Grasping the investment opportunity of "Artificial Intelligence +" - which six types of companies are worth paying attention to?

Full Summary:

The focus of AI+ is on the “+” rather than AI itself. The top-level design emphasizes that the key of “AI+” lies in its integration with specific scenarios, reflecting industry attributes rather than the technology itself. Therefore, the investment direction should focus on AI empowering specific scenarios. Attention should be paid to two major catalysts:

1. Industry Catalyst: Focus on the investment marginal changes brought about by the iteration of large model versions;

2. Policy Catalyst: On July 31, 2025, the Prime Minister passed the "Opinions on Deepening the Implementation of Artificial Intelligence+" at the State Council meeting, highlighting the importance of this round of policies.

Focus on six major investment logics in the AI industry:

1. Marginal changes brought by large model version iterations: The core of the upcoming GPT-5 release lies in lower hallucination rates, lower prices, and stronger programming capabilities, focusing on application landing. The logic of large models has shifted from "who is stronger" to "who is more useful";

2. Top-down — directly policy-related: Closely follow the full text of the "Opinions on Deepening the Implementation of 'Artificial Intelligence+'" and recent policy trends, targeting specific areas;

3. Bottom-up — event-driven or companies with good fundamentals: From a bottom-up perspective, select relevant companies with event catalysts or smooth progress in AI business landing;

4. Institutional varieties with market capitalization capacity: Large-cap companies (over 10 billion) are more suitable for institutional investment styles;

5. Mapping to US stocks: AI hotspot companies have shifted from upstream (such as NVIDIA) to downstream applications. We believe that the corresponding A-share companies of these US AI application companies also have investment value;

6. Undervalued: The overall valuation level of the AI application sector is currently low, with a significant risk-reward ratio in the future. Starting from low-position companies, it is recommended to focus on relevant companies with expected price-to-earnings ratios below 60X in the coming year.

The following is the main report by Wang Zijing, Chief of Computer at Soochow Securities, on August 16, 2025:

1. AI Application High Ground: What Stage Have We Reached?

1.1. AI applications are on the verge of accelerated growth. From the perspective of the enterprise lifecycle, AI applications are on the verge of accelerated growth. From the lifecycle of a company/a new industry, a company will go through: seed stage → startup stage → rapid growth stage → maturity stage → decline stage. For AI companies, the characteristics corresponding to the founding to maturity stage mainly include: seed stage (cost reduction), startup stage (rapid increase in penetration), rapid growth (ROI realization), maturity stage (platform unification, compliance completeness). We believe that AI applications are currently on the verge of rapid growth. AI applications have currently achieved a cost reduction by half, rapid increase in penetration rate, and revenue validation driving ROI realization, which basically aligns with the characteristics of a fast-growing phase.

1.2. The performance of large models has seen a leap in improvement, with costs experiencing a cliff-like drop. Since 2024, the performance of AI large models has significantly improved. In 2023, researchers introduced new benchmarks—MMMU, GPQA, and SWE-bench—to test the limits of advanced AI systems. Just one year later, performance has greatly improved: scores on MMMU, GPQA, and SWE-bench increased by 18.8, 48.9, and 67.3 percentage points, respectively. Beyond benchmark testing, AI systems have made significant progress in generating high-quality videos, with language model agents even surpassing humans in certain programming tasks with limited time budgets.

Chinese AI laboratories are gradually narrowing the technological gap with their American counterparts. With the release and update of the DeepSeek-R1 model, the performance of Chinese AI laboratory models has reached the o1 benchmark intelligence level and is approaching o3. Since the end of 2024, top Chinese AI laboratories have intensively released multiple high-performance models, significantly narrowing the intelligence level gap between Chinese and American models. Currently, some Chinese models have the capability to compete with models from top American laboratories, with open-weight models represented by DeepSeek and Alibaba reaching and approaching the o1 benchmark in intelligence level.

Large models are becoming more efficient and easier to popularize. Thanks to increasingly powerful models, a system with performance at the GPT-3.5 level has seen its inference costs drop by over 280 times from November 2022 to October 2024. On the hardware level, costs are decreasing by about 30% annually, while energy efficiency is improving by 40% each year. Additionally, the gap between open-weight models and closed models is continuously narrowing, with performance differences in some benchmark tests reduced from 8% to just 1.7% (in just one year). These trends collectively drive the rapid lowering of the threshold for advanced AI applications.

