#来点儿干货#
With the rapid development of artificial intelligence (AI) technology, the demand for computing power has shown an explosive growth trend. As a computing infrastructure, the number and technical level of AI data centers are expected to increase significantly. From 2017 to 2022, the data center market size will grow from 35.4 billion yuan to 172 billion yuan, and IDC expects it to grow to 500 billion yuan in 2027, with a compound growth rate of 24% from 2023 to 2027.
Pay attention to Leqing's industry observation and gain insight into the industrial pattern!
Overview of the AI data center industry
As a facility designed to support AI computing and data processing tasks, the AI data center is specially configured and optimized on the basis of the traditional data center. This optimization is designed to meet the specific requirements of the computing environment for machine Xi, deep Xi, and other AI-related applications.
AI data centers are not only equipped with a large number of high-performance servers, GPU accelerators, and dedicated storage systems, but also strive to provide powerful computing power to accelerate the training and inference process of deep learning Xi.
The infrastructure of an AI data center includes key components such as servers, storage devices, network equipment, cooling systems, and power supply equipment. These components work together to build an efficient and reliable computing environment, which provides solid support for the operation and data processing of AI algorithms.
AI data center industry chain
The industry chain of AI data centers can be divided into:
- Infrastructure: This includes hardware devices that provide computing power, such as servers, GPU accelerators, etc., as well as storage and networking devices. These devices are the foundation for the operation of AI data centers, where they store and process massive amounts of data and perform complex computing tasks. Infrastructure providers include hardware manufacturers, cloud service providers, and more.
- Software and algorithms: On top of the hardware infrastructure, AI data centers need to run a variety of software and algorithms, including machine Xi frameworks, deep Xi algorithms, etc. These software and algorithms are the key to enabling intelligent computing in AI data centers. Software and algorithm providers include open source communities, commercial software companies, and AI algorithm R&D institutions.
- Platforms and services: AI data centers also need to provide a range of platforms and services to support the development and deployment of AI applications. These platforms and services include AI development platforms, AI training platforms, AI inference platforms, model management services, and more. Providers of platforms and services include cloud service providers, AI platform companies, system integrators, and more.
- Applications and solutions: The value of an AI data center is reflected in the variety of AI applications and solutions it can support. These applications and solutions cover various industries and fields, such as intelligent customer service, autonomous driving, intelligent healthcare, etc. Application and solution providers include AI application developers and solution providers from a wide range of industries.
In the complete industry chain of AI data centers, each link is interdependent and mutually reinforcing, forming a close ecosystem.
In the upstream of AI data centers, chip manufacturers such as HiSilicon, Cambrian, and SMIC provide high-performance chip solutions to lay a solid foundation for the computing power of data centers. In terms of network equipment, the critical network infrastructure provided by Huawei, Cisco, and other companies ensures stable and efficient network connections in data centers.
In the midstream of the industry chain, telecom operators, cloud service providers and large Internet enterprises are the main players. Internet companies such as Amazon, Baidu, and Tencent have deployed their own AI data centers to support their huge AI computing needs. Cloud service providers such as Alibaba Cloud, Tencent Cloud, and HUAWEI CLOUD provide a full range of services such as cloud computing, storage, and AI platforms to help enterprises and research institutions rapidly apply AI technologies. Telecom operators such as China Telecom, China Mobile, and China Unicom provide a wide range of AI-related services.
The downstream user groups of AI data centers are extremely wide, including enterprises, research institutions, and government departments from all walks of life. These users choose suitable cloud services or build private data centers according to their business and development needs to promote the application and development of AI technology in their respective fields.
Diagram of the AI data center industry chain:
In addition, the traffic pattern of AI data centers is different from that of traditional cloud computing.
Compared with the traditional cloud computing traffic model, the traffic pattern of the AI data center is significantly different. Each task in cloud computing is usually asynchronous and has a small amount of traffic, so the overall traffic is relatively balanced. However, the data traffic in AI data centers is characterized by large traffic in a short period of time, which will lead to the network delay and training speed of neural network training under the traditional cloud computing network architecture.
server
As the core component of a data center, servers are responsible for performing complex computing tasks and running AI algorithms.
