In the first trading week of August, A-share AI concept stocks shook off the adjustment gloom from July. On August 5th, the computing power sector saw a collective surge, with the Wind AI Computing Power Index rising by 2.82% in a single day, leading multiple stocks to hit the upper limit. Hong Kong stocks in the computing power concept also exploded in tandem. Meanwhile, domestic large models historically dominated the top six in global usage rankings. Coupled with overseas cloud vendors collectively raising capital expenditure guidance and the dense realization of hundred-billion-level computing power orders from listed companies, market confidence in the AI narrative is re-solidifying. This article reviews the background of this round of market activity, analyzing the industrial logic and future focal points behind the rebound of AI concept stocks.
A collective surge in the computing power sector brings a long-awaited broad rally for AI concept stocks
On August 5th, the A-share AI computing power sector experienced a broad rally. The Wind AI Computing Power Index rose by 2.82% that day, significantly outperforming the general market. In terms of individual stocks, Shengshi Technology and Industrial Fan Link both hit the upper limit, Hongjing Technology surged by 12.60%, Xuguang Electronics rose by 9.54%, and HuoJu Electronics rose by 8.09%. The Hong Kong market was equally lively: Hongteng Precision rose by over 17%, Biren Technology rose by nearly 13%, MiniMax rose by over 10%, and Zhipu rose by over 3%. From A-shares to Hong Kong stocks, the computing power concept was the brightest direction in the technology main theme that day.
This round of rebound is not an isolated event. From August 3rd to 4th, signs of concentrated recovery had appeared in the AI sector, with a wide range of artificial intelligence concept stocks rebounding. Nearly 60 stocks hit the upper limit, and sub-sectors such as AI applications, intelligent agents, computing power leasing, and AIGC saw sequential fluctuations, with numerous 20% limit-up occurrences. Low-level stocks that had been continuously adjusting earlier experienced emotional recovery, and the market main theme re-focused on the artificial intelligence industry chain.
Domestic large models dominate global usage rankings, restoring confidence in the AI narrative
The direct catalyst for this round of activity is the concentrated explosion of domestic large models. Recently, domestic large models such as DeepSeek, Seedance, MiniMax, Kimi, and Qwen have intensively released new versions, continuously challenging top overseas models in advanced capabilities, while their general-purpose versions are gaining traction through the open-source ecosystem and cost-effectiveness, leading to continuously increasing usage.
Data released by the artificial intelligence model aggregation platform OpenRouter shows that the top six global large language models in the July usage ranking—Xiaomi MiMo-V2.5, DeepSeek-V4-Flash, Tencent Hy3, MiniMax M3, Zhipu GLM-5.2, and DeepSeek-V4-Pro—are all independently developed open-source large models from China. This marks the first time domestic models have dominated the top six of this ranking, signifying a significant increase in the global influence of the domestic large model army.
After a period of market skepticism regarding the massive returns of AI giants, the intensive release of domestic models, the accelerated implementation of AI applications, and the upward revision of capital expenditure guidance by overseas giants have collectively driven capital back into the AI industry chain, leading to a continuously rising vitality across the entire computing power industry chain.
Overseas giants increase capital expenditure, and the slope of computing power demand rises instead of falling
Looking globally, after this round of earnings season, four technology giants—Microsoft, Alphabet, Meta, and Amazon—all raised their collective capital expenditure guidance for 2026. Specifically, Amazon raised its 2026 capital expenditure guidance from approximately $200 billion to approximately $220 billion; Alphabet raised its full-year capital expenditure guidance from $180 billion to $190 billion to $195 billion to $205 billion. TrendForce's latest research indicates that amid surging demand for AI infrastructure, the total capital expenditure of the top nine global cloud service providers is estimated to grow by approximately 90% in 2026. Several industry insiders suggest that global tech giants' capital spending has not slowed down, and there are signs of a shift from mere investment to "expansion investment—driving cloud business—realizing AI revenue," effectively supporting the current boom in the entire AI industry chain.
CICC believes that capital expenditure by North American cloud service providers is not yet peaking and is expected to continue rising. With the deep penetration of multimodal large models, embodied AI, and AI Agents in industry, AI computing power demand is expanding from the "training side" to the "massive inference side," with the demand slope increasing rather than decreasing.
Listed companies secure dense orders, and computing power's boom is validated with real capital.
The fervor for AI infrastructure remains high, with multiple listed companies recently disclosing large-scale computing power orders, which serve as the most direct validation of the industry chain's vitality.
Bee Assistant: Announced on August 4th that its wholly-owned subsidiary, Ya'an Yunzhi, signed a computing service contract with Company B for a total amount of 4.608 billion RMB, with an estimated annual net profit of 60 million to 72 million RMB. It also signed a computing server procurement agreement with Company A for a total amount of 3.062 billion RMB.
