At the beginning of August 2026, the AI narrative in A-share and Hong Kong markets is experiencing a subtle power transfer. While the computing power hardware sector is repeatedly digesting transaction crowding at high levels, AI application concept stocks have declared with a sharp eight-day rally that the protagonist of the second half of AI investment may have changed. As of the morning of August 5, Xinkaipu, Desheng Technology,泛微网络, Guangzhou Media, and Tiandaixiu all hit the limit, Zhuoyi Information, Hanxin Technology, and Control Technology rose more than 10%, and the AI application concept index has accumulated a gain of more than 20% since its launch on July 27.
Market Overview: AI Application Index Soars 20% in Eight Days
The starting point of this round of market can be traced to the end of July. At that time, the AI application index was still in a downward channel since its high on January 13, and as of the close on July 31, the cumulative decline still exceeded 30%. However, in the past week, the index violently rebounded by nearly 10%, and on July 31 alone it surged by more than 5%, and market attention to AI application concept stocks suddenly increased. The "Software Yizhongtian" combination composed of Yidi Tianxia, Chinese Online, and Tianlong Group rose collectively for consecutive days, becoming the new popular targets that investors talked about; Hanyi股份, Sinocloud Technology, Tuersi and other software targets also successively showed performances of 20CM or more than 10%.
The software ETF (561010) focusing on AI applications also strengthened, rising 1.89% intraday on August 4. In Hong Kong, on August 6, against the background of overall market adjustment, large model leading stocks moved against the trend - MiniMax once rose 22.77%, reaching a high of 311.60 Hong Kong dollars/share, and Zhipu also rose more than 10% intraday. On the news side, MiniMax-W, Luxshare Precision, and Sanhuan Group were included in the list of Hong Kong Connect eligible securities and took effect on the same day. Coupled with MiniMax's official open source of the H3 full-modal generation model, domestic chip manufacturers such as Haiguang Information, Moore Threads, Brillian Technology, and Muxi Technology also announced completion of adaptation, and the "domestic model + domestic computing power" soft and hard collaborative ecosystem has entered substantive implementation, directly igniting market sentiment.
Performance Verification: Palantir's Report Legitimizes the Application Layer
The core logic driving this round of AI application market is the market's shift from questioning "when AI will bring revenue" to confirming "leading companies have already given answers". Locally on August 3 after hours, AI software giant Palantir released its second quarter financial report: total revenue increased by about 94% year-on-year to 1.94 billion US dollars, and adjusted earnings per share reached 41 US cents, both far exceeding market expectations. What made the bulls even more excited was that the company raised its full-year revenue guidance to 8.15 billion to 8.16 billion US dollars, significantly higher than the previous upper limit of 7.662 billion US dollars, and also far above the analyst average expectation of about 7.7 billion US dollars. After the financial report was released, Palantir's stock price soared nearly 15% after hours, dispelling previous market concerns about AI capital expenditure efficiency.
The sub-item data is more critical. Palantir CEO Karp emphasized that the AI laboratory has not replaced the core business, but has become a growth engine: US government business revenue increased by 90% year-on-year, and US commercial revenue even surged by 149%. This means that AI applications have been upgraded from "experimental expenditure" to "rigid expenditure" in the procurement budgets of governments and enterprises. Microsoft's previous financial report also confirmed this trend - Azure and other cloud services revenue increased by 43% year-on-year, with annualized revenue breaking through 100 billion US dollars, Microsoft 365 Copilot paid users exceeding 30 million, and GitHub Copilot and Azure AI are promoting more and more enterprise customers to embed AI in daily business processes.
Sharp Cost Reduction: OpenAI's Flagship Model API Price Cut by 80%
If performance verification solved the problem of "whether there is demand" for AI applications, then the drastic changes on the cost side solved the problem of "whether it can make money". Recently, OpenAI officially announced an 80% reduction in the API call price for its flagship large model GPT-5.6 Luna. Market analysts generally believe that the "avalanche-like" drop in inference costs directly alleviates the profit pain points of AI software, marketing and other application enterprises, significantly lowers the threshold for application layer commercialization, and thus drives the exponential explosion of the global AI downstream ecosystem and Agent workflow.
Industrial data indirectly confirms this judgment: taking OpenRouter as an example, its API call volume reached a new high in July this year, significantly higher than at the beginning of the year, reflecting that downstream developers and end-user demand has entered a stage of large-scale volume growth. Analysts point out that the unit cost reduction brought by computing power industrialization is being transmitted along the chain of "infrastructure → model services → application ecosystem", and AI applications happen to be the link with the greatest elasticity and the most open valuation space in this transmission chain.
