China's Artificial Intelligence Industry Outlook 2025

China's artificial intelligence (AI) industry chain has developed rapidly in recent years, forming a complete ecosystem in the fields of large model development , data annotation , chip computing power , data centers , autonomous driving and humanoid robots . The following will introduce the above six major fields from four aspects: major companies and market share, latest technological progress, industry chain relations and competitive landscape, financial status and investment opportunities.

Large model development

Major companies and market share

Since ChatGPT started to gain popularity at the end of 2022, a large number of large language model development companies have emerged in China. Baidu 's Wenxin Big Model ( Wenxin Yiyan ) is currently in a leading position and is known as the "one super" in the "China's big model one super, many strong pattern". According to a report from the China Internet Network Information Center (CNNIC), as of June 2024, Baidu Wenxin Yiyan's netizen usage rate in China's typical generative AI products reached 11.5%, ranking first ( CNNIC's latest report: Intelligent agents have become one of the mainstream forms of generative artificial intelligence applications_Sina Finance_Sina.com ). In addition, Alibaba 's Tongyi Qianwen , Tencent 's Hunyuan Big Model , iFlytek 's Spark Cognitive Big Model , and Huawei 's Pangu series are also influential general-purpose big model products in China ( Big companies crush or small companies counterattack? The data annotation and generation market has entered a white-hot competition|Artificial Intelligence_Sina Finance_Sina.com ). In terms of start-ups, unicorns such as Zhipu AI , Baichuan Intelligence , MiniMax , and Dark Side of the Moon (1X) have emerged; according to statistics, there have been 106 financings in the field of large models in China since 2023, and the industry structure has taken shape ( China has more than DeepSeek - Nikkei Chinese website ). By the end of 2023, China has nearly 80 large models with more than 1 billion parameters and more than 10 large models with 10 billion parameters, ranking first in the world (the "six little tigers" of large models are in trouble, so let the bullets fly for a while - Yiou ). The scale of the entire Chinese large model market is also expanding rapidly: the market output value in 2023 is about 14.7 billion yuan, and it is expected to grow to 21.6 billion yuan in 2024 ( commercialization is the key to the development of the large model industry | AI_Sina Finance_Sina.com ). Major technology companies have invested heavily in this field, and the industry concentration has begun to increase. Some small and medium-sized model companies are seeking to be acquired or listed due to insufficient financial strength.

Latest technological advances and products

In the past six months, China's big model technology has been significantly iterated. In the second half of 2023, China ushered in a "big model price reduction wave". Baidu, Alibaba Cloud, etc. have lowered the price of big model API calls (the reduction is as high as more than 90%) to lower the threshold for enterprise use and promote commercialization. In October 2023, Baidu released Wenxin Big Model 3.5 and opened the dialogue application open service, continuously upgrading core capabilities such as "slow thinking", so that the model has made breakthroughs in complex reasoning, retrieval enhancement, and multi-agent collaboration. iFlytek launched Spark Cognitive Big Model 2.0 in October 2023, claiming to have surpassed ChatGPT in Chinese understanding and question-answering capabilities, and plans to catch up with GPT-4 in 2024. Tencent released the Hunyuan Big Model and opened its self-developed NLP pre-training model to the outside world. Alibaba has open-sourced the Tongyi series of models (such as Qwen-7B/14B) and provides models as a service on its cloud platform. In August 2023, the first batch of models that passed regulatory review (including Baidu, iFlytek, Baichuan, Zhipu, etc.) will be open to the public, marking a new stage of standardized development of domestic big models. Overall, China's big model technology is advancing in sync with the international frontier, and is constantly innovating in knowledge enhancement, tool calling, cross-modality, etc. Some indicators have reached the world's first-class level.

Upstream and downstream relationships and competition landscape of the industry chain

The upstream of the large model industry chain mainly includes computing chips and cloud computing platforms (such as NVIDIA GPU and domestic Cambrian and Huawei Ascend chips , training platforms provided by Alibaba Cloud, Baidu Smart Cloud, etc.) and training data (a large amount of high-quality labeled data is required as raw materials for model training). The midstream is model research and development and algorithm framework (including model development teams from universities, scientific research institutions and enterprises, as well as deep learning frameworks such as PaddlePaddle and MindSpore). The downstream is application deployment , covering the implementation of applications in various industries such as search engines, office software, financial customer service, and medical consulting. The current competitive landscape presents a situation of "large companies dominating and startups supplementing innovation" : on the one hand, Internet and ICT giants such as Baidu, Alibaba, Tencent, and Huawei occupy a dominant position relying on their strong financial and computing power advantages, and regard large models as a strategic focus; on the other hand, a group of startups and research institutions quickly follow up in the subdivided fields, forming a prosperous scene of "hundred-model war" . However, as competition intensifies, a certain degree of resource integration has occurred in the industry : some believe that some small model companies that lack funds will be acquired or eliminated, and industry concentration will increase. At the same time, the debate between open source and closed source for big models is also unfolding in China: Baidu believes that open source is of little significance and emphasizes self-developed closed source to ensure performance, while companies such as 360 advocate an open source ecosystem. In the future, the field of large models may enter a new stage where applications are king , and players will compete not only on the scale of model parameters, but also on the implementation effect and ecological construction in industry scenarios .

Financial Status and Investment Opportunities

The R&D of large models requires huge investment in the short term, and the profit model is still being explored. Although giants such as Baidu and Alibaba are financially sound, their large model businesses are still in the investment stage. For example, Baidu's intelligent cloud department's R&D expenditure on large models continues to grow, and it is currently exploring monetization through enterprise cloud services and Wenxin Yiyan payment interfaces. It is estimated that by 2028, the scale of China's large model industry will reach 117.9 billion yuan, with an annual growth rate far higher than the global average. Investors pay close attention to the field of large models: since 2023, more than 100 venture capital investments have poured into this track, with a total amount of hundreds of billions of yuan. Many large model startups have obtained unicorn-level valuations (such as Zhipu AI and Baichuan Intelligence, which are valued at more than 20 billion yuan). However, it should be noted that commercialization is still a key challenge for big model companies. Corporate customers are not willing to pay and have a long profit cycle. For investors, the big model field contains long-term opportunities, but leading companies with technical barriers and application scenarios should be selected. For example, iFlytek (listed on the Shenzhen Stock Exchange) has advantages in Chinese speech and education, and its cognitive big model is expected to be transformed into a performance growth point. Another example is Baidu (listed on NASDAQ/Hong Kong Stock Exchange), which has the conditions to widely empower the industry with Wenxin big models with its search and cloud business ecosystem, thereby increasing its valuation. In general, big model investment needs to focus on the company's technological leadership and scene implementation capabilities . After the survival of the fittest in the industry reshuffle, a group of leading companies with continuous innovation capabilities are expected to win.

