Intelligent computing power is an important foundational resource for the development of the digital economy. Due to the US technology embargo and the domestic gap in artificial intelligence technology, China still needs to work hard on achieving full domestic industrialization of intelligent computing resources. In order to seek the release and efficient utilization of available computing power resources in physical space, the state has continuously promoted the 'East Data West Computing' project at the national level, and the related layout of the intelligent computing industry has been elevated to a strategic height for future national scientific and technological development.
Frost & Sullivan continues to monitor the development of the digital economy and hereby jointly with Tiangang ZhiCAL officially releases the '2024 China Intelligent Computing Power Industry White Paper'. The report aims to analyze the current situation, application prospects, technological trends, and development trends of the Chinese intelligent computing power market, identify market competition dynamics, and reflect the differentiated competitive advantages of the leading enterprises in this segment market. Frost & Sullivan looks forward to strengthening the understanding of the overall market landscape among industry participants through the organization and analysis of the intelligent computing power market.
This report provides an in-depth analysis of the intelligent computing industry, extending its research perspective to the supply side of intelligent computing power. It dissectes the ways market participants of different types enter the intelligent computing arena, their business models, differentiated strengths and weaknesses, forming a comprehensive market insight.
1
AI Era · Overview of Artificial Intelligence Development
In 2023, the global digital economy reached a scale of $45 trillion, accounting for 44% of global GDP, and has become a powerful engine for global economic development. Benefiting from national strategies that strongly promote the digital transformation of industries, China's digital economy has also grown rapidly, with a growth rate significantly exceeding that of the United States and Germany. As a technology driver and innovation accelerator in the digital economy ecosystem, the artificial intelligence industry has received high attention from global enterprises and sustained technological and financial investment.
01
Global: The wave of artificial intelligence sweeps across the globe, with countries actively deploying and accelerating its implementation.
In 2023, the total investment in the global artificial intelligence market reached $1835 billion, and it is expected to exceed $6000 billion by 2028. The annual compound growth rate of the global artificial intelligence market size from 2023 to 2028 will reach 28%. As of the third quarter of 2023, there were 29,542 artificial intelligence companies globally, with the number of companies concentrated in the United States and China accounting for 49% of the total global number. The distribution of artificial intelligence companies by country shows a pattern dominated by China and the United States.
Artificial intelligence can be divided into generative AI and discriminative AI according to model categories. Generative AI, based on technologies such as large models and deep learning, shows application potential in various fields such as virtual reality and content creation, triggering a new round of artificial intelligence revolution. The market scale has seen rapid development since 2020. Global leading large model enterprises continue to lead and promote the innovation and implementation of generative AI technology. Commercial applications are gradually advancing from text to fields such as images, audio, and video, thus generating a massive demand for high-performance intelligent computing power. To meet the development needs of the large model industry, major countries around the world have successively introduced supportive policies, planning and deploying high-performance intelligent computing resources in multiple aspects such as funding, infrastructure construction, data supply, talent, and downstream applications.
As the cornerstone of intelligent computing power, the market size for AI chips will see large-scale expansion. It is estimated that by 2028, the global AI chip market size will exceed $160 billion. AI servers, as the carriers of intelligent computing power, are experiencing both an increase in market size and shipments. It is predicted that by 2028, the global AI server market size will exceed $100 billion, with shipments reaching 3.43 million units.
02
China: Technological breakthroughs accelerate the development of artificial intelligence, becoming a new driving force for the digital economy
Currently, there is active investment in artificial intelligence technology in China, with investments exceeding $19.3 billion in 2023, a year-on-year increase of 48.2%, accounting for 10.5% of the global total investment. At the same time, the core artificial intelligence industry system is gradually becoming complete. In 2023, the scale of the core artificial intelligence industry reached 578.4 billion yuan, a year-on-year increase of 13.9%. The number of enterprises has exceeded 4,300, and downstream application scenarios cover multiple fields such as life services, smart healthcare, smart manufacturing, smart vehicles, and logistics warehousing.