The continuous maturity of large models has led to an accelerated increase in both user numbers and monthly active users. Since the release of GPT in November 2022, the number of users surpassed 1 million within 5 days, reaching 100 million WAU by November 2023, 200 million WAU by August 2024, and 400 million WAU by February 2025, with the time taken for the second doubling reduced by 3 months compared to the first round; DeepSeek gained 10 million users within 20 days of launch, reaching 200 million MAU within 2 months. Looking at the two most popular large models, GPT and DeepSeek, both user numbers and usage are showing an accelerated upward trend.

As of August 1, 2025, NVIDIA is absent from the top five AI stocks in the S&P 500 (with only a 29.2% increase this year), while the majority of AI companies with significant gains are AI application companies, reflecting the market's shift towards the application layer.

2. China needs to fully leverage its comparative advantages in the AI industry game

2.1. Differences in AI industry logic between China and the US There are significant differences in the development logic of the AI industry between China and the US, leading to completely different strategic planning in their path choices. The technological path of artificial intelligence has not converged, so we need to leverage our comparative advantages.

United States: Standing at the top of the global intellectual property pyramid, mining globally; (1) Cloud Services: The four major cloud providers in North America (Google, Amazon, Microsoft, Meta) plan to invest over $300 billion in AI data center construction by 2025, with US companies accounting for over 70% of the market share among the top eight cloud infrastructure service providers; (2) AI Chips: By the end of 2024, NVIDIA's data center GPU market share will reach 92%, forming technological barriers and ecological monopolies. For example, NVIDIA's net profit in the 2024 annual report exceeded $30 billion, with a net profit margin exceeding 53%, indicating high barriers and strong bargaining power. In 2024, the proportion of American chips in China's AI server market reached 63%, indicating a high dependence on the American chip industry in both product and technology.

China: Downstream applications are our core advantage and key breakthrough in the AI industry. China's advantages in the AI industry are reflected in downstream applications. The upstream cloud services and chips of China's AI industry are constrained by the US. Although domestic manufacturers are continuously achieving technological breakthroughs, the significant network effects in the industry make it extremely difficult for China to achieve complete victory in the upstream China needs to leverage its comparative advantages.

(1) Data Advantage: Over the past decade, there has been no significant progress in data openness, and the total amount of data continues to accumulate; a large amount of data remains undeveloped. Each year, municipal integrated media centers can generate over 50TB of cultural data, with over 80% undeveloped. In contrast, the total amount of social data has immense economic potential. Most data in the United States comes from leading companies, which have already been significantly transformed into economic effects during the training of large models.

Edwin Chen, founder of Surge AI, the world's largest data labeling company, mentioned: “In the current bottleneck of AI development, data quality absolutely ranks first, followed by computing power, and then algorithms. Simply investing more computing power cannot solve the problem because if there is no high-quality data for training, or no correct goals and evaluation metrics, you will fall into a trap of seeing false progress.” On June 12, 2025, the data labeling unicorn Scale AI officially announced a strategic investment of $14.3 billion from Meta, which will acquire 49% of the company's non-voting shares. This investment has propelled Scale AI's valuation to $29 billion. Scale AI's revenue reached $870 million in 2024, and after the acquisition on June 12, the corresponding PS multiple reached 33 times. On one hand, this strategic investment has prompted a reevaluation of the foundational role of data in the AI industry; on the other hand, this event also provides us with a window to observe the development of the global data labeling industry. The importance of high-quality datasets in China has been elevated to a new dimension. Before 2025, the development of China's data factor industry was significantly below expectations, mainly due to the technical difficulty of achieving "data not leaving the domain, usable but invisible" at low cost, which greatly reduced the willingness of localities to open data. However, with the development of artificial intelligence technology, this problem has been perfectly solved: as long as the data is encapsulated in vertical models, allowing the downstream to distill the models, the downstream can gain my knowledge but cannot access the data, forming a situation where "data remains static while the model moves," truly achieving data not leaving the domain, usable but invisible. In the future, two new business formats are worth noting: data labeling and data synthesis.