Data Annotation

Major companies and market share

Data labeling is the "data food" processing link for AI model training. China's data labeling industry started with crowdsourcing labeling in the Internet era, and now a group of professional service providers have emerged. According to market size statistics, the output value of China's data labeling market will be about 6.08 billion yuan in 2023, a year-on-year increase of 19.7%, and is expected to reach 7.73 billion yuan in 2024. Baidu Smart Cloud is one of the leaders in this field. Relying on its own AI technology, it provides data labeling services covering multiple scenarios such as speech recognition, NLP, and computer vision, serving industries such as autonomous driving, intelligent customer service, medical care, and finance . In terms of professional annotation service companies, Beijing Speechocean and Cloud Testing Data are regarded as the top three companies in the industry. Speechocean focuses on multimodal data annotation, and its customers include domestic and foreign technology companies such as Alibaba, Tencent, Microsoft, Amazon, and research institutions such as Tsinghua University. Its technologies cover voice, image, OCR, NLP and other fields. Its advantage lies in participating in the development of large models and promoting the formulation of data annotation standards, but its customization flexibility is slightly insufficient  Testin is known for its tool platform, providing full-stack annotation services such as voice, text, and vision. It has flexible componentized tools to improve annotation efficiency. Its disadvantage is that its overseas market layout is weak. In addition, some big companies have also laid out in the annotation field. For example, Alibaba's DAMO Academy has developed an annotation platform, and Didi and Meituan have also established internal annotation teams. Overall, the giants' annotation platforms occupy the main market with technology and funds, accounting for more than 60% of the market share; professional third-party companies serve vertical fields and small and medium-sized customers, providing differentiated services.

Latest technological advances and products

As AI has higher requirements for data quality, the data annotation industry has made progress in intelligence and standardization in the past six months. First, the application of automated annotation tools has accelerated: leading companies have developed pre-annotation models and auxiliary review tools based on deep learning, combining machine pre-annotation with manual verification to improve efficiency and consistency. For example, Baidu has integrated its self-developed large model capabilities to achieve intelligent quality inspection and error prompts during the annotation process. Secondly, synthetic data generation has become a hot topic: In order to make up for the lack of real data, more and more companies provide data generation services, generating training data through simulation or enhancement technology to improve the performance of models in various scenarios. Data annotation and data generation are jointly providing AI models with better training "sources". At the policy level, the government attaches great importance to the data annotation industry. At the end of 2024, the National Development and Reform Commission and other departments issued the "Implementation Opinions on Promoting the High-quality Development of the Data Labeling Industry", proposing that by 2027, the average annual growth rate of the data labeling industry will exceed 20%, and a group of influential labeling companies and industrial bases will be cultivated. The opinions encourage the construction of high-quality data sets in key industries such as transportation, medical care, finance, and manufacturing , promote the inclusion of labeling services in government procurement, and support key technology research (such as cross-modal alignment, 4D labeling, large model labeling, etc.). In the past six months, many places have also been building data labeling bases and talent training. It can be foreseen that standards, specifications and technical platforms will be further unified in the future: In July 2024, the China Academy of Information and Communications Technology took the lead in issuing the big data labeling quality evaluation standard to provide standardized guidance for the industry. In general, data labeling is transforming and upgrading from the traditional human sea tactics to human-machine collaboration, high quality and high standards .

Upstream and downstream relationships and competition landscape of the industry chain

In the industry chain, data annotation is in the upstream basic link . The upstream directly relies on various raw data sources , including data captured from the Internet, IoT sensor data, industry databases, etc.; it also relies on software manufacturers that provide annotation tools. The labeled high-quality data flows to the downstream AI algorithm development and model training links to support the training of autonomous driving perception models, medical imaging AI, speech recognition systems, etc. It can be said that data annotation provides "high-quality raw materials" for AI models, and its quality directly affects the model effect. The current competitive landscape presents a situation where large companies' platforms and professional companies coexist . The internal annotation platforms of large technology companies have financial and technical advantages, can process massive amounts of general data and provide one-stop services, squeezing out part of the market space. However, small and medium-sized professional annotation companies are still competitive in niche areas . For example, some startups focus on medical pathology data annotation or autonomous driving sensor fusion annotation , meeting specific needs with high professionalism. These companies have gained a foothold in areas where giants have not yet cultivated by improving the flexibility and accuracy of their services. It is worth noting that with the development of AIGC, data synthesis and enhancement will reduce the demand for repetitive labor for manual labeling to a certain extent, but at the same time, emerging applications continue to generate new labeling needs (such as multimodal dialogue data labeling). Therefore, it is expected that the data labeling industry will still be in high demand and intensified competition in the short term : on the one hand, giants will continue to integrate resources to enhance economies of scale, and on the other hand, small companies may seek survival space through differentiated services or mergers and acquisitions. Overall, data labeling has become an indispensable part of the AI ​​industry chain , and its development level is directly related to the basic strength of China's AI industry.

Financial Status and Investment Opportunities

The data labeling business is labor-intensive and has relatively low gross profit margins, and the profitability of companies in the industry is obviously differentiated. Taking Haitian Ruisheng , the domestic labeling leader , as an example, the company achieved operating income of 170 million yuan in 2023, a year-on-year decrease of 35%, and a net profit loss of about 30 million yuan . The decline in performance was partly due to the reduction in spending by major overseas customers, showing that labeling services are susceptible to downstream prosperity. However, driven by the AI ​​big model boom, the market demand for high-quality data has surged, and government policy support has provided a good environment for the industry, and the long-term investment value in the field of data labeling has attracted attention. In 2023, emerging AI fields such as humanoid robots and autonomous driving will flourish around the world, creating a large number of professional data labeling needs. At present, there are few pure data labeling concept stocks in the A-share market. Haitian Ruisheng is one of the companies listed on the Science and Technology Innovation Board. Its stock price has rebounded after rising due to AI themes in 2023, reminding investors to pay attention to the fundamentals of the company. Investment opportunities worth paying attention to in the future include: leading labeling services with diversified businesses (companies with long-term stable large customers and technology accumulation, with a high probability of winning in the industry reshuffle); labeling tool and platform companies (with self-developed labeling software platforms that can improve efficiency and reduce labor costs, and are expected to obtain higher profit margins); and crowdsourcing platforms and human resources suppliers that benefit from the growth in labeling demand . In addition, the wave of large models has highlighted the value of data resources, and some companies with rich industry data accumulation (such as data service companies in the fields of medical imaging and remote sensing imaging) are likely to be favored by industrial capital. Overall, the data labeling industry is under short-term profit pressure but has clear growth. Driven by both policies and markets, leading companies with technological and scale advantages are expected to usher in a performance inflection point, thereby bringing investment returns.

