Source: Guosheng Securities
Authors: Liu Gaochang, Chen Zhijing

Core Viewpoint

As the prices rise in the cloud upstream, the computing power enters a new round of upward cycle.

1) With the successive launch of deep learning models such as Deepseek R1 and openAI o1/o3, the demand for high-performance computing at the inference end is rapidly released. Major companies like Alibaba and Tencent are quickly expanding their capital expenditures, with a single-quarter capital expenditure growth rate exceeding 250% in Q4 2024. The cloud upstream officially enters a new round of upward cycle.

2) In various segments, price increase signals have emerged to varying degrees in the fields of cloud computing, IDC, servers, GPUs, and computing power rental, verifying the economic environment of the cloud upstream and benefiting leading manufacturers across the industry chain.

• Cloud platform: Led by AWS, cloud computing prices have gradually risen, with the H100 cloud computing invocation price increasing by approximately 22% compared to two weeks ago as of March 4.

• IDC: The global demand for intelligent computing centers has surged. In North America, supply is relatively tight due to factors like electricity and cooling. IDC service fees have shown a significant increase in 2023, with the highest fees exceeding 60%. Domestically, leading manufacturers have significantly increased their listing rates, and resources in core cities are quite tight. IDC is upgrading to AIDC, and with the continuous rise in AI application demand, the rent for AI-specific data centers is also expected to increase.

• GPU and Server: Core components of AI computing power are increasingly tightening in domestic supply. Driven by the DeepSeek craze, major manufacturers are focused on$NVIDIA(NVDA.US)$The demand for H20 chips is strong, with the unit price for an 8-card machine around 1.1 million yuan, an increase of at least 0.1 million compared to the beginning of the year.

• Computing power leasing: card sources and funding advantages are the core competitiveness, with the leasing price for Coreweave H100 servers in March 2025 expected to be around $49/hour, a nearly 29% increase from $38/hour in April 2023, reflecting strong demand from downstream.

There is a high enthusiasm for cloud deployment, and the demand for inference is expected to continue to be released.

1) From the perspective of the BG side, the government and central state-owned enterprises have recently actively connected to Deepseek, and the demand for cloud deployment is being rapidly released. For example, Shandong Public Utilities completed the localization of Deepseek with China Electronics Cloud, reconstructing business scenarios for heating, water supply, sanitation, and more; the AI platform under $Nanyang (7417.JP)$Xinggong also officially connected to Deepseek, significantly optimizing in areas such as Chinese comprehension, long text processing, and logical reasoning. Additionally, the all-in-one machine can meet large customers' privatization needs regarding information security and data privacy, with recent demand surging; over 60 companies, including major enterprises like Huawei and Lenovo, have launched all-in-one machines based on DeepSeek.

2) From the perspective of the C-end, individual users are using Deepseek more deeply, and Tencent Yuanbao has surpassed Doubao in downloads on February 22 thanks to the smooth experience of the "full-version" Deepseek. 4) The rise of hybrid cloud deployment and public cloud usage is expected to continuously release inference demand.

The utility of model deployment has improved, and integrated software systems will become the main carrier of AI applications.

1) In the past, the cost of large model inference was relatively high, which limited user expansion to a certain extent. IT companies found it difficult to sustain subscription-based payments, for example.$Salesforce (CRM.US)$In 2024, a usage-based pricing model was officially introduced, requiring 2 dollars for each conversation. The high performance and low cost of Deepseek will significantly drive down token prices, promote a decrease in AI application charges, user expansion, and ROI improvement, accelerating the arrival of AI application growth.

2) General-purpose software of substantial scale generally serves as the core IT system within enterprises, efficiently bringing management benefits to the business processes to retain high-engagement paying customers. This type of general-purpose software generally links core processes of internal and external business matters, covering various aspects such as administration, finance, procurement, sales, inventory, human resources, customer relations, marketing sales, and document assets, making it a primary carrier for upgrading AI applications, including CRM, ERP, OA, document office platforms, and data terminals.

