According to Zhitong Finance APP, Soochow Securities Company Limited released a research report stating that the market's main theme from June to August has revolved around artificial intelligence. Currently, the rally is primarily concentrated in upstream infrastructure hardware, with overseas supply chain optical modules and PCBs showing the highest certainty in prosperity, launching first in June. In mid-August, domestic computing power centered on Cambricon officially began catching up. Given the abundant market volume and absence of obvious flaws in industrial logic, they do not believe the computing power rally is about to end. However, for off-field cash holders, the strong profit-making effect of upstream hardware creates restlessness, while objectively, the continuous accumulation of profit-taking pressure brings certain selling pressure. Subjectively, the continuously accumulated gains inevitably cause anxiety among funds with lower risk appetite.
In this AI rally, the core reason for downstream application stagnation lies in insufficient short-term certainty - neither breakthrough explosive products with viral effects nor smooth business models have emerged. At the listed company level, this translates to insufficient earnings visibility, making them not the first choice for funds in this rally. However, from the perspective of technology wave evolution, the ultimate goal of AI empowering everything must be realized through the application end, meaning the explosion of the application end has medium-term certainty and broader space than upstream hardware. This was already verified in the "Internet+" wave and corresponding stock market rally 10 years ago, which also means the launch of AI application rallies is only a matter of time.
In the current market environment with abundant volume, with the main theme concentrated on AI-centric technology, if upstream hardware experiences chip loosening (such as core targets adjusting around 20%), then only marginal events sufficient to attract market attention are needed as catalysts (such as H20 security issues and DS model FP8 making domestic computing power rally move from dark to light), AI's internal low-position branches will show strong elasticity. However, joining on the right side at that time will lose part of the odds, facing the same choice dilemma as now regarding "whether to chase computing power gains." Therefore, it is recommended to view downstream application directions such as AI+ innovative drugs, AI+ military, AIGC, edge AI, humanoid robots, and intelligent driving as "call options" based on medium-term industrial logic certainty, and position aggressively on the left side.
Soochow Securities Company Limited's main viewpoints are as follows:
**The ultimate goal of technology waves is definitely to empower everything, which was already proven in the "Internet+" era. From an industrial trend evolution perspective, the rise of AI applications has inevitability, and the second half of the AI rally will revolve around the application end**
Taking the "Internet+" industrial wave from 10 years ago as an example, with upstream infrastructure construction and end-side interactive entrance-level hardware technology progress and penetration rate improvement, downstream application explosion is an inevitable trend from the ultimate perspective, but when and in what form it appears requires post-verification observation. Additionally, from both temporal and spatial dimensions, the sustainability of downstream application rallies is stronger than upstream hardware.
From the time dimension, the upstream hardware main rally was the consumer electronics rally around Apple supply chain from June 2009 to December 2010, lasting about 1.5 years. Although the server direction followed applications in 2013-2015, it was mainly point-like rallies, while the downstream application rally launched in January 2013 and didn't peak until June 2015, lasting two and a half years.
From the space dimension, during the mobile internet rally from January 5, 2009 to June 12, 2015, among TMT sectors, computer and media sectors showed greater upward elasticity, with cumulative gains of 1039% and 710% respectively, higher than the hardware-side electronics and communications sectors' gains of 697% and 300%.
Specifically:
2009-2012: Mobile internet began to take shape, 3G penetration continued to improve, iPhone 4 launch drove global smartphone volume, with leading directions mainly being hardware represented by electronics sectors, trading logic being prosperity growth. Additionally, the application side also saw explosive products represented by Sina Weibo, but commercialization models couldn't be established, internet applications were still in thematic investment stage.
From 2013: Accompanying 4G acceleration and fee reduction plus smartphone penetration reaching high levels, mobile internet rallies extended to downstream. In 2013, WeChat launched payment functions, and "I Am MT" created the "free game + in-app purchases" model that broke through mobile game monetization bottlenecks. Mobile payment, mobile games and other explosive applications saw user numbers surge dramatically during this stage, pushing rallies to switch from upstream hardware to downstream applications represented by media and computers.