Chip computing power

Major companies and market share

"Chip computing power" is the cornerstone of the AI ​​industry, mainly referring to high-performance computing chips and systems used for AI training and reasoning. For a long time, China's AI computing power market has been highly dependent on imported chips such as NVIDIA's GPU. In 2023, the scale of China's AI chip market is about 120.6 billion yuan, and it is expected to grow to 144.7 billion yuan in 2024, of which GPU chips account for as much as 85% ( Cambrian's stock price hit a new high during the trading session, the market value exceeded 300 billion yuan, and the company is still in a loss | Cambrian_Sina Finance_Sina.com ). This means that most of the AI ​​computing power on the market is currently provided by GPUs, and the GPU field is basically dominated by NVIDIA . Faced with "stuck neck", many AI chip companies have emerged in China in recent years to try to break through. Cambrian is one of the leaders in domestic AI chip design. Its products include the "Siyuan" series of cloud training chips and edge AI chips. It has provided NPU design for Huawei mobile phones. Cambrian's stock price hit a new high in early 2024, and its market value once exceeded 320 billion yuan, becoming one of the leading A-share chip companies. Huawei HiSilicon has built a computing power platform based on its self-developed Ascend series AI chips (such as Ascend 910) and has deployed them in some data centers. BiRen Technology focuses on high-end general-purpose GPUs and has launched chips such as BR100 to match the performance of Nvidia A100. According to China Business News, 85% of China's AI chip market in 2023 will be GPUs, and domestic manufacturers such as Cambrian's products are still in the ASIC or NPU category, and there is a gap with Nvidia in general computing power share ( Cambrian's stock price hit a new high during the trading session, and its market value exceeded 300 billion yuan, but the company is still in a loss | Cambrian_Sina Finance_Sina.com ). In the field of autonomous driving chips, the Journey series of automotive AI chips developed by Horizon Robotics occupy a major share of the local pre-installed market, competing with Nvidia's Orin chips. Alibaba's Pingtouge Semiconductor has launched AI inference chips such as Hanguang 800 and developed general-purpose GPUs. In addition to the above-mentioned companies, startups such as Enflame and Denglin Technology are also launching deep learning training accelerator cards. Overall, there are currently a large number of AI chip companies in China, but the market share is still dominated by overseas chips; however, in specific segments (such as security vision AI chips and vehicle-mounted AI chips), local companies already have certain advantages. According to data from the Prospective Industry Research Institute, the compound growth rate of China's AI chip industry in the past five years has been nearly 80%, and it is expected to continue to grow rapidly in the next few years.

Latest technological advances and products

Since 2024, the Sino-US "chip war" has prompted China to make a series of new advances in the field of chip computing power. In terms of high-end GPU replacements, the United States upgraded its export controls in October 2023, and also included Nvidia's A800 and H800 chips specially made for China in the restrictions (Nvidia details advanced AI chips blocked by new export controls | Reuters ). In response to this, China quickly introduced alternative solutions: Huawei released the latest Ascend AI cluster at the end of 2023, claiming that its performance in large-model training is close to that of A100; BiRen Technology adjusted its chip design to avoid restrictions, and its new GPU can be legally supplied after a slight reduction in performance. The construction of large-model training clusters has been accelerated: At the beginning of 2024, a certain intelligent computing center deployed more than 18,000 domestic AI acceleration cards with a computing power of 6.9 EFLOPS, achieving 100% localization of AI chips ( the "ten thousand cards" wave of intelligent computing is rising, and domestic AI chips are ushering in a highlight moment ). This is backed by the key technical support provided by Cambrian, and it also marks that China's domestic AI computing power cluster has reached a world-class scale. In terms of chip technology , Cambrian, Alibaba Pingtou Ge, etc. are all trying to design AI chips using 7nm and more advanced processes. In January 2024, Cambrian revealed that its new generation of chips will optimize large model training and provide better support for mainstream networks such as Llama2 and Tongyi Qwen. In the field of automotive AI chips , Horizon released an upgraded version of the Journey 5 chip in 2023, with improved computing power and support for multi-sensor fusion. The high-level autonomous driving computing platform began to be mass-produced in some vehicle models. It can be seen that in the past six months, domestic AI chips are evolving towards high performance and specialization : on the one hand, striving to make breakthroughs in general-purpose GPU/TPU, and on the other hand, launching customized AI SoCs in specific scenarios (such as smart driving and smart homes). Although there is still a gap with the world's top chips, driven by policies and demand, China's AI chip technology is advancing faster than expected ( CITIC Securities: In 2025, the attractiveness of China's AI assets will become more apparent - Wall Street Journal ). Nvidia CEO Jensen Huang also said that China's progress in the field of AI is "very fast" and is not limited to large models. Chips, algorithms and other aspects are also being promoted ( CITIC Securities: In 2025, the attractiveness of China's AI assets will become more apparent - Wall Street Journal ). Looking ahead to 2025, with the domestic substitution of EDA software and the improvement of wafer foundry technology, domestic AI chips are expected to launch more competitive products.

Upstream and downstream relationships and competition landscape of the industry chain

The AI ​​chip computing power industry chain includes upstream chip design, manufacturing and packaging and testing, midstream computing board/server integration, and downstream application deployment. In the upstream link, there are a number of domestic companies such as Cambrian and Horizon at the design level, but they are highly dependent on wafer fabs such as TSMC and SMIC in terms of manufacturing technology. At present, the manufacturing capacity of 7nm and below process chips is limited by international regulations, which is the biggest challenge facing domestic high-end AI chips. In the midstream, server manufacturers including Inspur Information, Huawei, and H3C are responsible for integrating AI chips into computing boards and complete machines, and building AI training clusters for downstream use. Downstream are various AI scenario application parties , such as cloud service providers renting computing power to large model training, and automobile manufacturers embedding AI chips into autonomous driving domain controllers. In terms of the competitive landscape, on the one hand, international giants monopolize the high-end : Nvidia, AMD, etc. have mastered the most advanced GPU/accelerator technology and ecology, and dominate the high-end training chip market. On the other hand, domestic manufacturers have divided their forces to break through : Cambrian, Suiyuan and others compete with international rivals in data center training chips; Huawei HiSilicon, Horizon, Yitu and others have gained a certain market share in terminals and vertical scenarios (mobile phone AI chips, security chips, and vehicle chips). This has formed a situation of misaligned competition. For example, Cambrian's cloud chips have been used in China Mobile's AI server clusters, forming a substitute relationship with Nvidia ( Cambrian's intraday stock price hit a new high, market value exceeded 300 billion yuan, and the company is still in a loss | Cambrian_Sina Finance_Sina.com ). Horizon has shipped more than one million pieces in China's front-end ADAS market, defeating Mobileye to become the local champion. It can be foreseen that in the short term, high-end general computing power will still be led by Nvidia, but as domestic substitution advances, domestic and foreign manufacturers will each occupy some market segments . It is worth mentioning that the Chinese government and leading Internet companies are supporting local chips through investment and cooperation: Alibaba and Tencent have invested in domestic AI chip start-ups, and the National Integrated Circuit Fund has also focused on AI chips. These measures will help domestic manufacturers occupy a more advantageous position in the next round of technology iteration.

