Technology cloud report original.
The dispute over the large model has gradually receded after the major entrants announced their products, but the high-frequency resonance of the industrial chain brought about by the large model has passed to the computing power layer.
The most intense performance is the AI server market. The demand for computing power brought by the large model directly triggered a wave of panic buying and price increases for AI servers.
According to the "Securities Times" report, a testing company revealed that the price of the eight AI servers they purchased in June last year had increased to 1.3 million each by March this year, and now the price has soared to 1.6 million each. In less than a year, the price has increased by nearly 20 times.
In addition, the surge in demand for AI servers has directly triggered a rush to buy the upstream material PPO (polyphenylene ether, used as a reinforcement material for high-speed copper-clad laminates). With the heavy volume of servers, PPO is likely to become one of the scarce links in the industry chain in the future.
In this context, news of AI server manufacturers expanding production is also emerging.
Hongbai Technology, a subsidiary of Hon Hai Group responsible for the AI server business, has been exposed to the news that it plans to add five to six new production lines to meet the requirements of AI server customers.
The enthusiasm of the market can be seen, which directly ignited the capital market.
Since January, the AI server concept stocks led by Inspur Information, Zhongji InnoLight, and Fii have skyrocketed, with multiple daily limit, and even the long-term loss-making Cambrian, the stock price has been soaring all the way.
Explosive "AI server"
What is an AI server?
AI server is a high-performance server specially designed to perform computing-intensive tasks such as artificial intelligence (AI), machine learning (ML), and deep learning (DL).
AI servers are usually equipped with high-performance central processing units (CPUs), graphics processing units (GPUs), tensor processing units (TPUs) or dedicated AI accelerators, as well as large amounts of memory and storage space.
In the heterogeneous mode, the AI server can be CPU+GPU, CPU+FPGA, CPU+TPU, CPU+ASIC or CPU+various accelerator cards.
The specific design and configuration can be tailored to the specific task requiring massive parallel processing.
Currently, the most widely used AI server is CPU+GPU. This also distinguishes it from traditional servers.
It is reported that traditional servers mainly use CPU as the computing power provider, but a large number of branch jump interrupt processing needs to be introduced during the operation process, which makes the internal structure of the CPU complex and cannot meet the needs of the AI era.
The AI server using GPU parallel computing has thousands of cores per card, and is good at processing intensive computing applications, such as graphics rendering, computational vision and machine learning.
The basic configuration of the AI server used by the above-mentioned testing companies includes 8 Nvidia A100GPUs and 80G memory.
AI servers are very useful for computationally intensive tasks of AI, ML, and DL. Key features include:
Big data processing : AI servers are able to process and analyze large amounts of data, which is the key to training AI and ML models.
Parallel Computing : Since AI and ML algorithms require complex calculations on large amounts of data, AI servers typically use hardware that can process large amounts of data in parallel, such as GPUs.
Storage and memory : AI servers usually have a large amount of storage space and memory in order to store and process large amounts of data.
Network capability : AI servers require high-speed and low-latency network connections in order to quickly transfer large amounts of data.
In fact, this also explains why after the heat wave of large models, there will be a rush to buy AI servers. Large models contain massive data parameters, training, and running require more computing resources to process, which requires higher-performance AI servers to support.
Of course, the most direct reason for the surge in demand for AI services this time is the arrival of the era of large models, but in fact, the explosion of AI servers at this node is related to the development of AI technology and big data.
In general, the explosion of AI servers can be attributed to the following key factors.
First, the rise of big data. Every corner of modern society, whether it's social media, e-commerce or Internet searches, is generating massive amounts of data.
These data need to be analyzed and interpreted by complex algorithms to discover useful patterns and information, and AI servers can provide enough computing power to handle these tasks.
Second, the popularity of AI and ML is also driving the demand for AI servers. AI and ML are now widely used in a variety of industries, including healthcare, finance, retail, and transportation, among others.
Advances in these fields require massive computing power to process and analyze data, and to train and run complex AI and ML models.
Finally, the development of cloud computing and edge computing also provides impetus for the explosion of AI servers.