Related symbols:

IAAS:$Alibaba (BABA.US)$、$Guangdong Aofei Data Technology(300738.SZ)$、$Kehua Data Co.,Ltd.(002335.SZ)$、$Hainan Huatie(603300.SH)$、$INESA Intelligent Tech Inc. (600602.SH)$、$Xiamen Hongxin Electronics Technology Group Inc. (300657.SZ)$、$Sharetronic Data Technology (300857.SZ)$、$Shandong Sacred Sun Power Sources (002580.SZ)$、$RunJian Co.,Ltd. (002929.SZ)$、$Sangfor Technologies Inc. (300454.SZ)$、$Digital China Group (000034.SZ)$、 $Shenzhen Sed Industry (000032.SZ)$ 、$Pinggao Group Co., Ltd. (688227.SH)$、Kingsoft Cloud (03896.HK).、$Zhejiang Narada Power Source (300068.SZ)$、 $Cloudwalk Technology (688343.SH)$ 、 $Ucloud Technology Co., Ltd (688158.SH)$ 、 $Cambricon-UW(688327.SH)$ 、$Inspur Electronic Information Industry(000977.SZ)$、$Dawning Information Industry(603019.SH)$、$Taiji Computer Corporation (002368.SZ)$、$Capital Online Data Service (300846.SZ)$、$Hang Zhou Iron & Steel (600126.SH)$、 $Shanghai AtHub (603881.SH)$ 、$Nanxing Machinery (002757.SZ)$、$Zhejiang Huace Film & TV (300133.SZ)$、$Hangzhou Shunwang Technology (300113.SZ)$、$EmbedWay Technologies (603496.SH)$、$Wangsu Science & Technology (300017.SZ)$、$Netac Technology (300042.SZ)$and so on.

Domestic computing power: $Cambricon-U(688256.SH)$ 、$Haiguang Information (688041.SH)$and so on.

SaaS:$Beijing Kingsoft Office Software, Inc (688111.SH)$、$Yonyou Network Technology (600588.SH)$、$KINGDEE INT'L (00268.HK)$、$Shanghai Weaver Network (603039.SH)$、$Dingjie Smart (300378.SZ)$、$Yusys Technologies (300674.SZ)$、 $Northking Information Technology Co., Ltd (002987.SZ)$ 、$Beijing Seeyon Internet Software Corp.(688369.SH)$、$Shanghai Golden Bridge InfoTech(603918.SH)$、$Hand Enterprise Solutions (300170.SZ)$、$Netac Technology(300682.SZ)$、$Shanghai Ganglian E-Commerce Holdings(300226.SZ)$、$Shanghai Newtouch Software(688590.SH)$、$Hithink RoyalFlush Information Network(300033.SZ)$、$萤石网络(688475.SH)$、$Shanghai Runda Medical Technology(603108.SH)$、$Sinodata Co., Ltd.(002657.SZ)$、$Hundsun Technologies Inc.(600570.SH)$、 $Star环科技-U(688031.SH)$ 、$Winning Health Technology Group (300253.SZ)$、$B-Soft Co.,Ltd.(300451.SZ)$、$Softtone (301236.SZ)$、$Hangzhou Raycloud Technology Co.,Ltd (688365.SH)$、$Iflytek Co.,ltd. (002230.SZ)$、$Wondershare Technology Group(300624.SZ)$、Haitian Information,$Dark Horse Technology Group(300688.SZ)$、$Maifushi (02556.HK)$、$Zhejiang China Commodities City Group (600415.SH)$、$Shenzhen Kingdom Sci-Tech (600446.SH)$、$Fujian Apex Software (603383.SH)$、 $Netac Technology(300682.SZ)$ 、 $Jingtai Holdings-P(02228.HK)$ 、$Chengdu Jiafaantai Education Technology(300559.SZ)$、$Jiahe Meikang(688246.SH)$、$Newland Digital Technology(000997.SZ)$、$Newcapec Electronics (300248.SZ)$and so on.