Financial Status and Investment Opportunities

AI chips are capital and technology intensive industries. Currently, most domestic AI chip companies have not yet made a profit, but the capital market highly recognizes their prospects. Take Cambrian as an example. The company is still in the loss stage - the loss in the first three quarters of 2024 was about 724 million yuan, and the R&D investment was as high as 355% of the revenue - but its stock price rose nearly 6 times in 2024, and the total market value exceeded 300 billion yuan at the end of the year. Investors expect that the future wave of artificial intelligence will drive the explosive growth of computing power demand and are willing to give core chip assets a high valuation. From the perspective of industrial investment, AI chips are a national strategic priority, and related companies continue to receive policy and financial support: since 2023, many start-ups have raised more than US$100 million, and the valuations of emerging companies such as Dark Side of the Moon and Bai Ge have risen rapidly ( After a valuation of 20 billion, the test of Dark Side of the Moon has truly begun - China Entrepreneur Network ). However, it should be noted that this field has a fast technological evolution and serious money burning, and some companies may be eliminated in the competition. Investment opportunities can be considered from several aspects: First, listed companies that have deployed AI chips . Cambrian-U in A shares is sought after due to its scarcity ( Market Portrait | 257.33 trillion yuan, the 2024 A-share report card is released! AI chip leader wins the increase ); in addition, Ziguang Guowei (002049) has also potential through acquisitions to deploy AI acceleration chip IP. Second, traditional GPU supply chain companies , such as TSMC (overseas) that manufactures for Nvidia and Inspur Information (SZ: 000977) that provides boards and cards, will still benefit from strong demand for computing power in the short term. Third, AI chip application companies that are expected to achieve revenue growth , such as Inspur and Haiguang Information (listed in Shanghai, focusing on server CPU and DCU), play an important role in improving domestic computing power. In addition, for investors with high risk tolerance, they can pay attention to the IPO dynamics of AI chip companies in the growth stage in the Science and Technology Innovation Board or Hong Kong stocks. Once these companies' products achieve key breakthroughs and enter the harvest period, their investment returns may be considerable. Of course, the chip industry is also facing uncertainties in international trade and technology, and investment needs to comprehensively consider the intensity of policy support and the company's core competitiveness.

Data Center

Major companies and market share

Data centers are physical infrastructure that carry AI computing power and data storage. China's data center industry has expanded rapidly in recent years and has become one of the largest markets in the world. By the end of 2023, the total scale of data center racks in use nationwide will exceed 8.1 million standard racks, and the total computing power will reach 230 EFLOPS, ranking second in the world. Chinese data center operators can be divided into three categories: Internet cloud vendors' self-built data centers , telecom operators' data centers , and third-party data center service providers . In the first category, technology companies such as Alibaba, Tencent, Baidu, and Huawei have built a large number of self-owned or leased data centers to support their cloud services and AI businesses. Taking Alibaba Cloud as an example, hundreds of data center nodes have been deployed across the country. The second category is the three major telecom operators (Telecom, Mobile, and Unicom), which have many IDC computer room resources across the country and provide rack leasing and network access services for government, enterprises, and cloud vendors. The third category includes professional IDC companies such as GDS , 21Vianet , and Chindata . GDS is one of the largest third-party data center operators in China, serving financial institutions and large Internet companies; Chindata has been deeply involved in the "East Data West Computing" hub area and has built multiple super-large data center clusters. According to Mordor Intelligence's analysis, China will have 443 large data centers by 2023, the largest number in the Asia-Pacific region. In terms of market share, Alibaba Cloud has long ranked first in the public cloud IaaS market and has the largest computing power infrastructure scale; Tencent Cloud and Huawei Cloud are closely behind. Some of these self-built data center resources are also open to leasing. Among third-party IDC vendors, GDS and 21Vianet have leading shares in first-tier cities, while Chindata has an advantage in Inner Mongolia, Zhangjiakou and other places. In general, the top five operators in China's data center market (including Alibaba, Tencent, etc.) account for more than half of the rack share, and the rest is shared by many small and medium-sized service providers. It is worth noting that the increase in China's data centers in 2023 will mainly come from the construction of the western hub node of the "East Data West Computing" project, and the regional layout will be further optimized.

Latest technological advances and products

In the past six months, China's data center sector has made progress mainly in green intensive expansion and new computing power clusters . The "East Data West Computing" project has been further promoted: as of the third quarter of 2024, the total scale of data center racks in use in the country's eight major computing power hubs (such as Inner Mongolia, Guizhou, Gansu and other western regions) has exceeded 2.11 million, doubling year-on-year. These regions have sufficient energy and suitable climate, and by taking on the computing power needs of the east, the national computing power resources can be optimized. Looking forward to 2025, the National Data Bureau requires that the new computing power in the hub node regions account for more than 60% of the new national additions, and the proportion of clean energy used in new data centers exceeds 80%. This means that in the future, large data centers will be more deployed in the west and powered by green energy. In terms of liquid cooling and high-density technology , in response to the high power consumption and heat dissipation problems of AI training clusters, many domestic IDCs began to deploy liquid cooling servers and high-voltage DC power supply solutions in 2024 to improve energy efficiency. For example, Tencent has applied immersion liquid cooling on a large scale in its Qingyuan data center, and the PUE has dropped to below 1.2. Ultra-large-scale AI computing power clusters have emerged: In 2024, major cloud vendors have established dedicated clusters for large model training. Alibaba Cloud announced the construction of a "Tongyi" computing power cluster containing tens of thousands of GPUs; Baidu Smart Cloud also released the Qianfan large model computing center, with a daily call volume of tens of billions. A landmark event is that Huawei has joined hands with governments in many places to build an AI computing power center that fully uses domestic Ascend chips, with a single cluster computing power of 100 trillion times per second (EFLOPS), providing a self-controllable computing base for domestic AI. The integration of government and enterprise data centers is also in progress: In 2024, local governments actively promoted the integration of government cloud and government data centers to reduce duplication and improve utilization. It can be seen that the data center industry is shifting from pursuing scale expansion to emphasizing efficiency, greenness, and intelligence . New forms such as edge data centers and distributed clouds are also emerging to support 5G and IoT applications. These new trends in technology and layout lay the foundation for the development of future data centers and also meet the computing power needs that are exploding in the AI ​​era.