In March 2015, the government work report first mentioned the "Internet+" concept, and in July of the same year, the State Council issued "Guidance on Actively Promoting 'Internet+' Actions." Under top-level design promotion, downstream application scenarios continued expanding. The "Internet+" concept continuously penetrated service industries and mid-stream manufacturing, fully expanding rally breadth. Combined with the 2015 liquidity bull market boost, "Internet+" became the absolute main theme of the market at that time.
Additionally, application-side "shovel sellers" servers gradually outperformed smartphones to become the hardware-side leading main theme during this stage.
Looking at software applications alone, the rally can be roughly divided into two stages: "Internet+ narrative universal rally" from 2013-2015 and "winner-takes-all leading company rally" from 2016-2017.
2013-2015: Accompanying "Internet+" penetration across all industries, emerging business models and applications emerged continuously, all with imaginative space. At this time, vertical markets were still in the land-grabbing stage, pattern issues hadn't yet appeared, combined with systematic valuation improvements under abundant liquidity background, software applications showed a "touch-and-rise" situation.
2016-2017: Mainly structural rallies, on one hand internet traffic early dividends showed signs of decline. According to China Internet Network Information Center data, China's internet penetration rates in 2016 and 2017 were 53.2% and 55.8% respectively, with growth slopes slower than 2010-2015 scale growth. Mobile internet gradually shifted to stock markets, leading manufacturers continuously expanded market share through capital, technology, user resource advantages, significantly compressing small manufacturer survival space. On the other hand, after the "water buffalo" rally peaked in June 2015, significant valuation regression occurred, market risk appetite clearly declined, requirements for company performance realization ability increased. In the internet field with very significant Matthew effects, leading manufacturers had higher earnings visibility. Tencent and Alibaba achieved high revenue and profit growth in 2016-2017, stock prices correspondingly continued upward. From January 1, 2016 to December 31, 2017, Hong Kong stock Tencent and US stock Alibaba gains were 167% and 112% respectively.
In comparison, what we expect to see next is the narrative universal rally stage of the "AI+ rally." On one hand, AI application field growth slopes haven't slowed yet, Token usage and AI software user numbers are still accelerating. The National Data Administration announced on August 14 that as of June 2025, China's daily Token consumption exceeded 30 trillion, growing over 300 times since early 2024. QuestMobile data also shows that as of March 2025, AI native app active users reached 270 million, up 536.8% year-over-year.
On the other hand, explosive AI applications and clear efficient commercialization models haven't yet appeared domestically, and market understanding of AI application ends is insufficient. Before "leading company victory," the application side is expected to first see "hundred flowers blooming" rallies.
**China's AI application end achieving qualitative landing has multiple factor backing including policy and resource endowments**
The central government's newly issued "Artificial Intelligence+ Action Opinion" can be compared at policy level to 2015's "Internet+ Action Opinion." Top-level design clearly defines "Artificial Intelligence+" development goals, with specific implementation measures from functional departments expected to follow, giving AI downstream applications clear development and landing "deadlines."
In August 2025, the State Council issued "Opinions on Deeply Implementing 'Artificial Intelligence+' Actions," clearly stating that by 2027 and 2030, China strives to achieve new generation intelligent terminal and intelligent agent penetration rates of 70% and 90% respectively. By 2035, China will fully enter new stages of intelligent economy and intelligent society development, emphasizing broad and deep integration of artificial intelligence with six key areas: science and technology, industrial development, consumption quality improvement, people's livelihood and welfare, governance capabilities, and global cooperation.
Previously, "Artificial Intelligence+" was written into government work reports twice in 2024 and 2025, but setting specific development goals and implementation rhythms is the first time. More local AI support measures will follow, and downstream application development is expected to accelerate.
China has natural soil for vigorous AI application development, with engineer dividends and huge user groups being the foundation for downstream application quantitative to qualitative changes. From hardware applications, AI terminal hardware's technology-intensive attributes will be further strengthened compared to previous two technology cycles, while China's unique engineer dividend provides stronger high-end manufacturing capabilities than newly industrialized countries and lower cost advantages than North America. Taking humanoid robots, AI's largest-scale terminal application scenario as an example, Tesla Optimus's mass production optimal choice will still be Chinese suppliers.