AI chain of major Internet companies: $Cambricon-U(688256.SH)$ 、$Bestechnic (Shanghai) Co., Ltd. (688608.SH)$、$Kid King (301078.SZ)$、$ Tianjian Co., Ltd. (301383.SZ) $、$ Shanghai Fortune Techgroup (300493.SZ) $、$ Shifeng Cultural Development (002862.SZ) $、$Espressif Systems (688018.SH)$、 $萤石网络(688475.SH)$ 、$Semiconductor Manufacturing International Corporation (00981.HK)$、$Range Intelligent Computing Technology Group (300442.SZ)$、$Shenzhen Honor Electronic (300870.SZ)$、$HMT (Xiamen)New Technical Materiasls Co.,Ltd (603306.SH)$、 $Inspur Electronic Information Industry(000977.SZ)$ 、$ZTE (00763.HK)$、 $Dawning Information Industry(603019.SH)$ 、$GigaDevice Semiconductor Inc. (603986.SH)$、$Guoguang Electric (002045.SZ)$、$Shenzhen Farben Information Technology (300925.SZ)$、$Yacong Co., Ltd. (301085.SZ)$、$Shenzhen Shining Environment (301018.SZ)$、$Zhejiang Zhaolong Interconnect Technology (300913.SZ)$and so on.

Military Industry AI:$Nancal Technology (603859.SH)$、 $Pinggao Group Co., Ltd. (688227.SH)$ 、$Cetc Potevio Science&Technology (002544.SZ)$、$Guangzhou Haige Communications Group Incorporated (002465.SZ)$and so on.

Brain-computer Interface:$Beijing Chieftain Control Engineering Technology (300430.SZ)$、$Beijing Relax Technology (688793.SH)$、$Sanbo Neuroscience (301293.SZ)$and so on.

Risk Warning:

Risks from AI technology iteration falling short of expectations, changes in trade relations, macroeconomic risks, and intensified competition in the Industry.

Report body

1. As the prices in the cloud upstream begin to rise, computing power enters a new upward cycle.

Deep learning models such as OpenAI o1/o3 and Deepseek R1 are constantly being iterated, and the demand for high-performance computing on the inference side is being rapidly released. The global cloud upstream market is experiencing a general wave of price increases. In areas such as cloud computing, IDC, servers, GPUs, and computing power leasing, price increase signals have appeared to varying degrees. We believe that the cloud upstream has officially entered a new upward cycle, indicating new opportunities for capital expenditure and expansion in the computing infrastructure sector, which is bullish for leading companies in various segments of the industry chain.

The theoretical profit margin of Deepseek exceeds expectations, and the ROI of computing investment has become increasingly clear.

On March 1, 2025, Deepseek officially released a document disclosing key information such as its theoretical costs and profit margins for the first time. It is estimated that assuming a GPU leasing cost of $2/hour and tokens charged at R1 pricing, the theoretical profit margin is as high as 545%, and the ROI of computing investment is becoming clearer. Although this is a theoretical value, considering the market's supply and demand volatility, and that the current DAU scale of Deepseek is still in the tens of millions, making it difficult to fully meet market demand, the profitability effects brought by computing power and tokens may gradually become apparent as computing demand continues to expand.

Alibaba, Tencent, and other large companies are rapidly expanding their Capex, validating the prosperity of domestic AI computing power.

1) On the inference side, as the ROI of computing investment improves and the capabilities of large models continue to break through, AI model performance is expected to undergo a qualitative change this year; we are bullish on the continuous expansion of cloud computing demand brought by inference applications.

2) On February 24, 2025, Alibaba CEO Wu Yongming announced that Alibaba would invest over 380 billion yuan in the next three years to build cloud and AI hardware infrastructure; on March 19, 2025,$TENCENT (00700.HK)$The announcement states that the capital expenditure involved in the AI project development for the fourth quarter of 2024 is expected to be 39 billion yuan. The company anticipates further increasing Capex in 2025, with an expected Capex-to-revenue ratio in the low double digits percentage. The top-down capital expenditure guidance from major cloud providers is expected to strengthen market confidence in the prospects of computing power investments. The industry is showing Bullish Signals.

Cloud Platform: Represented by AWS, cloud computing prices have gradually increased, with H100 usage prices rising over 20% in two weeks. The cost and performance of Deepseek R1 continue to optimize, opening up the demand for inference-side computing power, as high-performance computing power prices steadily increase. Taking AWS as an example, on March 4, the H100 cloud computing usage price increased by about 22% compared to two weeks ago.

IDC: Rental prices in North America have significantly increased in 2023, with the maximum rise exceeding 60%.