Upstream and downstream relationships and competition landscape of the industry chain

The data center industry chain covers upstream infrastructure and equipment suppliers, midstream data center construction operators, and downstream cloud services and corporate users. The upstream includes civil engineering, power equipment (transformers, uninterruptible power supply UPS), refrigeration systems (precision air conditioning, cooling towers) and IT equipment (servers, storage devices, network equipment, etc.) suppliers. China has a number of excellent companies in this regard, such as Vertiv Technologies and Shenzhen Invke , which provide data center temperature control solutions, and Inspur Information and Unisplendour Corporation provide servers and switches. The midstream is the planning, construction and operation of the data center , including telecommunications operators and IDC professional companies, who are responsible for site selection, design, computer room construction, operation and maintenance management, and customer leasing. Downstream customers are cloud computing vendors, Internet companies, and government and enterprises in various industries , who use data center resources by listing servers or leasing computing power. In terms of the competitive landscape, the scale effect of the strong is always strong . Large operators have advantages in funds and customer resources, and can continue to invest in the construction of ultra-large-scale parks to provide services at a lower unit cost. Small IDCs often focus on differentiated positioning (such as serving local small and medium-sized enterprises or edge data centers in special scenarios). At the same time, win-win cooperation is also a trend: cloud vendors often cooperate with third-party IDCs to rent computer rooms or jointly build data centers to serve customers together. The main barriers to the industry are large initial investments, high requirements for operating experience and safety and reliability, and it is difficult for new entrants. However, as policies encourage the circulation of data elements , some local governments or industrial parks have also begun to participate in data center projects, providing preferential electricity prices and taxes to attract computing power projects to land. Overall, China's data center industry is fully competitive and the market demand is strong. Major players are consolidating their positions by improving energy efficiency and expanding scale . For example, Alibaba and Tencent continue to upgrade the density of existing computer rooms in the east and build large-scale computing power bases in the west; GDS and others strengthen service quality and customized solutions to counter the self-built trend. In the future, with the advancement of infrastructure REITs, data center asset securitization is expected to inject new sources of funds into the industry ( China Data Center Industry Development In-depth Research and Investment Prospect Analysis Report (2024-2031) ). This will further intensify competition and promote the entire industry chain to become more mature and stable.

Financial Status and Investment Opportunities

Data center business is characterized by heavy assets and long payback period, but in the era of " digital economy", its stable cash flow and growth prospects are favored by capital.) , the company's continuous expansion has led to a high debt ratio, but its revenue has maintained an annual growth of more than 20%, its occupancy rate has been above 90% for a long time, and its profitability has gradually increased. Another example is that Qinhuai Data has seized the opportunity of "East Data West Computing" in recent years, achieved rapid performance improvement and was listed on the US stock market (now in the process of privatization and delisting). China's data center market is expected to exceed one trillion yuan in 2029 ( China Data Center Industry Development In-depth Research and Investment Prospect Analysis Report (2024-2031) ), with an average annual growth rate of more than 15% in the next five years. Therefore, investing in data centers is equivalent to investing in "digital economic infrastructure." At present, domestic investors can pay attention to several types of opportunities: IDC listed companies - Baosight Software (600845) listed on the A-share market has been working in the field of industrial data centers for many years and has a steady performance; 21Vianet ( http:// VNET.US) is listed in the U.S., but its business focuses on the domestic IDC leasing market. The IDC assets of operators , such as China Telecom's network operation company and China Mobile's information port company, will be worth looking forward to if they release value through spin-off listing or REITs. Data center equipment suppliers , such as UPS leader Easy (SZ: 300376) and modular computer room supplier Invic (SZ: 002837), are expected to grow in performance as new projects increase. In addition, the surge in computing power demand under the AI ​​wave is also good for data center-related industries: cloud service providers (Alibaba, Tencent, etc.) that provide computing power support for AI and GPU server manufacturers (Inspur, Sugon) will benefit directly. It is worth noting that the data center industry needs to be vigilant against regional supply and demand imbalances and energy consumption policy risks. But overall, driven by the long-term trend of digital transformation and artificial intelligence, data centers, as an important component of "new infrastructure" , are still areas with strategic investment value, and their leading companies and upstream and downstream high-quality companies deserve medium- and long-term attention.

Autonomous driving

Major companies and market share

China's autonomous driving industry has flourished in recent years, giving birth to many technology companies and test operation projects. In the field of Robotaxi (driverless taxis), Baidu Apollo is in a leading position. Baidu's Apollo Go has carried out passenger demonstration operations in nearly ten cities including Beijing, Shanghai, Shenzhen, and Wuhan, with a fleet size of hundreds of vehicles. As of October 2024, Apollo Go has provided more than 8 million rides in total, and the number of orders has increased by about 20% year-on-year ( Baidu Apollo Go is about to start running in Hong Kong! Related tests will begin as early as the end of the year-Moomoo ). Another unicorn , Pony.ai , has also put about 250 vehicles into operation in Beijing, Guangzhou and other places, and will be listed in the United States through SPAC in November 2024 (valued at about billions of dollars), and plans to expand to 1,000 fleets nationwide by 2026 ( Beijing unveils plans to boost driverless vehicle use in capital | Reuters ). WeRide has carried out Robotaxi and driverless minibus pilots in Guangzhou, Shenzhen, Nanjing and other cities, and has obtained the first fully driverless taxi operating license in Guangzhou. AutoX has been testing driverless cars on open roads in cities such as Shenzhen and Shanghai, and is known for its radical technology. In addition, Didi also has an autonomous driving department in Shanghai and other places. According to Reuters, as of 2023, at least 19 Chinese cities have carried out Robotaxi or driverless bus tests ( Beijing unveils plans to boost driverless vehicle use in capital | Reuters ). Among all companies, Baidu Apollo Go currently ranks first in operating mileage and order volume, and has a leading market share ( Baidu Carrot Run is about to start running in Hong Kong! Related tests will begin as early as the end of the year-Moomoo ). In terms of passenger car autonomous driving assistance, China's new car-making forces are not far behind: Xiaopeng Motors launched the city NGP pilot assisted driving, realizing high-level urban road automatic navigation driving in cities such as Guangzhou and Shanghai; Weilai Automobile and Li Auto also upgraded the city NOA function in 2024. The sales growth of these smart electric vehicles has rapidly increased the penetration rate of consumer assisted driving functions. In general, China's autonomous driving industry presents a "two-pronged" pattern: travel services are led by technology companies such as Baidu and Xiaoma in the Robotaxi pilot; mass-produced passenger cars are driven by car companies to promote the implementation of high-end ADAS. As the technology matures, the boundaries between the two sides are gradually blurring, forming a greater market integration.

