What signal does the collective strength of the computing power sector reveal?

Market Cap Observation

2024-06-03 15:13Published in Shaanxi financial field creators

Author: Taylor, Editor: Xiaoichi Mei

According to China Fund News, on June 3, affected by the news that Nvidia's Blackwell chip began to be put into production, the A-share computing power hardware side collectively strengthened, and the concept of memory chips fluctuated and rose. The direction of copper high-speed connectors continued to be strong, Deren Electronics, Shenyu shares, Zhongfu Circuit, etc. rose by more than 10%; The CPO sector fluctuated and strengthened, and Xinyisheng rose nearly 10% to a record high.

According to agency statistics, in 2021, the scale of the computing power industry represented by computers in the mainland has reached 2.6 trillion yuan, directly and indirectly driving the total economic output value of 2.2 trillion yuan and 8.2 trillion yuan respectively.

CPUs have led mankind into the PC era, mobile chips have set off a wave of mobile Internet, and AI chips have broken the decades-long computing bottleneck of the AI industry.

Nowadays, the computing power economy has been born. The arms race against AI computing power has already begun.

There is no shortage of elite soldiers in the A-share market. It can be divided into two main levels.

First, the AI supercomputing center.

The AI supercomputing center can be regarded as the core infrastructure of AI. Sugon is one of the best.

As early as 1993, Sugon broke through the international blockade and developed the high-performance computer "Sugon No. 1"; The self-developed Dawning 1000-Dawning 6000 has helped China's supercomputer win a number of firsts.

At present, the mainland is running three supercomputing prototypes with exascale (E-level) computing power, namely the Shenwei E-class prototype, the Tianhe E-class prototype and the Sugon E-class prototype, and behind them are the Jiangnan Institute of Computing Technology, the National Defense University of Science and Technology and the Sugon.

Some analysts believe that if the latest computing performance is ranked, the three major supercomputers in the mainland can rank among the top three in the world. And after the 2015 U.S. chip ban, these three computers are equipped with "Chinese chips".

Among the three supercomputers, Sugon is the only listed company and the company with the highest degree of commercialization, so it is not far behind to say that Sugon is the purest computing power stock in A-shares.

As leading Internet and technology companies launch large language models, the demand for computing resources in China will show a blowout. Computing power cloud leasing will become the optimal solution for computing power supply. Nvidia has launched related services, with monthly rental prices starting at $36,999.

At present, Sugon has deployed more than 50 cloud computing data centers across the country, mastering the largest computing resources in the country, and will officially launch the "computing power trading platform" in 2023, and third parties can access the computing resource pool. Moreover, the company has helped Baidu and Zidongtai to train large models, providing them with AI computing services and AI servers and other infrastructure.

Second, AI chips.

AI chips can be said to be the heart of computing power and the "spark" on the long march of AI.

Haiguang Information, which stands behind Sugon, is a "China Chip" national enterprise that provides kinetic energy for computing resources.

Sugon is the largest shareholder of Haiguang Information, backed by the Chinese Academy of Sciences, Haiguang has deep R&D strength, and at the same time, Sugon as a major shareholder and a major customer, Haiguang is not worried about sales.

Since 2021, the company's performance has shown explosive growth. According to the performance report, in 2022, Haiguang Information's revenue will be 5.125 billion yuan, a year-on-year increase of more than 1.2 times, and the net profit attributable to the parent company will be 802 million yuan, a year-on-year increase of 145.18%.

In the future, the company's CPU products will benefit from domestic substitution, and will continue to benefit from the procurement of government, finance and other enterprises; The DCU industry, due to the outbreak of the AI industry, will become a new growth point for the rapid rise.

In addition, Haiguang Information has AMD's x86 architecture technology license, and thus has become the only company in China that produces x86 chips. Based on the mature X86 system, compared with Loongson, Shenwei, Feiteng and other enterprises, Haiguang Information can reduce customer switching costs and have ecological advantages.

Haiguang Information is a leading domestic CPU + DCU (accelerated computing chip) company that is scarce in A-shares. Generally speaking, computing power can be divided into three parts, namely basic computing power, intelligent computing power, and supercomputing computing power, and what AI large models need is intelligent computing power.

As a product that focuses on general computing and simply provides artificial intelligence computing power, Haiguang Information has made some gains. At present, the only one in China that can support FP64 double-precision floating point arithmetic is Haiguang's DCU Deep Computing No. 1.

Compared with NVIDIA A100 products, Haiguang Shensuan No. 1 can basically play 70% of its performance level. According to the information of the investor interactive platform, Haiguang Shensuan No. 2 is in the research and development stage, and all work is progressing smoothly.

Another company with "hard power" is Cambrian, the first AI chip stock on the Science and Technology Innovation Board, and the company is one of the few companies in the world that has fully mastered AI chip technology.

The Cambrian product matrix involves cloud products, edge products, and IP licensing and software, with the first two benchmarking against NVIDIA and AMD, and IP licensing corresponding to ARM in the UK.

The cloud chip is the most needed source of kinetic energy for large language models, and in this field, the NVIDIA A100/H100 series occupies the top position of the pyramid.

As a catch-up, Cambrian launched the Siyuan series, and the Siyuan 290 and Siyuan 370 have entered many head customers such as Inspur, Lenovo, and Alibaba Cloud. Since the release of the Siyuan 220, the cumulative sales have also exceeded one million pieces.

In March 2022, Cambrian officially released the new training acceleration card "MLU370-X8", equipped with Siyuan 370, mainly for AI training tasks.

The latest Siyuan 590 chip is in rapid iteration, and the floating-point computing capability of the chip has been greatly improved compared with the 290, and the performance of the company's cloud products is expected to be further improved.

But Cambrian's biggest problem is that it continues to lose money, following a loss of 1.325 billion in 2022, it will lose another 800 million in the third quarter of 2023, which is also the core reason for the company's stock price to fall sharply before.

If its AI chips have not really achieved a performance explosion in the future, there will also be a risk of killing after the stock price soars.

As the "water seller" and infrastructure provider of the industry, the underlying computing power enterprises have undoubtedly benefited a lot. In this AI journey, the competition of China's computing power enterprises in Vientiane has just begun.

disclaimer

The content of this article related to listed companies is the author's personal analysis and judgment based on the information publicly disclosed by listed companies in accordance with their legal obligations (including but not limited to temporary announcements, periodic reports and official interactive platforms, etc.); The information or opinions contained herein do not constitute any investment or other business advice, and Market CapWatch disclaims any liability for any actions resulting from the adoption of this article.

——END——

View original image 64K

•