Compiled by Zhidongxi, Edited by Chen Junda, Cheng Qian
On May 24th, Zhidongxi reported that according to The Wall Street Journal, Nvidia achieved a 262% year-on-year revenue growth in the first quarter of the 2025 fiscal year, and its stock price surpassed $1000 per share for the first time after the financial report was released. However, there are emerging threats that could weaken Nvidia's dominant position in the chip market.
Competitors like AMD and Intel are developing chips that could potentially replace Nvidia products. Intel's Gaudi 3 boasts a 1.7 times advantage in training time compared to Nvidia's H100. AMD's MI300X chip has outperformed the H100 by 1.1 times when running the Llama 2-70B model. Microsoft, one of Nvidia's major clients, recently announced that it would offer AMD chips to cloud computing customers as an alternative to Nvidia chips.
The AI market itself is undergoing subtle changes. With the deployment of AI models, the demand for AI inference is growing steadily. However, as this demand for computational power is not as high as for training, Nvidia's chips are not essential in this field, which could impact Nvidia's market share.
To address these challenges, Nvidia is accelerating the development of next-generation products, shortening the product iteration cycle to one year. Nvidia's financial report for this quarter also mentioned that Nvidia has achieved rapid development in new areas such as automotive, consumer internet, and sovereign AI.
Nvidia's sales in the first quarter of the 2025 fiscal year nearly tripled, with expectations of doubling in the next quarter. This has driven Nvidia's stock price to reach new highs, rising 9.3% to $1037.99 per share at the close of trading on Thursday. Nvidia announced a 1-for-10 stock split plan and doubled its dividend, a move taken by listed companies to enhance liquidity when stock prices are high.
Despite Nvidia's current momentum, investors are facing a significant question of whether Nvidia can sustain its growth in the long term. The high demand leading to shortages and the high prices of Nvidia chips have prompted AI companies of all sizes to seek alternative solutions.
Nvidia's competitors have seized this opportunity. AMD introduced the MI300X chip last year, claiming to outperform Nvidia's H100 chip in AI core performance. When running the Llama 2-70B model, the MI300X chip's mid-core performance is 1.2 times that of the H100, and the large core performance is 1.1 times that of the H100. AMD's CEO Lisa Su revealed last month that the company expects to generate around $4 billion in revenue from AI chips this year. Microsoft recently announced that it would provide AMD's AI chips to cloud computing customers as an alternative to Nvidia chips.
Intel launched its new generation AI chip, Gaudi 3, in April this year. Compared to Nvidia's H100, Intel's Gaudi 3 boasts a 1.7 times advantage in training time. Intel CEO Pat Gelsinger stated in a call with analysts that the company expects to generate $500 million in revenue from these chips in the second half of this year.
Major customers of Nvidia are also venturing into developing their own chips. Tech giants like Amazon, Google, Meta, and Microsoft are all designing their own AI chips, posing a new challenge to Nvidia.
Google has been collaborating with semiconductor company Broadcom for years on developing AI chips in-house. Google recently unveiled its sixth-generation Tensor Processing Unit (TPU) chip, Trillium, this month. Industry analysis firm TechInsights announced this week that Google is set to become the third-largest company in data center chip design by 2023, following Nvidia and Intel.
Broadcom CEO Hock Tan revealed in an internal speech this year that the company's custom chip division brings in over $1 billion in operating profit each quarter. The primary business of this division is manufacturing AI chips for Google, indicating Google's significant investment in developing its own chips.
Amazon introduced Graviton 4 and Trainium 2 AI chips in November last year, while Microsoft launched its Maia 100 AI chip in the same month.
Furthermore, apart from facing direct challenges from other chip players, Nvidia must adapt to the evolving AI market to maintain its leading position. In the early stages of this AI wave, enterprises focused on training AI models, requiring substantial computing power where Nvidia's chips excelled.
As AI models are deployed, the demand for AI inference continues to grow. Nvidia's CFO Colette Kress mentioned in a financial call this Wednesday that over the past year, more than 40% of the company's data center chip sales were for inference. However, unlike model training, inference does not demand as much from chip performance, indicating that other companies' AI chips can also handle inference tasks, making Nvidia's chips not indispensable for inference.
Some of NVIDIA's customers are also facing challenges. Despite the vast prospects of the AI market, startups still find it difficult to afford the high cost of NVIDIA chips. According to Redpoint Ventures at the AI Ascend Summit in March this year, the AI industry has spent $50 billion on purchasing NVIDIA chips for training large language models, but generative AI has only brought in $3 billion in revenue for the global AI industry.
This significant gap between investment and returns has put some AI startups developing products using NVIDIA AI chips in a difficult position. Mustafa Suleyman, co-founder of the large model startup Inflection AI supported by NVIDIA, joined Microsoft with a group of employees in March, while Emad Mostaque, CEO of Stability AI behind the well-known AI image generator Stable Diffusion, suddenly resigned in March as well. These executive changes in startups have brought uncertainty to the demand for NVIDIA chips.
The growth of the AI industry also faces broader risks, such as the shortage of energy for building and operating data centers that house these AI chips.
To address these challenges, NVIDIA is accelerating the development of next-generation products. NVIDIA CEO Jensen Huang revealed during the first quarter earnings call for the 2025 fiscal year that after the Blackwell architecture GPU, NVIDIA will introduce a new chip, with plans to update at a pace of one generation per year in the future. Additionally, he mentioned that NVIDIA will rapidly advance the development of all other types of products, including new CPUs, new GPUs, new network NICs, and new switches.
Huang vigorously promoted the advantages of NVIDIA's products during the earnings call. He stated that all NVIDIA products support CUDA, enabling the operation of the entire software stack. By choosing to invest in NVIDIA's architecture, customers can enhance innovation capabilities while reducing the total cost of ownership.
Analyst Hans Mosesman from Rosenblatt Securities wrote in a report that due to competitive pressures, it is expected that NVIDIA's share in the global AI chip market will decrease. However, he also pointed out that with NVIDIA's expansion into computing and software, its overall share in the AI computing sector may remain stable or even increase.
Conclusion: Challenges and Opportunities Coexist in NVIDIA's Future
NVIDIA has achieved commendable success in the market, with rapid profit growth and record-breaking market value serving as strong evidence. NVIDIA's products still offer optimal solutions in many application scenarios, with their latest products H200 and Blackwell architecture GPUs in high demand.
Yet, NVIDIA's future development path faces challenges. NVIDIA is taking various measures to ensure its competitiveness in the rapidly evolving AI market. During NVIDIA's financial conference call, it was mentioned that NVIDIA is experiencing rapid growth in sectors such as automotive, consumer internet, and sovereign AI, with automotive expected to become NVIDIA's largest enterprise vertical market within data centers this year.
Market expectations for NVIDIA have been raised to new heights, and whether NVIDIA can continue to outperform expectations as it has in the past few quarters remains uncertain.
Source: The Wall Street Journal
