How to + AI?

I. The AI Market is Expected to Gradually Expand into More AI+ Fields

DeepSeek accelerates the process of AI democratization, and policy-level encouragement for the large-scale commercialization of artificial intelligence is expected to speed up the domestic AI commercialization process. High computing power costs and the high costs of development and deployment have always been the main constraints on the implementation of AI applications. The release of DeepSeek at the beginning of the year has made significant contributions to reducing costs and increasing efficiency for large models. With the advancement of AI democratization, more innovative AI application scenarios and business models are expected to emerge rapidly.

The policy level also places great importance on the implementation of AI applications. On July 31, the State Council's executive meeting reviewed and approved the "Opinions on Deepening the Implementation of the 'Artificial Intelligence+' Action." From last year's government work report's "launch" to this year's government work report's "continuous promotion," and now to this "in-depth implementation," the national level's emphasis on AI empowering economic and social development in various fields is increasingly heightened.

Looking back at the evolution of the previous "Internet+" market, the market exhibited a rotation pattern from upstream to downstream. More importantly, with the implementation of more downstream applications, the market is no longer limited to the TMT sector but is further expanding into more "Internet+" fields:

• From 2009 to 2010, the mobile internet began to sprout, benefiting upstream hardware, represented by electronics, under the emergence of 3G technology, the launch of the iPhone 4, and the increase in smartphone penetration.
• Since 2013, the mobile internet has entered an accelerated penetration phase, with the industrial focus gradually extending to downstream content areas. The rapid growth in user numbers of downstream applications such as mobile games and WeChat Pay has pushed the market into its first "peak" phase, structurally expanding to the downstream sectors represented by media and computing.
• In 2015, the government work report first mentioned "Internet+," at which point the technology market was no longer limited to the TMT sector but gradually began to evolve towards Internet+, the Internet of Everything. From social media, e-commerce, and gaming to finance, education, and healthcare, the application scenarios of the mobile internet continue to expand and deepen, pushing the market to another "peak." Structurally, in addition to the TMT downstream industrial chain benefiting more, many strong stocks have emerged in traditional fields such as logistics, retail, healthcare, and machinery, with technology empowerment driving a flourishing market.

Therefore, similar to "Internet+," AI, as a general technology, will also accelerate the empowerment of various industries in the future. With the "blossoming" of AI applications, it may also drive the market to extend and expand into more "AI+" fields. 

Since the beginning of this year, with DeepSeek releasing confidence in the implementation of enhanced AI applications, the market has gradually spread to the mid and downstream sectors. However, compared to the previous round of "Internet+", the empowerment of AI in most industries is still in its early stages. With DeepSeek promoting AI equity and strong support from national policies, the commercialization process of AI in China is expected to continue accelerating, driving the market to further spread to the mid and downstream sectors. The AI sector as a whole will also enter a stage with a broader scope, stronger profit effects, and a more solid mainline position.

II. How to grasp AI+?

AI+ is a vast industrial chain. How to grasp the configuration? AI and TMT account for more than 20% of market capitalization, covering a large industrial chain that includes upstream computing hardware, midstream software services, and downstream applications. Among them, the mid and downstream sectors involve more fields and cover a wider range, making industry comparisons even more significant.

First, overseas AI development experience is a good reference. Since the fourth quarter of last year, the US stock market has seen a batch of mature AI mid and downstream application companies represented by AppLovin, Palantir, and Duolingo. This year, more application companies are beginning to realize their performance Referring to the US stock market, the fields where the AI+ model is expected to be the first to run through and where applications are likely to land first are mainly concentrated in cloud services, ERP enterprise management systems, customer management SaaS, financial automation software, data analysis, e-commerce platforms, advertising platforms, and other B2B fields, as well as education, social platforms, gaming, healthcare, and voice in B2C fields. From the recent stock price performance, the areas that the market recognizes more and are likely to form a mapping are mainly concentrated in B2C virtual communities, social platforms, healthcare, gaming, and B2B data analysis, advertising, ERP enterprise management systems, office software, customer management SaaS, and other fields.

Secondly, from the perspective of domestic economic expectations, with breakthroughs in the AI industry and the recovery of domestic demand, many mid- and downstream sub-sectors are expected to see marginal improvements in prosperity by 2025. Many mid- and downstream sectors are expected to enter a phase of turnaround or accelerated prosperity in 2025, including accelerated prosperity (AI wearable devices, AIPC, AI smartphones, humanoid robots) and turnaround (industry application software, SaaS, online education, cybersecurity, basic and general software, e-government, large models, operating systems, gaming).

Thirdly, from the perspective of congestion comparison, the current congestion level in most AI mid- and downstream fields is still at a medium or low level, including midstream software services (office software, basic and general software, operating systems, industrial software, SaaS, AIAgent, etc.), and downstream applications (gaming, autonomous driving, humanoid robots, smart healthcare, etc.).