(2) Industrial Chain Advantage: China has the most complete industrial chain in the world. In 2024, China's manufacturing value-added reached $3.6 trillion, accounting for 28.9% of global manufacturing

Manufacturing + Artificial Intelligence = Robots, pointing the direction for the future evolution of embodied intelligence in the market: The core of robots lies in intelligence rather than limbs and torso, therefore the future evolution direction of embodiment is primarily the brain. Policies and industrial development will revolve around the robot brain.

(3) Market Advantages: China has a demand advantage with an ultra-large-scale market, boasting a population of over 1.4 billion and more than 400 million middle-income groups, creating a vast domestic demand market with rich application scenarios. China has a supply advantage with a complete industrial system, being the only country in the world that possesses all industrial categories classified by the United Nations. Additionally, there are over 60 million various enterprises that complement each other and compete for development. China has a talent advantage with a large number of high-quality laborers, including over 700 million laborers, more than 240 million skilled workers, and over 10 million university graduates each year. Behind these advantages lies enormous market potential, which will become a strong backup force for the development of AI applications.

(4) Application Scenario Advantages: Currently, “Artificial Intelligence +” is deeply integrated into specific environments of people's production and life, particularly outstanding in promoting the intelligent upgrade of manufacturing, office, home, and consumption scenarios. ① In the manufacturing sector, traditional assembly line models are gradually being replaced by flexible and customized intelligent production; in office scenarios, artificial intelligence technology is enhancing the efficiency of knowledge workers and expanding their capability boundaries; in home life scenarios, various intelligent products are truly serving human needs; in entertainment and consumption scenarios, artificial intelligence is aiding the development of new consumption models such as the experience economy.

The core of Artificial Intelligence + application scenarios lies in AI Agent. AI Agent is a broad application of AI that ultimately needs to be implemented in actual production scenarios. China has long faced efficiency bottlenecks, a labor-intensive service industry, and continuous pressure on labor costs: (1) Efficiency Bottleneck: China's manufacturing industry has long faced pain points such as low efficiency of manual operations and high isolation between systems (ERP/MES/WMS data silos). Agents can effectively enhance manual operation efficiency and connect system data, breaking through the efficiency ceiling; (2) Labor-Intensive Status: In fields such as customer service and sales, the proportion of repetitive tasks is high. For example, Alibaba's Lingyang customer service Agent can reduce manual operations by 80% and shorten return and exchange times by 60%, allowing human customer service to shift to higher-value tasks (3) Population Structure and Cost Pressure: From a macro perspective, labor costs in our country continue to rise. Between 2011 and 2021, the average annual growth rate of wages for urban employees in our country was 9.8%, while the nominal GDP's average annual growth rate was 8.9%. The speed of labor cost increase is faster than economic growth, showing a compensatory growth trend. Agents will effectively alleviate cost pressure.

The core of the intelligent economy lies in production efficiency, and the evolution of digital employees is a core manifestation of enterprise digital transformation, being one of the key links in the iteration of productivity and production efficiency. Its development can be divided into three generations: Process Automation → Limited Scenario Autonomous Decision-Making → Complex Decision Logic Chain. Digital Employee 1.0 - RPA: Relies on manually configured process automation, solving repetitive labor but lacking intelligence. Digital Employee 2.0 - Intelligent Automation: Integrates AI technologies such as OCR and NLP to achieve autonomous decision-making in limited scenarios. Digital Employee 3.0 - Agent: A full-link closed loop driven by large models, capable of handling complex business logic and continuously evolving.

The above four advantages correspond to four investment directions. We believe that future policy directions and industrial development will revolve around these four advantages.

(1) Data Advantage: High-quality datasets, data labeling, data synthesis;

(2) Manufacturing Advantage: Embodied intelligence, brain;

(3) Market Advantage: Domestic computing power, chips;

(4) Scenario Advantage: AI+, industry attributes; AI agent.**

Artificial intelligence will ultimately empower various industries: (1) Primary Industry: Using large models for AI breeding, yield prediction, and monitoring of crop growth processes; (2) Industry: Using artificial intelligence to reshape industrial software such as CAD/CAE/EDA/PLC; (3) New Energy: Using large models for power prediction, power plant site selection, and the consumption of new energy; (4) Smart Cultural Tourism: Integration of the hotel industry with large models; (5) Others: Including AI+ government affairs, AI+ healthcare, AI+ e-commerce, etc.