Latest technological advances and products

In the past six months, the field of autonomous driving has made a number of milestones. Autonomous driving policies are more open: In early 2024, Beijing issued new regulations (effective from April 1, 2024) to support the use of autonomous vehicles in private cars, buses and taxis, and plans to gradually allow fully autonomous buses and taxis to be put into operation ( Beijing unveils plans to boost driverless vehicle use in capital | Reuters ). Wuhan also simultaneously issued innovative policies for intelligent connected vehicles to encourage industrial development ( Beijing unveils plans to boost driverless vehicle use in capital | Reuters ). These landmark policies have cleared institutional obstacles for autonomous vehicles to hit the road. Test scope expanded : By the end of 2024, the cumulative mileage of autonomous driving test roads opened in Beijing and Shanghai exceeded 1,000 kilometers, and Shenzhen, Chongqing and other places have also designated unmanned manned demonstration areas. From the second half of 2023, Beijing's Yizhuang and Shunyi districts will allow Baidu and Pony to conduct unmanned tests with driver seats unmanned without safety officers, marking the technology entering a higher stage ( Pony AI begins robotaxi test with driver seat unmanned on Beijing's ... ). The fleet size and commercialization continue to increase: Baidu has started charging for its Robotaxi service in Wuhan, and plans to deploy 1,000 autonomous vehicles in Wuhan to provide travel services by the end of 2024 ( Beijing unveils plans to boost driverless vehicle use in capital | Reuters ). Pony.ai was approved to operate Robotaxi at Beijing Daxing Airport in 2024, providing unmanned logistics transfers for airport passengers ( Baidu and Pony.ai receive permit to offer robotaxis at Daxing airport ). On the technical level, multi-sensor fusion and AI algorithm refinement have improved the safety of unmanned vehicles on complex urban roads. Baidu Apollo demonstrated the sixth-generation unmanned vehicle technology architecture at the 2024 Apollo Day, equipped with a new generation of laser radar and a more powerful on-board AI computing platform, which can better handle night and bad weather scenarios. Vehicle-to-road collaboration (V2X) is also one of China's special solutions: In 2024, smart transportation projects in many places began to allow road infrastructure to interact with autonomous vehicles. For example, Shanghai deployed millimeter-wave radars and AI signal machines on some sections of the road to assist autonomous vehicles in obtaining real-time road conditions. It is worth mentioning that Tesla announced plans to launch its Full Self-Driving (FSD) system in China in 2025, but it needs to wait for regulatory approval ( Beijing unveils plans to boost driverless vehicle use in capital | Reuters ). This news has made domestic companies feel the competitive pressure and accelerated the iteration of technology and testing. Overall, after six months of development, China's autonomous driving has transitioned from "small-scale demonstration" to "large-scale trial operation", and major players are working hard to achieve the leap from R&D verification to early commercialization of technology.

Upstream and downstream relationships and competition landscape of the industry chain

The upstream of the autonomous driving industry chain includes chip and sensor suppliers, such as Nvidia (Orin, Drive series) and domestic Horizon Robotics (Journey series), which provide in-vehicle AI computing chips, as well as LiDAR manufacturers Hesai Technology and RoboSense. The midstream is the autonomous driving system development companies, that is, companies that integrate algorithms, software and hardware (Baidu, Pony.ai, WeRide, etc.), some of which also cooperate with OEMs to modify test vehicles. The downstream refers to travel service operators (Robotaxi fleet operators, such as Apollo Go) and automobile OEMs (installing high-level autonomous driving functions in mass-produced vehicles, such as Xiaopeng and Weilai). The entire ecosystem also involves support links such as high-precision map service providers (NavInfo, AutoNavi, etc.) and cloud monitoring and scheduling platforms. The current competitive landscape can be described in two parallel lines: " technology companies vs. automobile manufacturers ." On the one hand, technology companies such as Baidu and Xiaoma focus on the research and development of fully driverless technology, competing fiercely with each other to achieve reliable driverless services first; on the other hand, new car-making forces and traditional car companies compete in the L2-L4 assisted driving market, competing through different sensor solutions (lidar route vs. pure vision route) and user experience. It is worth noting that China's characteristics lie in collaboration : Baidu Apollo not only cooperates with GAC, WM Motor and other car companies to mass-produce autonomous driving models, but also operates Robotaxi services on its own; Huawei provides ADS advanced driving systems for multiple models as a supplier, and does not directly build cars. In this model, competition and cooperation are intertwined: technology companies need OEMs to produce vehicles, and OEMs need technology companies to enable intelligent driving. The advantage is that it can accelerate the implementation of technology, and the disadvantage is that the profit model requires exploration and coordination by all parties. At present, Baidu Apollo has a certain first-mover advantage due to its technological maturity and government relations, while start-ups (Xiaoma, WeRide, etc.) have introduced large car companies and travel platforms as shareholders (such as Toyota investing in Xiaoma and Renault-Nissan Alliance investing in WeRide), forming an alliance confrontation. In addition, there is fierce competition in key parts such as LiDAR: China's Hesai Technology (Nasdaq: HSAI) maintains a leading position in global automotive LiDAR shipments, and its products are installed in Li Xiang cars and some Robotaxi (In 2025, the humanoid robot industry will usher in a "hundred-machine war" | Robot_Sina Finance_Sina.com ), while American Velodyne and others are gradually withdrawing from the Chinese market. Overall, China's autonomous driving competition is in a state of competition among many players , and each company is staking out users and data: Robotaxi competes in service coverage of cities and order volume, and passenger cars compete in smart driving reputation and sales. In the foreseeable few years, there will be a wave of integration in this field , and those with insufficient technical and financial strength may be acquired or marginalized, and eventually a few leading players will stand out.