From software perspective, China's user scale advantages are evident. According to QuestMobile's "2025 AI Application Market Semi-Annual Report," as of June 30, 2025, mobile AI application monthly active user scale reached 680 million. China's huge user group not only forms enormous potential market demand and catalyzes more diverse application scenarios, but also provides more diversified massive data to assist model training and vertical application capability improvement, empowering AIGC industry chain development.
**Insights from this domestic chip rally: As long as industrial logic has certainty, rally launch is only a matter of time**
Fundamental visibility affects sector priority in investors' minds. Branches with slightly weaker immediate prosperity may lag in performance sequence. As long as overall market volume is abundant and AI rallies continue, directions with industrial trend logic will launch rallies - it's only a matter of time. This domestic computing power rally is a typical example.
The reason overseas computing power hardware became the vanguard of this rally lies in the most solid immediate performance, high-visibility medium-term prosperity, and unfalsifiable future growth. In comparison, domestic computing power, edge-side and software applications are "somewhat lacking" in these dimensions.
Looking back, this overseas computing power chain rally started in late May with relatively long duration. The core was that immediate and medium-term prosperity of computing communication like PCBs and optical modules was continuously verified and revised upward by domestic manufacturer performance like O-Net and overseas CSP major capital expenditure guidance. However, during the two-month period from June to early August, domestic computing power sector performance remained "lukewarm."
But investors familiar with technology sectors clearly understand that regardless of whether NV computing cards are opened for China export, achieving chip self-sufficiency is a survival right issue in the AI era. National Big Fund Phase III, sci-tech innovation board creation layer establishment, etc., all indicate upper-level guidance for capital and resource support for "technology self-reliance and self-improvement." That is, the medium-term logic of domestic chips and upstream manufacturing (wafer foundry, equipment, materials, etc.) is very certain, just lacking landmark events to attract more capital attention and catalysts for rally launches.
On August 8, SMIC's Q3 performance guidance slightly fell short of market optimistic expectations, encountering significant selling pressure. This shows that even on the eve of domestic computing power and semiconductor rally official launches, sectors were in states lacking profit-making effects. But simultaneously, based on strong medium-term certainty and abundant overall market liquidity, sector bottom centers were slowly rising. Still taking SMIC as an example, after August 8 mispricing, funds actively bought dips with quick rally recovery.
Until August 10, when CCTV's media Jade Pool Tan Tian published "How America Installs 'Backdoors' in Chips" revealing H20 as "neither environmentally friendly, nor advanced, nor secure," followed by Cambricon's 20cm limit up on August 12, domestic computing power lines officially moved "from dark to light."
On August 21, DeepSeek V3.1 used FP8 parameter architecture enhancing domestic chip compatibility. On August 27, Financial Times reported Chinese chip manufacturers plan to triple AI processor production capacity next year. Bull narratives continuously accumulated to strengthen sector confidence, with Sci-Tech Innovation 50 bursting with enormous upward elasticity: August 1-27, Sci-Tech Innovation 50 rose over 20%, significantly outperforming CSI 300 and ChiNext indices. Cambricon, as domestic chip leader, gained over 90% during this period.
The transformation of domestic chips and semiconductors from "ignored" to market main theme is essentially a process of building market consensus and accumulating bullish momentum. But if chasing gains after "two positive lines," inevitable odds losses occur. If entering later lacking profit cushion protection, holding mentality will also be affected.
With medium-term industrial trends having certainty, it's impossible to accurately predict when important catalysts will occur to reverse fund attitudes. But when overall market liquidity is abundant, sector downside space is also very limited. At this time, the better strategy is based on odds thinking, treating relatively low-position sectors with industrial logic certainty as "call options" for left-side positioning.
Currently in AI rallies, downstream application ends clearly lag behind upstream hardware, representing potential allocation directions with odds advantages.