3. Industrial Catalysis and Policy Direction AI+ focuses on the “+” rather than AI itself.

**

The top-level design emphasizes that the focus of "AI+" lies in its integration with specific scenarios, reflecting industry attributes rather than the technology itself. Therefore, the key investment directions should concentrate on AI empowering specific scenarios. Attention should be paid to two major catalysts: (1) Industry Catalyst: Focus on the investment marginal changes brought about by the iteration of large model versions (GPT-5 and DeepSeek V4); (2) Policy Catalyst: On July 31, 2025, the Premier approved the "Opinions on Deepening the Implementation of the 'Artificial Intelligence+' Action" at the State Council meeting, highlighting the importance of this round of policies.

3.1 Industry Catalyst: Iteration of Large Model Versions

In the early morning of August 8, 2025, Beijing time, OpenAI released GPT-5. The biggest highlight of this update is the breakthrough progress in landing obstacles: The hallucination rate has dropped to <1%, addressing the biggest pain point of AI applications and clearing core obstacles for practical implementation; On the other hand, it reflects the long-term nature of technological qualitative change and application shift: The leap in large model capabilities requires long-term investment, and GPT-5 has not achieved a qualitative change in functionality, forcing the industry to shift its focus to exploring application scenarios. Cost reduction accelerates commercialization: API prices continue to decline, pushing companies to shift from technological competition to pragmatic application development. Domestic opportunities are released:** The slowdown in OpenAI's technological iteration enhances the imagination space and release expectations for domestic models like V4. The narrative of domestic applications is strongly tied to DeepSeek and has little relation to GPT, so the overall underperformance of GPT's release this time is actually beneficial for the current narrative expectations of DeepSeek V4.

3.2 Policy Direction Interpretation:

Top-level design intensifies artificial intelligence+ On July 31, 2025, Premier Li Qiang presided over a State Council executive meeting, which reviewed and approved the "Opinions on Deepening the Implementation of the 'Artificial Intelligence+' Action." The meeting pointed out that the current artificial intelligence technology is accelerating iteration and evolution, and it is necessary to deeply implement the "Artificial Intelligence+" action (established strategy), vigorously promote the large-scale commercialization of artificial intelligence applications, and fully leverage China's advantages of a complete industrial system, large market scale, and rich application scenarios (developing around advantages), to accelerate the popularization and deep integration of artificial intelligence in various fields of economic and social development (the goal is to transform traditional industries), forming a virtuous cycle of innovation driving application and application promoting innovation. Government departments and state-owned enterprises should strengthen demonstration leadership (state-owned assets cloud, data governance, data circulation), supporting technology landing through open scenarios. Efforts should be made to optimize the artificial intelligence innovation ecosystem, strengthen the supply of computing power, algorithms, and data, increase policy support, enhance talent team building, construct an open-source and open ecological system (renewing the platform), providing strong support for the growth of the industry. It is necessary to enhance safety capability levels and accelerate the formation of a dynamic, agile, and multi-coordinated artificial intelligence governance pattern 3.2.1 Policy Alignment with Internet+ This round of Artificial Intelligence+ policy is highly aligned with, and even surpasses, the "Guiding Opinions of the State Council on Actively Promoting the 'Internet+' Action" issued 10 years ago. The Internet+ policy was formulated on July 1, 2015, and released on July 4, 2015, with a three-day gap in between. We expect that the full text of the Artificial Intelligence+ policy will be released soon, thereby redirecting the capital market's direction. The content of the full text will become an important information node.

3.2.2 Establishing the Strategic Position of Implementing "Artificial Intelligence+" The action is placed at the strategic forefront, highlighting its programmatic position. Our country will fully leverage the advantages of the industrial system, market scale, and application scenarios to promote the deep integration of AI with traditional industries, with the core demand being industrial upgrading. The government and state-owned enterprises need to play a demonstrative role, balancing both demand and supply; in the infrastructure sector, the demand for state-owned cloud and privatized vertical model deployment will significantly increase. ShenSanda A, as a national state-owned cloud service provider and data processing enterprise within the system, is expected to fully benefit from the policy-driven AI industry. Secondly, the emphasis on building an open-source ecosystem is highlighted, with significance that transcends literal interpretation. Open-source and openness are key paths to breaking technological barriers and promoting inclusive development, which can lower the technological threshold for small and medium-sized enterprises and promote the widespread adoption of AI technology. This orientation aligns closely with the goal of breaking down closed-source ecological barriers and achieving equitable access to technology, demonstrating the policy's strategic emphasis on open collaborative development.