Financial Status and Investment Opportunities

The autonomous driving industry is still in the investment stage, and most companies have not yet achieved profitability, but financing is active and valuations are high. Companies in the field of Robotaxi rely on venture capital and strategic investment to operate: Pony.ai will raise more than $1.1 billion in cumulative financing by 2023, and will go public through a special purpose acquisition company (SPAC) in 2024 to raise funds. Although it is still not profitable, its valuation has reached billions of dollars ( Beijing unveils plans to boost driverless vehicle use in capital | Reuters ). WeRide and Yuanrong Qixing are also preparing for listing or the next round of financing. Burning money to operate has caused financial pressure on many start-ups. For example, Jingchi Technology and Roadstar.ai , which were once active , have been eliminated. At present, companies with large shareholder support have more advantages: Baidu Apollo's R&D and operating expenses are supported by Baidu, and iFlytek, GAC and other shareholders in WeRide provide endorsement. In the consumer passenger car market, smart electric vehicle manufacturers make profits through car sales, and their cash flow is relatively more stable. If investors are optimistic about the direction of autonomous driving, there are several paths: First , they can invest in vehicle companies , such as Xpeng Motors (NYSE: XPEV), which has a leading position in autonomous driving and has seen its market value increase due to its intelligent advantages. Second, they can focus on key component companies . For example, Hesai Technology, the leading laser radar company, received financial support after its listing to expand production capacity and research and development. Its revenue in 2023 increased by 73% year-on-year ( In 2025, the humanoid robot industry will usher in a "hundred-machine war" | Robot_Sina Finance_Sina.com ). Third, there are investment opportunities in software algorithm companies . Most of them are not listed at present, but they can be participated in through venture capital funds or wait for their IPO. It should be reminded that the business model of autonomous driving is not yet mature, and the large-scale profitability of Robotaxi still depends on the relaxation of regulations and cost reduction. According to ARK Investment Research, to achieve large-scale adoption, the cost of future humanoid robots (similar to the cost of unmanned vehicles) needs to be reduced to less than US$20,000, which is about half of the current cost. This analysis also applies to the field of driverless cars: reducing the cost of sensors and computing power and improving fleet utilization are the keys to breaking even. Optimistically, China's huge travel market has bred a huge space for driverless replacements. Once the technology matures and can replace drivers, it will release considerable profit potential. Therefore, for investors with high risk tolerance and a long-term perspective, autonomous driving is still a high-risk and high-potential field. At present, we can focus on companies with clear mass production plans (such as technology companies that have deep cooperation with car manufacturers) and companies that are the first to achieve profitability in specific scenarios (such as Easy Control Intelligent Driving, which has achieved commercialization of driverless driving in mining areas). In short, investment in the autonomous driving track needs to balance short-term losses and long-term return expectations, and gradually intervene in the pace of industry maturity.

Humanoid Robot

Major companies and market share

Humanoid robots are known as the "crown jewel of robots" and will become a new hotspot for global technology investment in 2023. China's humanoid robot research and development layout started later but has developed rapidly. At present, there are more than 80 complete machine companies engaged in the research and development of humanoid/humanoid robots in China, and nearly 150 companies worldwide. Major players include: Fourier Intelligence - Its universal humanoid robot GR-1 was unveiled at the 2023 World Artificial Intelligence Conference. It is 1.65 meters tall, weighs 55 kilograms, has 40 degrees of freedom, can walk at 5 kilometers per hour and carry 50 kilograms. Fourier plans to produce 100 GR-1s for scientific research and rehabilitation by the end of 2023. It is one of the fastest-growing companies in China in this field ( GR-1 general-purpose humanoid robot will carry nearly its own weight ). UBTECH , a long-established service robot company, has previously released the humanoid robot Walker, which has the ability to walk and go up and down stairs. Although it faces great financial pressure, it is still continuing to develop humanoid robot technology. Unitree , a company that started out as a quadruped robot, founder Wang Xingxing said that it is actively developing humanoid robots and plans to launch consumer-grade humanoid robots with lower prices in the future. Zhiyuan Robotics , founded by former Huawei genius Peng Zhihui (nicknamed "Zhihuijun"), received consecutive rounds of large-scale financing in 2023, focusing on the development of open-source embodied intelligent humanoid robots, has attracted attention in the industry. In addition, laboratories under giants such as Xiaomi and Tencent are also developing humanoid or humanoid robots: Xiaomi has demonstrated the CyberOne concept robot, and Tencent Robotics X Laboratory has invested in research on multimodal robots. As for overseas competitors , Tesla's Optimus robot aims to mass-produce thousands of units in 2025, Norway's 1X Robotics, the United States Agility, etc. have also attracted attention from the domestic industry. Therefore, it can be said that the field of humanoid robots is currently in a stage of " many heroes emerging " and no one dominates. Chinese companies have potential advantages in cost control and diversity of application scenarios. Although there is no clear market share statistics yet (because most of them are in the prototype and small batch trial production stage), if we look at the scale of financing, Fourier Intelligence, Zhiyuan Robotics, UBTECH, etc. are in the forefront. In terms of financing, according to statistics, there were 23 financings in the field of humanoid robots in the world in 2023 alone, with a total amount of approximately RMB 5.47 billion, a ten-year high; this investment enthusiasm continued in the first half of 2024, with more than 10 financings in China, with a total amount of more than RMB 2 billion.

Latest technological advances and products

2024 is regarded by the industry as the start of the first year of mass production of humanoid robots . In the past six months, this field has made breakthroughs in technology and products. Prototype demonstrations are moving towards practical applications : At the end of 2023, Fourier Intelligence announced that its GR-1 humanoid robot had entered the small-scale production stage, and demonstrated application scenarios such as autonomous walking, carrying objects, and assisted walking training. This means that domestic humanoid robots are beginning to move from the laboratory to actual scenes. AI big model empowers robots : Many companies are trying to connect large language models to humanoid robots to give them more natural communication and cognitive capabilities. Fourier's GR-1 is said to be equipped with an LLM-driven dialogue and decision-making system that can perform tasks based on voice commands ( GR-1 - FOURIER-Robotics ). Tencent and others have proposed the concept of "embodied intelligence", hoping to give physical robots an intelligent center for understanding the environment and learning. Progress in key components : The core components of humanoid robots, such as servo motors, reducers and battery packs, are factors that restrict their performance and cost. In early 2024, domestic manufacturers such as Green Harmonic and Shuanghuan Transmission launched new lightweight robot joint reducers in an effort to reduce costs. Tsinghua University and other research institutions have also announced the research and development results of high-power density servo motors. Industry collaboration : In Shanghai, Shenzhen and other places, a robot industry alliance has been formed, and universities, enterprises and capital have jointly tackled common problems of humanoid robots, such as balance control and visual navigation, to accelerate the transformation and implementation of results. Foreign benchmarking and catching up : Tesla's Optimus released the latest progress video at the end of 2023, and the robot can autonomously identify and pick up objects. The Chinese R&D team pays close attention to this and follows up on the details of the algorithm through open source communities and papers. Overall, the current technical inflection point has not yet fully arrived , but the industry expects that it will gradually move out of the "concept verification" stage and move towards the mass production application stage from 2024 to 2025 ( 2025 Humanoid Robot Industry Research: The first year of mass production begins, and the market space is broad-Future Think Tank ). Especially as more players enter the market, humanoid robots are expected to see a "hundred-machine war" situation like the "large models" of the past ( In 2025, the humanoid robot industry will usher in a "hundred-machine war" | Robots_Sina Finance_Sina.com ). However, it should also be recognized that the commercialization of humanoid robots still faces many challenges: costs need to be significantly reduced, battery life and reliability need to be improved, and application scenarios need to be verified. However, around 2025, some vertical scenarios (such as collaborative robots in manufacturing and medical care) are expected to be the first to have usable humanoid robot products.