1) IDC data centers are an important part of computing power infrastructure. As global demand for intelligent computing centers surges, the data center industry has gained significant development benefits, with investment and financing activities showing a win-win situation for supply and demand. Investment in IDC data center construction continues to rise, while operating costs for electricity, cooling, land, etc., also remain on the rise.

2) According to feedback from UBS regarding North American cloud providers, supply in North American data centers is increasingly tight due to factors such as electricity and heat dissipation. Some data centers have already been reserved before starting construction, and cabinet rental prices have also risen rapidly, from around 80 USD/kW to 120-130 USD/kW, with the highest price increase exceeding 60%.

Resources in key domestic cities are tightening, and rents for AI-specific data centers may increase.

1) The market for intelligent computing services in China has grown by 85.8% year-on-year in 2023. Leading IDC service providers have benefited from the strong demand for emerging GenAI infrastructure, with the data center listing rate steadily increasing. Taking IDC service providers as an example.$21Vianet (VNET.US)$For example, according to the financial report, as of June 30, 2024, the equipment rack utilization rate of the company’s base-type data center is 75.9%, showing a year-on-year increase of 63.4% and a quarter-on-quarter increase of 71.0%.

According to Statistics from Smart Frontline, in the public bidding market in 2024, various winning projects related to intelligent computing centers consistently show a tiered upward trend, jumping to a new level almost every quarter. By December 2024, the monthly number of winning projects has already reached at least 100. According to Yang Yang, research manager at IDC China Enterprise Research Department, as IDC upgrades to AIDC, larger-scale computing clusters and higher power density, along with value-added service opportunities brought about by data center transformations, will push the IDC market into a new phase of rapid growth. With the continuous rise in AI application demand, the rent for AI dedicated data centers is also expected to increase.

GPU and Server: The core components of AI computing power, domestic supply may become increasingly tight, and H20 prices have shown significant increases.

From an overseas perspective, taking Nvidia’s A100/H100 as an example, the A100 is based on the Ampere architecture, produced with a 7nm process, and has 6,912 CUDA cores, capable of providing a maximum of 312 TFLOPS of AI computing power; the H100 adopts the Hopper architecture and is equipped with HBM3e memory, which has increased its maximum memory capacity within the same package size compared to the A100, and its pricing shows a clear premium compared to the previous generation products. According to third-party price monitoring platforms, in August 2024, the price of H100 on e-commerce channels reached the highest point of 44,999 USD.$Amazon (AMZN.US)$On the e-commerce channels, the price of H100 reached the peak at 44,999 dollars.

From a domestic perspective, according to reports from Caixin, driven by the DeepSeek craze, large manufacturers have a strong demand for Nvidia H20 chips due to cost-effectiveness and compliance considerations. Some Dealers have indicated that the H20 series is the computing card with the largest price increase. 'Currently, a machine with 8 cards is priced around 1.1 million yuan, which has increased by at least 0.1 million compared to the end of last year.'

Computing power leasing: Card sources and funding advantages are the core competitive strengths, and leasing prices are expected to gradually increase.

In the short term, building AI computing power independently requires significant capital investment, and there may also be shortages on the supply side. For some manufacturers, leasing computing power is a more cost-effective solution in the short term. Manufacturers are increasing their computing power leasing services, leading to a rise in leasing prices.

Taking CoreWeave, a computing power leasing company partially owned by Nvidia, as an example, as of March 2025, the leasing price for an 8-card H100 server is $49.24/hour, compared to $4.76/GPU/hour in April 2023 (which is $38.08/hour for the 8-card server), an increase of 29.31%. The prices for high-end computing services are on an upward trend, reflecting the urgent demand from enterprises for real-time inference computing services.

II. There is a high enthusiasm for cloud deployment, with inference demand expected to continue to be released.

Government and enterprises are actively deploying Deepseek, with high demand for government cloud and integrated machine. Compared to previous large models, Deepseek has significantly improved cost-effectiveness, attracting numerous government and enterprise clients for deployment.

In terms of cloud services:

1) According to reports from China Electronics Cloud, ShanDong Public Utility has completed the localized deployment of the national open-source large model DeepSeek, becoming one of the first enterprises in the domestic public utility sector to achieve deep application of AI large models. This move marks the official entry of ShanDong public utility services into the "AI+" era, reconstructing core business scenarios such as heating, water supply, and sanitation with autonomous and controllable intelligent technologies.