Cloud computing enables businesses and organizations to obtain powerful computing capabilities without purchasing and maintaining expensive hardware, while edge computing requires data processing and analysis on servers close to where the data is generated.
AI server domestic market pattern
The AI server market has continued to grow in the past few years, and now, with the blessing of large models, the AI server market is getting bigger and bigger.
According to the latest data released by Beijing Yanjing Bizhi Information Consulting, the global AI server industry market shipments will reach 850,000 units in 2022, a year-on-year increase of about 11%. 600,000 units, an increase of about 39% compared to the same period last year.
In the future, with the development of large AI models such as natural language processing and images and videos, the demand for computing power will continue to grow. It is estimated that by the end of this year, the global AI server market will exceed 20 billion US dollars.
By 2025, market shipments are expected to increase to around 1.9 million units, with an average annual growth rate of 41.2% during 2022-2025.
As far as the specific industry chain is concerned, the upstream of the AI server industry chain is the supply of core components such as CPU, GPU, memory and hard disk, as well as software such as database, operating system and basic management software; the downstream is the application market, including the Internet, cloud computing and data service center etc.
Currently, the market is dominated by some major AI server manufacturers, including Huawei, Inspur, Lenovo, and Sugon, whose servers are widely used in AI and ML research and commercial applications.
However, it is worth noting that Inspur Information recently released a semi-annual performance forecast with both revenue and net profit declining.
Among them, Inspur Information's non-net profit in the first half of 2023 fell by 88%-99% year-on-year. In this regard, Inspur Information stated that in the first half of 2023, due to factors such as the tight supply of global GPUs and related special-purpose chips, operating income will decline.
In fact, some industry insiders analyzed that in the context of the AI server fire, Inspur Information’s performance fell short of expectations. The deep-seated reason lies in the overall downturn in the traditional server industry, and the actual proportion of Inspur Information’s AI servers is not large.
Inspur Information previously stated that overall, the company’s AI server business accounted for an increasing proportion of its overall business. The performance brought about by the surge in demand for AI servers may only be reflected in the 2023 annual report of Inspur Information.
However, according to the "China Server Market Tracking Report Prelim for the Fourth Quarter of 2022" released by IDC, at present, Inspur's share in the entire server market (covering AI servers and traditional servers) is still leading with 28.1%, but compared to last year's 30.8%. share has declined.
In fact, this also shows that the traditional CPU server industry is affected by AI, and the market is gradually sluggish. In the future, heterogeneous servers represented by AI servers may be the general trend.
In the application market, the data shows that the four North American cloud providers represented by Microsoft, Google, Meta and AWS have a relatively high purchase volume in the global market.
At the end of 2022, Microsoft will occupy the first place in the year with nearly 20% of the purchase volume; Google, Meta and AWS purchase volume will be next, reaching 17%, 15% and 14% respectively.
In China, with the entry of technology manufacturers into large-scale models and the rise of large-scale entrepreneurship, and the accelerated construction of AI computing power infrastructure, the proportion of AI server purchases has also increased accordingly.
By the end of 2022, ByteDance's AI server purchases will increase significantly, accounting for 6% of the market.
However, the market also faces some challenges. First, there is the issue of energy consumption. Although the performance of AI servers is constantly improving, their energy consumption is also increasing. This is a problem both for the environment and for power supply.
Second, the rapid development and changes of AI and ML require server manufacturers to continuously invest in research and development to ensure that their products can meet the latest needs.
Regarding the future, the domestic AI server market has great potential. With the further development and application of AI, ML, and DL, the demand for AI servers is expected to continue to grow.
In addition, with the popularization of 5G and Internet of Things technologies, the demand for AI servers in the field of edge computing will increase in the future.
In general, although the market is facing some challenges, the rapid development and wide application of AI servers shows that this is a market full of vitality and potential.
[About Science and Technology Cloud Report]
Focus on original enterprise-level content experts - technology cloud reports. Founded in 2015, it is the top 10 media in the cutting-edge enterprise IT field. Recognized by the Ministry of Industry and Information Technology, Trusted Cloud, one of the official media designated by the Global Cloud Computing Conference. In-depth original reports on cloud computing, big data, artificial intelligence, blockchain and other fields.