Finally, from the perspective of chip structure, institutional funds are still relatively low in the allocation ratio in most AI mid- and downstream industrial chain links. The current allocation ratio of active public funds in mid- and downstream sub-sectors at historical lows is mainly concentrated in software (cybersecurity, operating systems, office software), services (cloud computing, edge computing, SaaS), applications (digital marketing, Financial technology, smart healthcare, e-government, online education).

III. Current AI Mainline Focuses on the Spread of Domestic Computing Power and Mid-to-Downstream Applications

Since June, the AI sector has continued to outperform the entire A-share market due to its favorable conditions and risk appetite. However, from the perspective of rolling return differences, trading volume proportions, and crowding indicators, the sector has not yet reached overheating levels:

• In terms of rolling return differences, with the recent overall market rise, the rolling 40-day return difference between TMT and the entire A-share market remains around 5%, and has not risen to the overheating range of over 10%.

• In terms of trading volume proportions, the cyclical sector has performed stronger recently, and the current trading volume proportion of TMT has fallen below 30%, still some distance from the historical overheating range of 40%-45%, and has not issued overheating signals.

• In terms of crowding, most fields are still at a moderately crowded level.

However, there have been two obvious divergences within the sector. On one hand, upstream computing power hardware has begun to outperform midstream software services and downstream applications; on the other hand, within upstream computing power hardware, the North American computing power chain represented by optical modules and PCBs has significantly outperformed domestic computing power chains such as chips

We believe that from multiple dimensions such as stock price position, overseas mapping, policy support, and event catalysis, the AI sector should focus on the diffusion opportunities towards domestic computing power and downstream software applications:

First, looking at the stock price position, since June, the North American computing power chain, including PCB and optical modules, has accumulated a significant excess increase compared to other directions in the industry chain.

Secondly, from the perspective of overseas mapping, the latest performances of Microsoft and META once again confirm the acceleration of AI application commercialization. Recently, overseas AI downstream software applications have begun to outperform upstream computing power, which is expected to create a mapping effect domestically. The rotation between domestic AI upstream and downstream is greatly influenced by overseas mapping. For example, in the fourth quarter of last year, as the performances of downstream companies such as AppLovin, Palantir, and SoundHound AI began to materialize, the trading theme of the overseas AI market started to shift towards downstream, which also led to the diffusion of the domestic AI market towards downstream. Since June this year, the North American computing power chain market, in addition to the inherent prosperity advantages of the sector, has also been significantly influenced by the mapping brought by the continuous outperformance of overseas computing power.

Since mid-July, overseas AI downstream has started to outperform upstream again. The latest performances of Microsoft and META once again confirm the acceleration of AI application commercialization: AI-enabled Microsoft cloud business revenue exceeded expectations with high growth, with FY2025Q4 single-quarter Microsoft cloud revenue up 27% year-on-year; AI intelligent recommendation algorithms drove META's advertising business volume and price to rise simultaneously, with Q2 25 advertising revenue up 21% year-on-year, including a year-on-year increase of 11% in the number of ads and a year-on-year increase of 9% in the average price per ad. The better-than-expected performance of North American tech giants once again confirms the effectiveness of business growth empowered by AI, and has recently created a mapping effect on the stock prices of some domestic AI software application companies As more North American AI giants release their earnings, it is expected to further enhance the market's confidence in the commercialization of AI applications.

Finally, from the perspective of subsequent catalysts, as the earnings disclosure period passes, the technology sector's focus on immediate performance will weaken, and it will begin to price in medium- to long-term industry trends. Coupled with the recent lifting of restrictions on H20 chips, the release of new models like Kimi and Alibaba, the 2025 World Artificial Intelligence Conference, and the August release of GPT-5, these events provide better opportunities for the AI market to expand towards domestic computing power and mid- to downstream applications:

• In terms of domestic computing power, the capital expenditure growth of domestic internet companies remains a long-term trend. With the recent lifting of restrictions on H20 chips and the expectation that Nvidia's new card B30 will complete market introduction in the third quarter, the previous market concerns about internet companies' capital expenditures are expected to continue to weaken. Additionally, the potential listing of domestic chip companies like Suiyuan Technology and Birun Technology in the second half of the year serves as a catalyst, reinforcing the certainty of the industrial chain under the logic of domestic substitution.

• In terms of applications, the recent release of new models like Kimi K2 and Alibaba Tongyi Qwen3 indicates that domestic large models are still rapidly iterating. With the lifting of H20 chip restrictions and major companies increasing their investments in AI again, the progress of domestic AI applications is also expected to accelerate further. Moreover, in the third quarter, there are multiple potential catalysts in the mid- to downstream sectors, including the progress of large models like GPT-5 and DeepSeek R2, as well as Meta's upcoming release of AR glasses in September, all of which are expected to bring new catalysts for AI applications and edge computing.

Author of this article: Zhang Qiyao's team from Xingzheng Strategy, Source: Yao Wang Hou Shi, Original title: "【Xingzheng Strategy Zhang Qiyao Team】AI+ How+?"

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at one's own risk