2) According to reports from Nanyang Wanbang, Nanyang Xing Gong, as an important participant in the field of artificial intelligence applications, has officially connected to the Deepseek series models, significantly optimizing Chinese comprehension, long text processing, and logical reasoning.

In terms of integrated machines:

According to reports from the Financial Associated Press, more than 60 companies, including major enterprises such as Huawei and Lenovo, have launched all-in-one machines based on DeepSeek. According to the Kechuangban Daily, the all-in-one machines currently on the market are divided into inference all-in-one machines and training all-in-one machines, equipped with different sizes of models like DeepSeek-R1 32B, 70B, and full-strength version 671B, with prices ranging from hundreds of thousands to millions. The all-in-one machines can meet the privatization needs of large clients regarding information security and data privacy, and quickly meet client demands and go live in a short time.

C-end user enthusiasm is high, and Tencent's Yuanbao download volume has surpassed Douchou.

1) According to AI Fan, on February 22, Tencent's application 'Yuanbao' surpassed 'Douchou' to rank second in the free app download chart in the Apple App Store in China, while DeepSeek continues to hold the top position.

2) According to user feedback, Tencent Yuanbao has become 'one of the best DeepSeek clients' due to its smooth experience with the 'full-strength version', high-quality public account article retrieval, image recognition, and other differentiated functions, even being deemed 'better than the original application'. Its technical capabilities may be closely related to Tencent's recent strategy of integrating DeepSeek across its entire product line—including core products like QQ Music, QQ Browser, WeChat Work, and WeChat—each gradually incorporating DeepSeek services. After WeChat's integration, traffic surged, prompting Tencent to direct some users to Yuanbao, further boosting downloads.

Three, model utility enhancement, comprehensive software systems will become the main vehicle for AI applications.

In the past, the high cost of large model inference restricted user expansion.

1) Taking Salesforce as an example, its AI pricing strategy was originally on a monthly subscription basis; for instance, the price for Sales GPT in 2023 was $50 per user per month, which included a certain number of Einstein GPT credits; similar to Sales GPT, Service GPT will be included in Service Cloud Einstein, priced also at $50 per user per month.

2) However, in 2024, Salesforce introduced a usage-based pricing model on the Agentforce platform, where each conversation costs $2. 3) It is believed that with the rapid increase in usage, the inference cost associated with user tokens will also rise quickly, making the monthly subscription model potentially unable to generate reasonable profits for IT companies. Pricing can shift toward a usage-based approach while also significantly increasing AI application prices. In the current situation where the effects brought by AI cannot be clearly quantified, this will limit the speed of user expansion.

The high performance and low cost of Deepseek will significantly drive down token prices, promote a decrease in fees for AI applications, expand user bases, and enhance ROI, marking the arrival of a tipping point for AI applications.

General software serves as the core IT system for enterprise management and will become the main carrier of AI applications.

1) Mature general software typically acts as the core IT system within an enterprise, and only when it brings significant management benefits to the operational processes can it attract highly engaged customers and strong willingness to pay.

2) Such general software usually connects the core processes of internal and external enterprise matters, covering various aspects such as administration, finance, procurement, sales, inventory, human resources, customer relations, marketing, and document assets, thus becoming the main carrier for upgrading AI applications.

1. CRM - Customer Relationship Management.

Agentforce 2.0 has been launched, covering various fields including sales, customer support, and market analysis. On December 17, 2024, global tech giant Salesforce officially introduced its new platform - Agentforce 2.0. CEO Marc Benioff announced at the launch event that Salesforce plans to hire 2,000 sales personnel to accelerate the promotion of this significant AI platform, with services expected to be offered to global enterprises by February 2025.

2. ERP: Enterprise Resource Planning.

HR SaaS and AI+ financial management directions are gradually being implemented.

1) HR SaaS can be used in various scenarios, specifically including abnormal analysis of attendance monthly reports, intelligent talent discovery reasoning search scenarios, and so on.

2) The potential for AI in financial management is large, and the tasks are relatively complex, including comprehensive budgeting, risk control, cost control, financial process automation, and various other aspects.