The Chinese generative artificial intelligence market is driven by both continuous investment in upstream technology and the ongoing implementation of commercial applications downstream. It has shown strong growth potential and market vitality. Although the domestic large model industry started later than Western companies, with rapid technological iteration and commercial implementation in the market, the industry began to enter a development explosion phase in 2023. Currently, the level of large models of leading domestic enterprises has caught up with the international average. In 2023, the market scale of Chinese large models reached about 132 billion yuan, and it is expected that the market scale will reach 817 billion yuan by 2028.
2
Industrial upgrading · The rise of intelligent computing power in China
Computing power, algorithms, and data are the three major elements in the development of artificial intelligence. Artificial intelligence computing power (referred to as 'intelligent computing power' or 'AI computing') refers to the ability to process large amounts of data, models, and other computational tasks used in AI training and inference. Currently, the improvement of intelligent computing power has become a key factor driving innovation in AI applications and industrial upgrading, as well as crucial for achieving rapid iteration and optimization of large models.
01
Panoramic Analysis of the Industrial Chain: From Hardware Foundation to In-depth Development of Application Scenarios
Artificial intelligence chips, as the foundational hardware of the intelligent computing power industry, provide the necessary computing capabilities for intelligent algorithms and applications.Due to the tightening of US semiconductor regulation towards China and the global economic slowdown, the market scale of China's chip industry has shown a downward trend since 2021. Thanks to the high attention from the state, the artificial intelligence chip market has risen rapidly against the backdrop of adversity. In 2023, the market scale of AI chips in China reached 652 billion yuan. With the continuous development of autonomous and controllable artificial intelligence chip technology and the sustained increase in demand for AI chips, there is great potential for future market growth.
Artificial intelligence servers equipped with AI chips provide necessary computing power support for the intelligent computing industry, enabling efficient completion of tasks such as large-scale data processing, model training, and inference computing.The Chinese artificial intelligence server market has shown strong performance driven by the rapid development of the intelligent computing industry, with both market size and shipments growing in tandem. As the maturity of artificial intelligence applications improves, the market's demand for artificial intelligence servers is expected to continue to rise.
There are many types of mid-tier enterprises in intelligent computing power, which can be roughly divided into seven categories: ICT hardware integrators, telecom operators, third-party data center service providers, cloud service vendors, AI enterprises, AI Infra vendors, and cross-border enterprises.By integrating the infrastructure and core hardware devices of upstream intelligent computing power, various midstream enterprises can leverage their differentiated advantages in supply chain, channels, products, platforms, and technology to deliver high-performance intelligent computing resources to downstream demanders. For example: Data centers can build and operate physical platforms for intelligent computing resource providers, offering massive data storage, high-speed network transmission, and real-time processing services; cloud service providers can provide computing resources and algorithm value-added services with their comprehensive technical capabilities and complete product matrix in cloud computing.
At the downstream level, artificial intelligence technology has gradually penetrated into every corner of society, and has been widely applied in fields such as the internet, service industry, government departments, finance industry, and industrial manufacturing.With the continuous maturity and popularization of artificial intelligence technology, AI acceleration servers are expected to play a key role in more industries. At the same time, as the penetration of large models in vertical industries gradually deepens and commercialization continues to advance, the demand for intelligent computing resources in various industries will gradually emerge, boosting the digital transformation and intelligent upgrading process of society.
Since the intelligent computing power industry is a technology-intensive sector involving numerous key technologies such as computing power perception, scheduling, and operation, and the domestic intelligent computing market is currently in the early stages of pilot development, it is necessary for participants across the entire industry chain to leverage their respective advantages and achieve collaborative construction.Investors, builders, operators, model developers, and other stakeholders of intelligent computing need to explore different profit models based on their respective core resources, achieving a healthy cycle of cash flow..