The size and number of servers directly determine the computing power of the data center. Larger server clusters mean more processing units and computing resources, which can provide more computing power for AI applications.
AI servers play a vital role in many fields such as computer vision, natural language processing, and machine Xi learning, and they are widely used in various application scenarios such as image recognition, speech recognition, text analysis, and model training.
In terms of the architecture of AI servers, it has become a trend to adopt the combination of CPU+ acceleration chips. This architecture shows higher efficiency advantages in model training and inference, which provides strong support for the rapid development of AI applications.
In terms of cooling technology, cold plate liquid cooling technology occupies a mainstream position in the liquid cooling technology route due to its high maturity, assuming that its current market share reaches 80%. With the continuous maturity of immersion liquid cooling technology, its market share is expected to gradually increase.
According to comprehensive estimates, the training and inference of AI large models will bring up to 4 billion yuan of liquid cooling market space. With the increasing number of model parameters and the promotion of applications, the liquid cooling market is expected to achieve a compound annual growth rate of more than 60% in the next four years.
With the continuous development of AIGC and cloud computing technologies and the further expansion of application models, the demand and application scenarios of supercomputing continue to grow. The combination of supercomputing and cloud computing has brought new growth points to the server market, and the high-end servers required by supercomputing are expected to further drive the growth of the entire server market.
In terms of the AI server market structure, according to IDC statistics, in the first half of 2023, from the perspective of sales, Inspur, Xinhua III, and Ningchang ranked the top three, occupying more than 70% of the market share, and from the perspective of server shipments, Inspur, Kunqian, and Ningchang ranked among the top three, accounting for nearly 60% of the market share. In addition, manufacturers such as Sugon, China Great Wall, Industrial Fortune Union, and Tuowei Information are also actively deploying in the field of AI servers.
Optical transceivers
The need is getting more urgent. In this context, the demand for 800G optical modules for intra-cluster communication has become particularly prominent.
Optical modules are located at the core of the optical communication industry chain, covering key raw material suppliers such as optical communication devices (including chips) and optical fiber and cables in the upstream, and docking with optical equipment manufacturers in the downstream. Optical module manufacturers purchase upstream optical chips, electrical chips, and optical components, etc., and go through a sophisticated integration, packaging, and testing process to finally provide equipment integrators with optical communication equipment that meets specific needs, which are widely used in the telecom and data center markets.
With the rapid development of the field of artificial intelligence, especially the sharp increase in the amount of model training, the construction of high-performance AI server clusters Although 1.6T optical modules have not yet entered the mass production stage, due to the harsh requirements of the next generation of high-performance computing equipment for communication bandwidth, cloud vendors have shifted their procurement focus to 800G optical modules, which makes 800G optical modules occupy a higher position in terms of procurement priority and duration compared with 400G and 100G optical modules. #Artificial Intelligence##Data Center##光模块#
Optical module industry chain and some representative manufacturers combing:
Source: Minsheng Securities
epilogue
As a key part of artificial intelligence, AI data centers are responsible for processing massive amounts of data, supporting complex algorithms, and promoting model training. With the in-depth development and wide application of artificial intelligence technology, AI data centers play an irreplaceable role in providing computing power, storage capacity, and network connectivity.
The AI data center provides powerful computing power. The training and Xi of AI requires a lot of computing resources, especially as models become more complex and the amount of data increases dramatically. AI data centers are equipped with high-performance computer clusters, including a large number of CPUs, GPUs, and dedicated accelerators, capable of processing complex mathematical operations and machine-learning Xi algorithms, thereby accelerating the training and inference process of models.
AI data centers provide large-scale storage capacity. AI applications involve a huge amount of data and need to be supported by secure, reliable, and efficient storage systems. The AI data center uses distributed storage technology to store data on multiple nodes to improve data reliability and access speed. At the same time, the data center also adopts data compression, deduplication and other technologies, which effectively reduces storage costs.
As a key link in artificial intelligence, the AI data center provides powerful computing power, storage capacity, and network connection support for the development and application of artificial intelligence technology. In the future, with the continuous development of artificial intelligence technology and the continuous expansion of application scenarios, the importance of AI data centers will be further highlighted.
Pay attention to Leqing's industry observation and gain insight into the industrial pattern!