3. OA: Collaborative Office Automation

OA serves as a high-frequency entry point for daily work, with many intelligent management scenarios in marketing, contracts, customer service, projects, and so on.

1) OA is a high-frequency entry point for general employees to use the company's system. With the development of mobility and cloud technology, OA is gradually evolving into a broadly defined collaborative office platform, including functions such as process approval, document management, communication and collaboration, schedule management, and mobile office.

2) Taking the 泛微数智大脑Xiaoe.AI as an example, it features intelligent Q&A and AI search, intelligent summaries and content analysis, intelligent content writing and proofreading, AI image recognition and data generation, AI intelligent data analysis and intent recognition, intelligent process engine and automation processing assistant, and RPA automation engine for business and data processing.

4. Document Office Platform

Documents are the core carrier of work knowledge. AIGC creation evolves from usable to easy to use. Taking WPS AI as an example, it will launch a new 2.0 version in July 2024, releasing AI writing/reading/data/design assistants for personal office use; and officially launching an intelligent document library for enterprises that supports intelligent Q&A and creation; for the government version, it will continue to optimize for vertical scenarios, including government AI writing/Q&A, and so on.

5. Data Terminal

Big Data has become the foundational support for work decision-making, and for high-frequency data scenarios in finance and commodities, AI is expected to provide deep empowerment.

1) Hithink RoyalFlush Information Network's 'Wencai' is a representative product as a financial assistant, offering capabilities such as intelligent agents, stock selection, stock diagnosis, market overview, investment learning, and more.

2) For Shanghai Ganglian, its 'XiaoGang' digital intelligent assistant version 1.6 is launched on multiple channels such as 'My Steel Network' website, APP, and Ganglian Data Terminal. Through interactive dialogues with the intelligent assistant, 'XiaoGang' can provide content services such as price checking, news reading, report writing, encyclopedia inquiries, and business opportunities finding. Additionally, it possesses market analysis, multi-modal generation, and intelligent customer service functions, assisting with article reading and providing AI summaries. The generated content serves various fields in the commodity industry every day.

Four, Related Symbols

IAAS:$Alibaba (BABA.US)$、$Guangdong Aofei Data Technology(300738.SZ)$、$Kehua Data Co.,Ltd.(002335.SZ)$、$Hainan Huatie(603300.SH)$、$INESA Intelligent Tech Inc. (600602.SH)$、$Xiamen Hongxin Electronics Technology Group Inc. (300657.SZ)$、$Sharetronic Data Technology (300857.SZ)$、$Shandong Sacred Sun Power Sources (002580.SZ)$、$RunJian Co.,Ltd. (002929.SZ)$、$Sangfor Technologies Inc. (300454.SZ)$、$Digital China Group (000034.SZ)$、 $Shenzhen Sed Industry (000032.SZ)$ 、$Pinggao Group Co., Ltd. (688227.SH)$、Kingsoft Cloud (03896.HK).、$Zhejiang Narada Power Source (300068.SZ)$、 $Cloudwalk Technology (688343.SH)$ 、 $Ucloud Technology Co., Ltd (688158.SH)$ 、 $Cambricon-UW(688327.SH)$ 、$Inspur Electronic Information Industry(000977.SZ)$、$Dawning Information Industry(603019.SH)$、$Taiji Computer Corporation (002368.SZ)$、$Capital Online Data Service (300846.SZ)$、$Hang Zhou Iron & Steel (600126.SH)$、 $Shanghai AtHub (603881.SH)$ 、$Nanxing Machinery (002757.SZ)$、$Zhejiang Huace Film & TV (300133.SZ)$、$Hangzhou Shunwang Technology (300113.SZ)$、$EmbedWay Technologies (603496.SH)$、$Wangsu Science & Technology (300017.SZ)$、$Netac Technology (300042.SZ)$and so on.

Domestic computing power: $Cambricon-U(688256.SH)$ 、$Haiguang Information (688041.SH)$and so on.