Gaole Co.: Announced on August 3rd that its wholly-owned subsidiary, Harbin Zhichen, signed a computing service contract with a client for a total service value of 3.195 billion RMB (including tax) over a 5-year period, which could add approximately 200 million RMB in revenue for the company in 2026 if successfully implemented.
Xingyun Technology: Disclosed that as of the end of July, the company has long-term framework orders for computing power totaling 15.082 billion RMB, all for long-term agreements exceeding 5 years. In July, three core orders were successively increased in price and volume.
CICC research points out that as large model development shifts from "parameter stacking and Q&A" to Agent capabilities, lower inference costs, and multi-model collaboration, AI invocation is moving from low-frequency Q&A to high-frequency, complex, multi-step tasks. The decline in token prices does not necessarily suppress total computing power demand; combined with the higher token consumption from Agent multi-turn planning and tool calling, inference computing power demand is still expected to rise. Open Securities believes that computing power leasing, with its light asset model of "pay-as-you-go and elastic scaling," is expected to be a core path to fill supply gaps and reduce corporate capital expenditure.
Policy tailwinds blow frequently, opening up investment space for the "15th Five-Year Plan" computing network.
Positive signals are also emerging at the macro level. Jiang Yi, a spokesperson for the National Development and Reform Commission, stated at a press conference that China's progress in independent AI innovation accelerated in the first half of the year, achieving rapid breakthroughs in multiple segments of the entire industry chain: As of the end of June, the national intelligent computing power scale reached 2.8 times that of the same period last year; the global total downloads of domestic large models exceeded 10 billion times, and domestic models are accelerating adaptation with domestic computing power chips; over 120,000 high-quality datasets have been built, accelerating the formation of a virtuous cycle where computing power, models, and data mutually promote each other.
According to calculations by relevant institutions, the construction of the computing network during the "15th Five-Year Plan" period will add 400 billion RMB in direct investment. Considering that computing power construction is mainly driven by corporate investment, this creates vast space for private investment and provides long-term performance imagination for A-share listed companies.
New themes emerge: AI RAN concepts rise like a sudden star. On August 6th, AI RAN (AI Radio Access Network) concept stocks surged following news of Nvidia's urgent search for Chinese AI base station suppliers. Tongyu Communications hit the daily limit up within two minutes of opening, exhibiting a four-day three-board trend. Guangha Communications rose 20% and the group, including Wuhan Fanggui, Benchuan Smart, ShuoBede, Xinke Mobile, and Shenglu Communications, followed suit. The market believes that the deep integration of AI and communication networks is expected to open up new incremental space, resonating with the computing power main theme and becoming an important supplementary direction for AI concept stocks.
Institutional Views: Rebound or Reversal? Performance Verification is Key
Institutional views on the nature of this rally are cautiously optimistic. Zhuang Yang, an analyst from Shenwan Hongyuan Research Institute, stated that the logic behind the technology sector's rebound is mainly threefold: first, a rebound from oversold conditions; second, the July pullback was more due to chip and sentiment factors rather than fundamental disturbances, and the future industry trend of AI is still viewed favorably; and third, domestic support for AI industry development has boosted market confidence and risk appetite with impressive interim results from related companies.
Shen Lingjun, chief financial advisor at Guotai Haitong, analyzed that the technology sector trend has not ended, and the medium-to-long-term logic of the industry chain remains strong. The market may place more emphasis on performance verification subsequently, and true leading technology manufacturers are expected to reach new highs after risks have been fully released. Directions such as electronics, communications, innovative drugs, and power equipment are worth paying attention to.
CICC noted that AI hardware varieties are "shrinking their scope," and the attractiveness of cloud and applications is increasing. In the software industry, the narrative of "AI swallowing software" is gradually being disproven. Whether the sector can move from valuation repair to reversal in the short term still depends on whether AI products can drive the overall revenue growth of software companies to recover. From an allocation perspective, with breakthroughs in domestic component technology and increased gross profit driven by scale, domestic AI computing power—from hardware to chip design, semiconductor equipment, and materials, down to downstream complete machines and switches—is expected to see both volume and price increases. Currently, the domestic chain with a logic of self-reliance and stronger certainty can be prioritized.
Conclusion
From the global usage volume dominance of domestic large models, to the upward revision of capital expenditure guidance by overseas cloud vendors, and to the landing of ten-billion-level computing power orders by listed companies, the positive logic of the entire AI industry chain is being verified from multiple dimensions. For investors, the August market may have moved from the "sentiment repair" stage to the "performance verification" stage. The three main tracks—computing power, applications, and domestic substitution—are worth continuous tracking, while also being vigilant about the volatility risks of high-level stocks and grasping structural opportunities rationally.