02
Analysis of intelligent computing power: driving factors, trend prospects, and competitive barriers
Driving factor: The government has introduced policies to accelerate the layout of the intelligent computing industry, which is an important foundation for its development.In recent years, the state has introduced a series of policy documents to actively develop the computing power industry ecosystem, strengthen the connection between supply and demand of computing power, and promote regional coordinated development of the computing power industry. Local governments at all levels have responded to the national call, and policies covering multiple dimensions including infrastructure construction, regional coordinated development, talent cultivation, and technological innovation have been guided and encouraged;The vigorous development of artificial intelligence drives the demand for intelligent computing, which is the core driving force for the development of the intelligent computing industry.Currently, artificial intelligence is accelerating its penetration into various production processes across industries and fields. The application of AI in emerging areas is constantly emerging, and the transformation and upgrading of demand-side industries are driving the development of intelligent computing;The iteration of computing power technology has further driven the rapid growth of the intelligent computing industry.The continuous development of artificial intelligence has led to exponential growth in the amount of data used by large models and the scale of their parameters. The demand on the supply side for large-scale data processing capabilities and the ability to perform complex computations is becoming increasingly stringent, requiring the long-term development of the intelligent computing industry.
Development Trend: In the future, the regional collaboration effect of intelligent computing centers will become more pronounced.Cross-regional collaboration of intelligent computing centers is increasing, distributed training demand is gradually rising, and the integration of large models with business scenarios is deepening. With terminal device performance improving, 'cloud-edge-terminal' collaborative applications will become mainstream.In addition, the ubiquitous and inclusive AI computing service is under developmentThe mode of platform-scheduled computing power is becoming increasingly important, and the computing power service model is gradually shifting from resource-based to task-based. The threshold for using computing power and the operational difficulty of intelligent computing centers will gradually decrease, while the utilization efficiency of computing power will gradually improve;The green and low-carbon transformation of intelligent computing is also a key focus for future development.The green development of the entire link of computing power production, operation, management, and application is an important direction for industrial development. It requires intelligent computing centers to actively apply green energy and further achieve deep integration of green computing power with key carbon-emitting industries, realizing the dual sustainable development of the environment and business.
Competitive Barriers: As a technology-intensive industry, the intelligent computing sector faces high barriers, mainly reflected in talent resources, capital, value-added service technology, and brand effect barriers.Smart computing enterprises not only require talents in data processing, industry knowledge mining, large model construction, and optimization but also face high upfront investment and ongoing operational costs. This necessitates the adaptation, management, and scheduling of computing resources. As a result, market resources and opportunities are gradually concentrating on a few leading companies with strong capital strength. At the same time, after sufficient technical investment and iteration, the brand effect of leading manufacturers becomes increasingly evident, giving them a wide user base. Moreover, users face higher costs for data migration and system reconstruction when switching platforms, further enhancing user stickiness. Under the influence of multiple factors, new entrants often find themselves at a disadvantage in competition.
In addition, the Chinese intelligent computing industry has shown clear pain points in its development. The supply side incurs high costs, while the demand side exhibits differentiated characteristics in scheduling computing power. The mismatch between supply and demand, coupled with technical limitations, hinder flexible allocation of intelligent computing resources. The market urgently needs to explore a more diversified computing power supply pattern to solve various problems emerging in the intelligent computing ecosystem.
3
Demand Migration · AI Computing Leasing Opens a New Chapter
Generative artificial intelligence is developing rapidly, requiring large models to process longer data sequences and complete more complex tasks. The computational power required by large models is closely related to algorithm structure, parameter quantity, data volume, and the number of training rounds. When the model size exceeds a certain threshold, its performance will significantly improve. The total computational power consumption from training to fine-tuning and then to inference has become a key to the functionality of large models, and model training on large-scale intelligent computing clusters has become an inevitable choice for model optimization.