Domestic Strength: Top Five on OpenRouter List Are All Chinese Models
Another highlight of this round of market is the collective breakthrough of domestic large models. The latest weekly call volume list of the global multi-model aggregation platform OpenRouter shows that the top five products are all developed by Chinese companies: Xiaomi MiMo-V2.5 topped the list with 10.5 trillion tokens of call volume in a single week, a month-on-month increase of 12%; DeepSeek's two models ranked second and fifth, forming a high and low match; Tencent Hunyuan 3 ranked third, with a month-on-month increase of more than 999% since its open source on July 6.
At the same time, Moonshot officially released and fully open sourced Kimi K3 - a 2.8 trillion parameter MoE architecture model with native visual understanding capabilities and a 1 million token context window. After its release, domestic and foreign manufacturers quickly announced adaptation; Alibaba launched Qwen3.8 on August 3 and started the "Qwen Office" public test. The open source strategy combined with high cost-effectiveness is helping domestic models shift from "capability catching up" to "ecological competition". Bohai Securities pointed out that Chinese manufacturers have narrowed the model capability to the level of top overseas levels through architectural innovation, and relying on the open source strategy, they are expected to accelerate the overseas process and attract some overseas enterprise customers to switch from closed-source solutions to open-source solutions.
Capital Resonance: Main Positions Increase and Margin Traders Rush to Buy
The sustainability of the market depends on capital participation. From the capital side, a total of 28 AI application concept stocks have received net inflow of main capital exceeding 100 million yuan in the recent period. Among them, BlueFocus ranked first with a main net inflow of 2.419 billion yuan and also achieved a 20CM limit; Shengyi Technology and Kunlun Wanwei both had main net inflows exceeding 1 billion yuan, and Kunlun Wanwei also closed at a 20CM limit. In terms of leverage capital, Hikvision has received a net purchase of 256 million yuan from margin financing since July, Giant Network, DeepinFu, Youzu Network, BlueFocus, and others have net purchases of margin financing exceeding 100 million, and Zhangyue Technology, Yonyou Network, Kingsoft Office, etc. also rank high in net margin purchases.
It is worth noting that brokerage research institutions are turning their attention to "August Golden Stocks". As of early August, the software and service industry has already included Kunlun Wanwei (recommended twice), Fengzhoushou, Century Huatong, Zhipu and other stocks in the monthly golden stock portfolio. At the performance level, Century Huatong announced that it expects a net profit of 4.3 billion to 4.77 billion yuan in the first half of the year, a year-on-year increase of 61.87% to 79.57%, providing another footnote for the performance realization of the AI application sector.
Institutional Perspective: AI Trading Logic Shifts from Computing Power to ROI Verification
From a more macro perspective, many institutions believe that this round of hot spot switching is not short-term speculation, but a structural migration of the main AI investment line. Morgan Stanley China Chief Economist Xiang Ziqiang recently said that AI investment has entered a "half-time adjustment", and the logic is shifting from chasing computing power chips upstream to two new main lines: AI application end (cost reduction and revenue realization) and HALO resource support (energy, energy storage, raw materials). CIT Securities judges that AI has entered the "midfield time", hardware varieties are beginning to "shrink circles", and cloud and application attractiveness is increasing; the sector is likely to enter a stage of waiting for a new narrative and rebalancing of hardware and software. China Galaxy Securities also pointed out that the market valuation logic is shifting from capital expenditure expansion to ROI return verification. As domestic large model capabilities approach the global mainstream first echelon, domestic AI applications and Agents will enter a stage of blooming, and the performance inflection points of office, government and other first-landing scenarios have arrived.
DeFAI Enlightenment: The Application Layer Explosion Logic Also Applies to Encrypted Finance
For investors in the encrypted AI field, the performance path of A-share AI application market is quite referential. From computing power arms race to application value realization, this capital migration logic is being staged simultaneously in the DeFAI track where DeFi and AI are deeply integrated: when the calling costs of basic models and on-chain data continue to decline, AI-driven intelligent risk control, credit assessment and automated lending protocols have the prerequisites for large-scale commercialization - just as Palantir has verified the rigid demand for enterprise-level AI procurement, DeFAI protocols are also using the improvement of on-chain credit scale and risk control efficiency to prove the "cost reduction and revenue realization" capability of AI in financial scenarios. In other words, the shift of hot spots from hardware to applications in AI concept stocks, and the migration of capital from computing power narrative to DeFAI application narrative in the encrypted market, are essentially mirror projections of the same industrial cycle in different asset categories.
However, the strategy team of China Merchants Securities also reminded that the core contradiction of current AI application commercialization lies in: user trial enthusiasm is high, but stable payment scenarios and enterprise ROI still need further verification. After the short-term carnival, whether the market can move from "theme-driven" to "performance-driven" still depends on whether each company can convert model capabilities into sustainable cash flow - this point is equally applicable to AI concept stocks and the DeFAI track. (This article is compiled based on public information and does not constitute investment advice. The market has risks, and investment requires caution.)
