers, copper cables to glass substrates, it continues to pull the industrial chain
Since the launch of Nvidia Blackwell products, it has further driven the growth space of upstream and downstream industrial chains. From storage, packaging technology, communication connectivity technology, to server assembly, computing power leasing, the capital market has also continued to set off a boom.
Globally, semiconductor related suppliers of GB200 super chips include SK Hynix, which produces HBM, TSMC, ASM Pacific, which manufactures packaging equipment, KYEC, BMC chip manufacturer ASpeed, etc; Suppliers related to AI server hardware include Foxconn, Wistron, Delta which provides power and cooling solutions, transceiver manufacturer Innolight, guideway manufacturer KingSlide, etc.
Looking back at the domestic market, on the domestic computing industry chain, Industrial Fulian, Inspur Information, Lixun Precision, optical module supplier Zhongji Xuchuang, and computing leasing company Hongbo are all concerned. From the stock price perspective, as of the press release, the share price of Industrial Fulian has risen by 65% since the beginning of the year, Inspur Information has risen by 18%, and Zhongji Xuchuang has risen by 51.47%. However, Lixun Precision has fallen by 5%, and Hongbo shares have fallen by 36%.
Specifically, in the field of servers, Zheng Hongmeng, chairman of Fulian, said at the recent performance conference that in 2024, AI is expected to contribute 40% of the company's total cloud computing revenue, and AI servers will account for 40% of the global market share. Peng Zhen, chairman of Inspur Information, also said that the company has a considerable number of orders in hand at present and is expected to maintain growth in the second quarter.
From the institutional data, the sales of servers will increase this year. TrendForce Jibang Consulting estimates that in 2024, the global server shipment will increase by 2.05% annually, and AI servers will account for about 12.1%. At the same time, the market still focuses on deploying AI servers, which account for about 12.1% of shipments.
Previously, the concept of copper high-speed connection became popular. Nvidia's new connection mode of copper backplane led the market to pay attention to the concept of "copper high-speed connection", and many communication connection industry chain companies' share prices climbed as a result.
In terms of suppliers, although Amphenol, Molex and Tyco are the leading enterprises in the overseas market, accounting for 60% to 80% of the backplane connector market, domestic enterprises are also developing rapidly, and enterprises such as Lixun Precision, Huafeng Technology and Zhaolong Interconnection are also accelerating their research and development.
Shu Di, chief electronic analyst of Guotai Jun'an Research Institute, pointed out to the reporter of 21st Century Business Herald that the GPU in Nvidia's new chip uses high-speed copper cable to realize electrical connection, which can further reduce losses, reduce costs and improve heat dissipation efficiency. From the perspective of supply chain, for example, Lucent Precision currently has high-speed copper cable products such as DAC and ACC. In combination with its optical connection (layout of high-speed large bandwidth and silicon optical high-end products), thermal management and other product layout, there are complete solutions for high-speed interconnection between cabinets and inside cabinets.
Recently, glass substrate is becoming a new hotspot. It is pointed out that the advanced packaging process adopted by Nvidia GB200 will use glass substrate. In addition, Intel, Samsung, AMD, Apple and other major manufacturers all said that they will import or explore glass substrate chip packaging technology.
A few days ago, Zhou Hua, chief analyst of CINNO Research, said that Intel would announce the introduction of glass substrates for advanced packaging in 2023 and planned to supply them to the market in the next few years. At present, the domestic representative enterprises in the field of organic substrates include Xingsen Technology, Shennan Circuit, Zhuhai Yueya, etc. They may provide solutions for glass substrates in the future.
Morgan Stanley pointed out in the report that the glass substrate is very hard and flat, and also has the advantages of reducing power loss, high temperature resistance, moisture resistance, etc. It is expected that the glass substrate will be used for advanced packaging in the next two years. With the establishment and adoption of glass substrate manufacturing process becoming more clear, although the cost is higher, it is expected that chip manufacturers will use glass substrates for GPU, CPU, DPU and HBM.
As Nvidia continues to iterate, changes in the industrial chain will continue to update.