Upstream and downstream relationships and competition landscape of the industry chain

The upstream of the humanoid robot industry chain covers the supply of core components , including high-torque density servo motors, reducers, sensors (IMU, force sensors, etc.), batteries and materials. These parts largely determine the upper limit of the robot's performance and the lower limit of its cost. At present, Japanese manufacturers (Harmonic, Nabtesco) are technologically advanced in the field of reducers, but domestic companies are gradually catching up; there are a large number of domestic supplies available for motors and batteries. The midstream is the robot body R&D and manufacturing companies, that is, companies that design mechanical structures, integrate software and hardware, and develop control algorithms, such as Fourier and UBTECH. The downstream is the application scenarios and services , which potentially include medical rehabilitation, elderly care, hotel reception, dangerous work replacement (such as disaster relief), home assistants, etc. At present, humanoid robots are still mainly in the prototype testing stage and have not yet really entered the downstream large-scale commercial use. In terms of the competitive landscape, it is currently a typical technology-driven early stage competition : companies are competing in comprehensive R&D strength and iteration speed. As the market is not yet mature, the strategies of various companies are slightly different - some emphasize open source cooperation (such as Zhiyuan Robotics claims to accelerate industry popularization through open source ( In 2025, the humanoid robot industry will usher in a "hundred-machine war" | Robot_Sina Finance_Sina.com )), some choose to keep it secret and iterate quickly; some take the high-performance route (pursuing more real-life athletic ability), and some focus on low-cost practicality (even if the functions are simple but cheap). It can be foreseen that as more and more teams join, this field will soon be divided : a few technologically advanced and well-funded ones will stand out, and most latecomers may be integrated or turn to specific segments. The industry has predicted that in 2025, there will be a "hundred-machine war" for humanoid robots, exploring commercialization in a crowded track ( In 2025, the humanoid robot industry will usher in a "hundred-machine war" | Robot_Sina Finance_Sina.com ). China has some advantages in this emerging industry: relatively low-cost engineering talents and manufacturing capabilities, and broad local application needs (such as the demand for companion robots in an aging society). But there are also disadvantages, such as the gap between basic theory and original technology and the United States and Japan. Therefore, in the next few years, Chinese and foreign companies may divide the work at different levels: for example, domestic companies may be the first to run through the model in specific scenarios , while the breakthrough of sensors/controllers at the basic level may rely more on global collaboration. The competition pattern will ultimately depend on who can first solve the practical bottleneck of humanoid robots , including reliability, safety and cost. It can be expected that by the time humanoid robots are truly mass-applied, the industry pattern may have become clear, and some giants with ecosystem support (such as Tesla, or "Tesla-style" companies that may appear in China) and manufacturers that master key technologies will dominate.

Financial Status and Investment Opportunities

At present, most humanoid robot companies are in the R&D investment period, and there is no stable profit model. They mainly rely on equity financing to support operations. In 2023, venture capital financing in this field reached a peak in ten years, indicating that capital recognizes its vision. Among China's representative companies, UBTECH received investment from Tencent, China Merchants Group and others in the early stage, and was once valued at more than US$5 billion. However, due to slow commercialization progress, it failed to sprint for an IPO in 2022 and is currently seeking to transform the enterprise market. Fourier Intelligence has also continuously raised funds in recent years to expand its rehabilitation robot and humanoid robot businesses. It is reported that in the first half of 2023 alone, domestic humanoid robot investment and financing exceeded 2 billion yuan ( In 2025, the humanoid robot industry will usher in a "hundred-machine war" | Robot_Sina Finance_Sina.com ). It can be foreseen that if a company achieves a technological breakthrough in the next 1-2 years, it will quickly obtain huge financing or even IPO. However, for investors, humanoid robots belong to a high-risk field, and short-term returns are difficult to predict. Investment opportunities can be considered from the following perspectives: Core component suppliers - no matter which robot wins, it will need support from high-performance joint motors, reducers, power batteries, etc. Therefore, pay attention to the trends of companies such as Green Harmonic and Qinchuan Machine Tool (reducer field). These companies may usher in new markets due to the robot wave. Related AI and software companies - humanoid robots require powerful vision and decision-making AI. Large model companies and computer vision companies may use this to expand new business lines. Related listed companies such as Hongsoft Technology (visual algorithms) are worth paying attention to. Potential listing targets - if companies such as Zhiyuan Robotics and Fourier are listed on the Science and Technology Innovation Board or Hong Kong stocks, they will become a window for direct investment in humanoid robots. It should be noted that there are currently no pure humanoid robot listed companies, and investments are mostly in the form of PE/VC, but as mass production approaches, some companies may start the IPO process. Finally, in the long run, if humanoid robots are successfully commercialized, their market space is extremely broad (theoretically, every human job may be replaced by robots). ARK Investment predicts that its scale will be comparable to the automotive industry ( In 2025, the humanoid robot industry will usher in a "hundred-machine war" | Robots_Sina Finance_Sina.com ). Therefore, early layout of related leading companies has the opportunity to obtain several times or even dozens of times the return. However, in the short term, the industry may also experience bubbles and adjustments : if the technological progress is not as expected or the cost remains high, the investment enthusiasm will cool down. In general, for the frontier field of humanoid robots, a "small proportion, long cycle" investment strategy should be adopted, and it should be regarded as a strategic opportunity for the next 5-10 years. Continue to track the evolution of technology and the dynamics of benchmark companies, and seize the opportunity to enter at the right time.