**Launch is only a matter of time**
Since the overall market formed a "golden pit" under April 7 tariff turmoil, using Sci-Tech Innovation ChiNext 50 as benchmark, only upstream hardware outperformed, consumer electronics and robotics sectors followed, with software application sectors showing the most lagging gains.
From April 7 to August 26 interval gains and losses, upstream hardware optical modules/PCB/high-speed copper connections/server sectors outperformed benchmarks by 67.8/17.3/14.6/6.3pct respectively. Mid-downstream software and application ends only saw gaming sectors outperform benchmarks by 14.3pct, while other AIGC/intelligent agents/cloud computing/humanoid robots/AI wearables/intelligent driving/e-commerce sectors underperformed benchmarks by 13.1/12.5/16.5/13.3/15.1/19.4/14.7pct respectively.
Looking at domestic software application directions, insufficient model capabilities lead to limited vertical application abilities, with markets not yet seeing earnings realization possibilities and imaginative space from explosive applications in software application sectors. From "overseas mapping" perspectives, currently US stock AI software applications also only show individual company single-point performance like Palantir and AppLovin, without large-area rallies. The core remains waiting for industrial development "inflection points," which is also one factor many investors worry about. Therefore, AI software application sector stagnation is most obvious.
Based on previous analysis, application landing is inevitable result of technology waves. Before leading companies emerge victorious, there will first be comprehensive rallies based on industrial logic narratives. Trigger points might be breakthrough progress in domestic foundation model capabilities and sudden token call volume increases, steep rises in application monthly active users/rankings, specific "AI+" policies providing concrete R&D or landing subsidies, etc.
In fact, under benign "slow bull" patterns, excess within sectors is difficult to infinitely expand. Some cash holders tend to "open new tables" at low-position branches due to fear of heights, while some holders' profit-taking mentalities expand with profit accumulation. Subsequently, if high-position upstream hardware experiences chip loosening (volatility/adjustments), liquidity spillover will also help improve low-position branch upward elasticity.
For funds that missed upstream hardware, there are subsequent motivations to position at low levels in downstream application directions with low short-term earnings visibility but medium-long term ultimate certainty and current position cost-effectiveness.
Based on odds thinking, it's recommended to actively position in downstream application direction investment opportunities in AI+ innovative drugs, AI+ military, AIGC, media gaming, AI edge-side, humanoid robots, and intelligent driving sectors.
Based on this, core recommendations for the following AI application directions (including hardware and software applications):
**AI+ Innovative Drugs**: AI applications in pharmaceutical fields are expected to significantly reduce drug discovery costs and time cycles, accelerate target development and validation processes, while reducing initial trial failure risks through simulated clinical trials.
**AI+ Military**: Artificial intelligence empowers military informatization construction, effectively integrating multi-source intelligence data from satellites, radar, drones, etc. in real-time, constructing comprehensive and precise battlefield situation maps, revolutionizing command systems. Machine wolves, robot dogs and other unmanned equipment and autonomous combat systems are another key area of AI+ military.
**AIGC**: Complete ultimate narratives, but short-term earnings visibility is low due to explosive application launches still awaiting time. Subsequently focus on domestic large model capability upgrades, AI Agent industrial progress and other catalysts.
**Humanoid Robots**: AI's largest-scale terminal application scenario, domestic robot manufacturers gradually moving toward order verification stages. Key focus on Tesla Optimus V3 new blueprint updates.
**Consumer Electronics**: Dense new product launches after September, key focus on September 10 Apple consumer electronics new product launch and Meta AI glasses product feedback.
**Intelligent Driving, Vehicle-Road-Cloud**: VLA technology paradigms reshape automaker patterns, cloud-vehicle collaboration competition enters white-hot stages, also important branch of edge AI, but short-term elasticity relatively insufficient due to automaker competitive pattern impacts.
**AI+ Others**: AI+ finance, AI+ agriculture, AI+ logistics, AI+ legal, AI+ government affairs, AI+ e-commerce, AI+ programming, etc.
**Risk Warnings**: Domestic economic recovery speed below expectations; Fed rate cuts below expectations; macro policy strength below expectations; technology innovation below expectations; geopolitical risks.