01
The Rise of Intelligent Computing Leasing: Penetrating the Market Logic Behind Supply-Demand Contradictions
At present, the training, fine-tuning, and inference of large models mainly rely on high-performance GPUs represented by NVIDIA, which provide powerful parallel computing capabilities, large memory capacity, and mature software ecosystem support. In 2023, the chip shipments in the Chinese market accelerated to 1.4 million units, with NVIDIA accounting for 85% of the market share with 1.19 million units.In recent years, the demand for computing power has shown a leapfrog growth. Despite the continuous increase in NVIDIA's shipments, market demand has not been fully met. The imbalance between supply and demand in the market has led to an increase in GPU prices.
In addition, the US government has successively introduced a number of technology embargo policies targeting China since 2018, making the supply of high-performance GPUs outside Shanghai in the Chinese market even more scarce. Domestic GPUs have not yet achieved complete substitution in terms of technical level and practical application in a short period.Therefore, the scarcity of high-performance intelligent computing resources has further increased.
According to Frost & Sullivan's calculations, based on the training cost of ChatGPT 3.5 as a benchmark, the demand for A100 GPUs in China's large model training sector in 2024 will reach 26,380 units, while the demand for A100 GPUs in the inference sector will reach 3,298 units. Under the pattern of mismatch between supply and demand for high-performance GPUs, how to allocate intelligent computing resources and optimize GPU usage efficiency has become a key issue.
Objective factors such as supply-demand mismatch have constrained the development speed of domestic model R&D enterprises to a certain extent. In addition, deployment costs, operational maintenance costs, time costs, the flexibility of computing resources, and data security have also put forward more requirements for the allocation of intelligent computing power.Against this backdrop, compared to self-built intelligent computing power, AI computing rental offers significant advantages in terms of capital cost and operational flexibility. It is more friendly to model R&D enterprises other than large internet companies, significantly lowering the threshold for enterprises to use high-performance AI resources. It has become a new choice for domestic large model manufacturers as a source of intelligent computing power..
02
The new trend in intelligent computing leasing: How the leasing model reshapes the industry landscape
Currently, China's intelligent computing rental market is in its early stages of development. There are many types of participants, each with differentiated advantages. The service forms of intelligent computing rentals vary, and the business models are also in the process of exploration.The mainstream intelligent computing rental models in the market can be broadly categorized into four types: one-stop solutions from cloud service providers, GPU compute pool rental, GPU compute pool scheduling, and the embedded hardware training-on-demand integrated machine model.Different intelligent computing leasing business models have their own advantages and disadvantages, which can be targeted at differentiated downstream customer needs. For example, cloud service providers can leverage their comprehensive technical capabilities and complete product matrix in cloud computing, and with their underlying high-performance intelligent computing resources, offer value-added services to provide one-stop intelligent computing solutions to downstream customers. Meanwhile, GPU computing resource pool scheduling platforms provide services such as computing resource discovery, supply and demand matching, transaction purchase, and scheduling usage, making computing resource delivery more flexible.
Different types of participants in the upstream, midstream, and downstream of the intelligent computing power industry chain are entering the market. Relying on their existing businesses, they are building unique product matrices. According to different business sets, they are exploring various business models and demonstrating differentiated advantages in the intelligent computing leasing business.For example: IDC enterprises, due to their comprehensive data center construction capabilities, large-scale hardware cluster deployment, and supporting operations and maintenance capabilities, can build a product matrix mainly consisting of IDC and infrastructure construction services, intelligent computing center operations and maintenance services, cloud services, etc. By collaborating with ICT hardware integrators, telecom operators, or internet giants to jointly build intelligent computing cluster projects, they undertake operations and maintenance services as well as the rental of intelligent computing resources. AI Infra vendors can support model training and inference, server deployment and operation based on their outstanding AI and intelligent computing technology capabilities, providing products mainly consisting of software and platforms such as inference engines, training acceleration, and intelligent computing leasing. Some cross-border enterprises that are engaged in other industries have also taken the intelligent computing industry as their second growth curve, deploying resources such as GPU chips based on their intelligent computing resource channels and sufficient capital reserves, and investing in the intelligent computing industry by renting out their own intelligent computing hardware equipment. The development of differentiated business models and the entry of diversified market participants are making the industry matrix of computing power leasing more three-dimensional.
