On August 6, 2026, a report from Caixin triggered market attention: against the backdrop of high-low switching in the AI sector, the holdings and capital intentions of public funds are gradually diffusing toward AI application areas with low stock prices. As stocks related to software, the internet, and enterprise digitalization experience a continuous recovery trend, the investment logic of public funds has quietly changed—shifting from speculation in upstream hard technology to focusing on application-end enterprises with commercialization capabilities.
This signal, combined with the intensive research reports recently released by multiple leading institutions, is providing new answers to the core questions: "Why invest in AI, and where to invest in AI in the second half of the year." The second half of AI investment is characterized by "applications" and "realization."
High-Low Switching: Capital Flows from Compute "Hardcore" to Application "Soft Landing"
Looking back at July, global capital markets experienced a period of intense volatility. The monthly declines for the ChiNext Index and the STAR 50 Index, which have high technological content in A-shares, reached 23% and 25.9% respectively, while the Philadelphia Semiconductor Index fell by over 20% cumulatively. The Korean stock market also triggered circuit breakers multiple times under leveraged capital panic. The deep adjustment in the AI direction is due to a multi-resonance of crowded trading, leveraged clean-up, and global semiconductor fluctuations.
However, the adjustment has not changed the industry trend; instead, it has accelerated the rebalancing of capital structure. Entering August, AI application-related stocks such as software, the internet, and enterprise digitalization led the way in a recovery trend, and the preferred holdings of public funds have consequently diffused toward these areas. Market participants point out that this is essentially the inevitable result of AI investment switching from "expectation-driven" to "performance-driven"—after the valuation of upstream compute has been fully priced, capital begins to look for the next "undervalued" value sink, which is application-layer enterprises capable of converting AI technology into actual revenue and profit.
Institutions Speak Up: AI Enters "Mid-Season Rest," Two New Main Tracks Emerge
Multiple leading institutions have reached a high degree of consensus regarding AI investment recently. Xing Ziqiang, Chief Economist at Morgan Stanley China, stated in a media conference that the recent volatility in the AI sector is not due to deteriorating fundamentals, but rather the resonance of multiple factors, such as crowded trading, capital withdrawal from large companies, and rising interest rate expectations due to oil price increases. He judged that AI investment has entered a "mid-season rest" phase, and the investment logic is shifting from chasing upstream compute chips to two new main tracks: first, AI applications (cost reduction and realization); and second, resource segments such as energy, energy storage, and raw materials that support AI compute expansion.
CICC in its research report released on August 5 also pointed out that the large-scale pullback in the global technology sector over the past month is not an essential change in the vitality of the AI industry, nor have any companies downgraded their earnings forecasts; the fundamentals remain high and tightly balanced. The triggers for the pullback are more concentrated negative narratives, such as concerns about price increases, doubts about the return on capital expenditure, the disruption of the model landscape and Token price expectations by domestic open-source models, and interest rate hike expectations, compounded by over-crowded hardware trading and positions.
Looking ahead to August, CICC believes the sector is likely to enter a phase of "awaiting new narratives and rebalancing of hardware and software," and suggests focusing on three directions: first, leading internet companies and cloud vendors with rich AI application ecosystems whose valuations are still low; second, domestic equipment and wafer manufacturing whose vitality is still upward; and third, the optical communication sector and some application companies with good mid-year earnings whose valuations are low.
Open-Source Models Break Through: "Intelligent Parity" and "Token Parity" Reshape Industry Logic The profound force driving the shift in AI investment logic from hardware to applications is causing a dramatic change in the competitive landscape of the model layer. Latest data shows that open-source models developed in China account for 41% of the global downloads, surpassing the US to rank number one globally. Furthermore, all six models in the top six on the global mainstream large model calling charts originate from Chinese teams. Over the past 12 months, domestic models have maintained the scale ceiling of global open-source models for nine months, and the cumulative downloads of open-source models from China have exceeded 10 billion.
The significant improvement in open-source model capabilities has brought about two major trends: "intelligent parity" and "token equalization." On one hand, enterprises no longer need to pay high costs to use top-tier models, and the calling price of large models continues to decline, significantly lowering the barrier to accessing AI capabilities. On the other hand, secondary development and industry fine-tuning based on open-source models have become possible, providing vast innovation space for SMEs and developers, and directly driving AI applications from "demonstration" to "large-scale implementation."
Some analysts point out that the AI industry over the past two years followed a typical "shovel seller" logic—the shovel seller (computing hardware) made money regardless of the success or failure of the gold miners. However, as the underlying computing infrastructure gradually improves entering the second half of 2026, the industry's golden track is fully migrating from "hardware dominance" to "downstream application diffusion." Computing power is the starting point; applications are the destination; only when the commercial closed loop in the application end is truly established can the long-term value of the AI industry be verified.
The Performance Verification Period: From "Blind Buying" to "Informed Selection"
It is worth noting that the shift in AI investment logic is being substantiated by fundamental data. Intermediary reports recently disclosed show that some AI application companies have delivered impressive results: for example, Dipu Technology in Hong Kong reported a year-on-year revenue increase of 115% in the first half of the year, with AI business revenue increasing by approximately 209%, and a near 50% narrowing of year-on-year losses. Its revenue significantly increased after upgrading its enterprise AI solutions to an AI-grade enterprise operating system platform. Concurrently, some companies recorded expanding losses due to poor gross margins in AI businesses and credit impairment—which precisely illustrates that the divergence in AI application investment is intensifying, and the era of universal gains just by being "AI-related" is over.
Multiple leading private equity firms have expressed similar views during the review of this round of adjustments. Danshuiquan Investment believes that the AI-led technology growth trend has not ended; the adjustment of related assets is essentially a correction for the stock price rising too fast and deviating excessively from actual industry progress, not an reversal of industry trends. Ding Ying, founder of Command Capital, stated frankly that the era when the AI industry was "easiest to make money" may be over, but the era when it can "make the most money" is just beginning—the core judgment that AI is a trend-level industrial wave remains unchanged; what has changed is the investment method, moving from "blind buying" to "informed selection."
Yude Investment judges that AI investment has entered a period of divergence, but the industry trend has not changed: the demand for AI computing power continues to exceed supply capacity, and capital expenditure guidance has not shown a substantial downward revision, but rather a continuous upward revision, confirming that this adjustment is short-term market volatility, not a reversal of AI industry fundamentals.
Conclusion: Why Invest in AI? The Answer is Shifting from "Story" to "Financial Statements"
Returning to the fundamental question of why to invest in AI, the latest portfolio rebalancing directions from public funds offer highly valuable commentary: the appeal of AI investment is no longer solely derived from exciting technological narratives, but is increasingly built upon verifiable commercial returns. As the cost of large model inference decreases, the open-source ecosystem flourishes, and application scenarios extend from text generation to enterprise software, smart manufacturing, healthcare, and finance, AI is evolving from a "concept track" into a "real industry track." For individual investors, the keywords for AI investment in the second half of the year should be "structure" and "realization": structurally, pay attention to the rebalancing opportunities as the focus shifts from computing hardware to application software and cloud services; in terms of realization, closely monitor quantifiable indicators such as interim earnings, order execution, and customer willingness to pay. The long-term trend of AI has not changed; only the way to profit has changed—only by selecting the right direction, choosing the right targets, and holding patiently can one share the true dividends of this era-defining industrial wave.