3.2.3 Subsequent policies will catalyze the gradual introduction of detailed rules from various ministries and localities, forming a dense and continuous catalyst. From the documents already released, subsidies and computing power vouchers and other actionable measures will be reflected, intensifying the Artificial Intelligence+ market. Numerous AI applications both domestically and internationally are accelerating in volume, driving a surge in token usage, with the core product's ARR slope steepening, as AI gradually enters the application explosion phase of the second half.

Four: Six Stock Selection Logic:

1. Marginal changes brought by the iteration of large model versions; 2. Top-down — directly policy-related; 3. Bottom-up — event-driven or companies with good fundamentals; 4. Institutional varieties with market capitalization capacity; 5. Mapping to US stocks; 6. Undervalued

1. Pay attention to the changes brought by GPT5 and DeepSeek: Multi-modal and AI programming-related targets GPT5 has been released, and DeepSeek V4 is about to be implemented. The core of this round of GPT-5 release lies in a lower hallucination rate, lower price, and stronger programming capabilities. The logic of large models has shifted from "who is stronger" to "who is more useful": marginal performance improvements are not significant, and empowering the B2B end is necessary to benefit both large models and downstream application vendors. After the release of GPT-5, AI application-related stocks in the US stock market opened with a collective rise, reflecting the market's initial recognition of the AI application logic.

2. Top-down: Policies directly benefit the second stock selection idea, which is closely tied to the full text of the "Opinions on Deepening the Implementation of 'Artificial Intelligence+' Action" and recent policy trends, aiming at targeted investments. Focus on high-quality datasets driven by policies, embodied intelligent brains, and investments in AI + specific industries, including AI + agriculture/industry/new energy/culture and tourism/education, etc.

3. Bottom-up: Event-driven or company AI fundamentals are of good quality. From the perspective of event-driven or company AI fundamentals, select relevant stocks that have event catalysts or smooth progress in AI business implementation. Currently, several companies related to AI applications have preliminarily achieved performance results.

4. Institutional varieties: Consider stock selection strategies from the perspective of market capitalization capacity. From the perspective of market capitalization capacity, large-cap companies (over 10 billion) are more suitable for institutional investment styles.

5. US stock mapping: Reference A-share targets corresponding to the performance of the US stock market. We have sorted out AI-related varieties in the US stock market that have performed relatively well. As mentioned earlier, AI hot stocks have shifted from upstream (such as NVIDIA) to downstream applications. We believe that the A-share companies corresponding to these US AI application stocks also have investment value.

6. Undervalued targets: The overall valuation level of the current AI application sector is relatively low, with a large risk-reward ratio in the future. Starting from low-positioned targets, it is recommended to focus on relevant targets with expected price-to-earnings ratios below 60X in the coming year.

For more investment opportunities in the AI industry, please pay attention to the master class "US-China AI Investment Paradigm: Value Chain Differences and Strategic Choices" jointly launched by Wall Street Insights and Wang Zijing on September 6, 2025. This course includes interactive Q&A, where students can discuss their most concerned issues with teacher Wang Zijing and get answers face-to-face.

Currently, early bird discounts are available for purchases; click on the image above to unlock benefits.

For more course information, please pay attention to the content of the poster below:

Risk Warning: The Master Class is a platform for selected third-party compliance professionals to teach investment research theory courses. The content taught does not constitute buy or sell recommendations for any specific products or investment advice. The opinions expressed in the platform courses are for learning and reference only and do not represent the opinions or views of Wallstreetcn, nor do they address the user's specific investment objectives, financial situation, or needs. The market is volatile and uncertain, and the platform is not responsible for any losses incurred by you relying on the opinions or information from the courses. Investment involves risks; please make decisions cautiously