SaaS:$Beijing Kingsoft Office Software, Inc (688111.SH)$、$Yonyou Network Technology (600588.SH)$、$KINGDEE INT'L (00268.HK)$、$Shanghai Weaver Network (603039.SH)$、$Dingjie Smart (300378.SZ)$、$Yusys Technologies (300674.SZ)$、 $Northking Information Technology Co., Ltd (002987.SZ)$ 、$Beijing Seeyon Internet Software Corp.(688369.SH)$、$Shanghai Golden Bridge InfoTech(603918.SH)$、$Hand Enterprise Solutions (300170.SZ)$、$Netac Technology(300682.SZ)$、$Shanghai Ganglian E-Commerce Holdings(300226.SZ)$、$Shanghai Newtouch Software(688590.SH)$、$Hithink RoyalFlush Information Network(300033.SZ)$、$萤石网络(688475.SH)$、$Shanghai Runda Medical Technology(603108.SH)$、$Sinodata Co., Ltd.(002657.SZ)$、$Hundsun Technologies Inc.(600570.SH)$、 $Star环科技-U(688031.SH)$ 、$Winning Health Technology Group (300253.SZ)$、$B-Soft Co.,Ltd.(300451.SZ)$、$Softtone (301236.SZ)$、$Hangzhou Raycloud Technology Co.,Ltd (688365.SH)$、$Iflytek Co.,ltd. (002230.SZ)$、$Wondershare Technology Group(300624.SZ)$、Haitian Information,$Dark Horse Technology Group(300688.SZ)$、$Maifushi (02556.HK)$、$Zhejiang China Commodities City Group (600415.SH)$、$Shenzhen Kingdom Sci-Tech (600446.SH)$、$Fujian Apex Software (603383.SH)$、 $Netac Technology(300682.SZ)$ 、 $Jingtai Holdings-P(02228.HK)$ 、$Chengdu Jiafaantai Education Technology(300559.SZ)$、$Jiahe Meikang(688246.SH)$、$Newland Digital Technology(000997.SZ)$、$Newcapec Electronics (300248.SZ)$and so on.

AI chain of major Internet companies: $Cambricon-U(688256.SH)$ 、$Bestechnic (Shanghai) Co., Ltd. (688608.SH)$、$Kid King (301078.SZ)$、$ Tianjian Co., Ltd. (301383.SZ) $、$ Shanghai Fortune Techgroup (300493.SZ) $、$ Shifeng Cultural Development (002862.SZ) $、$Espressif Systems (688018.SH)$、 $萤石网络(688475.SH)$ 、$Semiconductor Manufacturing International Corporation (00981.HK)$、$Range Intelligent Computing Technology Group (300442.SZ)$、$Shenzhen Honor Electronic (300870.SZ)$、$HMT (Xiamen)New Technical Materiasls Co.,Ltd (603306.SH)$、 $Inspur Electronic Information Industry(000977.SZ)$ 、$ZTE (00763.HK)$、 $Dawning Information Industry(603019.SH)$ 、$GigaDevice Semiconductor Inc. (603986.SH)$、$Guoguang Electric (002045.SZ)$、$Shenzhen Farben Information Technology (300925.SZ)$、$Yacong Co., Ltd. (301085.SZ)$、$Shenzhen Shining Environment (301018.SZ)$、$Zhejiang Zhaolong Interconnect Technology (300913.SZ)$and so on.

Military Industry AI:$Nancal Technology (603859.SH)$、 $Pinggao Group Co., Ltd. (688227.SH)$ 、$Cetc Potevio Science&Technology (002544.SZ)$、$Guangzhou Haige Communications Group Incorporated (002465.SZ)$and so on.

Brain-computer Interface:$Beijing Chieftain Control Engineering Technology (300430.SZ)$、$Beijing Relax Technology (688793.SH)$、$Sanbo Neuroscience (301293.SZ)$and so on.

V. Risk Warning

Risk of AI technology iteration not meeting expectations: If AI technology iteration does not meet expectations, it will have a certain adverse impact on companies in the industry chain. Risk of changes in trade relations: International trade relations are affected by multiple factors, and if changes occur, they may adversely affect the supply chain.

Macroeconomic risk: If the macroeconomic downturn occurs, fixed asset investment slows down, affecting companies' willingness to reinvest, which in turn affects consumers' willingness to spend and the production willingness of the industry chain, causing adverse effects on the entire industry.

Risk of intensified industry competition: If related companies accelerate technology iteration and application layout, the overall level of competition in the industry will intensify, posing a threat to the growth of companies currently in the industry.

Editor/rice