Friday's session saw market consolidation with major A-share indices closing in negative territory, as the Shanghai Composite oscillated within a narrow band while the ChiNext underperformed. Combined turnover across Shanghai, Shenzhen, and Beijing markets exceeded 2.1 trillion yuan, slightly contracting from the prior day's levels.
According to data from the National Oceanic and Atmospheric Administration, the probability of a strong El Nino event this autumn and winter exceeds 90%. Market expectations suggest that anomalous weather patterns such as drought and erratic precipitation will disrupt food supply and lift grain prices, which in turn would boost demand for agricultural inputs like pesticides and fertilizers. The agricultural chemical sector could see a surge in demand, with the underlying indices of the Chemical ETF Huabao (516020) and the Agriculture, Livestock and Fishery ETF Huabao (159275) showing repeated recent activity. Notably, both currently trade at attractive valuations with strong cost-performance characteristics.
Salesforce's better-than-expected earnings powerfully dismantled the "AI eats software" panic narrative, with software stocks briefly leading market gains in early trading. Cybersecurity concepts also delivered standout performances, as Kirin Xinan hit the 20% limit-up. The underlying index of the Software Development ETF Huabao (159036), which targets frontier AI application areas, once climbed over 4%, while the Hong Kong Internet ETF Huabao (513770), heavily weighted toward internet leaders, rose as much as 1.7% intraday.
US technology firms are grappling with the potential imposition of new tariff policies by the White House. According to sources familiar with the matter, the Trump administration is considering a fresh round of comprehensive tariffs on semiconductors. Morgan Stanley chief China economist Robin Xing pointed out that China and the US are pursuing fundamentally different models of AI development, with the balance subtly shifting: American leading tech companies plan substantial increases in computing power investment but face three hard constraints—insufficient electricity supply, local community resistance, and policy regulatory hurdles. Already 14 US states have enacted restrictions on computing infrastructure construction, and over $160 billion in computing projects were shelved last year.
The US AI ecosystem relies primarily on private capital, making large model usage costs prohibitively high. Enterprises are beginning to tier their AI tool adoption, shifting simple reasoning tasks to more cost-effective open-source models. Meanwhile, China is building inclusive computing infrastructure, creating unique competitive advantages. First, cost advantages are pronounced, with domestic large model token usage costs at just one-tenth of comparable US models, suited to SME and industrial deployment needs. Second, a dual-track computing architecture has been established—frontier model training selectively imports high-end chips while inference and application layers widely adopt domestic computing cards, with future reliance on national-level computing centers to scale and amortize hardware costs. Third, infrastructure deployment speed leads globally, drawing on the mobile internet era's experience of affordable, inclusive infrastructure, opening computing resources to domestic AI vendors at parity pricing, compensating for single-chip performance gaps through system integration and computing interconnection to form overall competitiveness.
On positioning, Xing believes AI investment has entered a "half-time intermission," with market logic shifting from single-minded pursuit of computing power, chips, and other upstream hardware toward two new main lines: first, AI application-layer enterprises leveraging large models to reduce costs, improve efficiency, and realize revenue growth; second, AI industry chain resource support (HALO assets), where energy, energy storage, and strategic raw material demand continues expanding with long-term pricing advantages.
In the chemical sector, stocks mounted a counter-trend surge with Haohua Technology hitting the limit-up at close, while Hangyang, Tongkun, and Do-Fluoride each gained over 5%. Hualu Hengsheng, Dongcai Technology, and Hengli Petrochemical also posted notable gains. Strong El Nino events typically disturb food supply through drought in Southeast Asia, India, and Australia, plus abnormal precipitation in South America. China Securities stated that while global grain prices remain relatively low with buffer inventories, weather risks are accumulating rapidly; if production reductions in major growing areas materialize in coming months, supply-demand dynamics could shift from "loose inventories" to "production cuts and destocking," warranting close attention to upward grain price elasticity. Rising grain prices would also improve global planting profitability and farmer input willingness, further driving pesticide and fertilizer demand through replanting, acreage expansion, and increased per-unit investment, benefiting the agricultural chemical cycle.
From a valuation perspective, data shows the sub-index chemical index's price-to-book ratio stood at 2.54x as of yesterday's close, positioned at the 49.02nd percentile over the past decade—a relatively reasonable level offering solid medium-to-long-term cost-performance. Looking ahead, China Securities suggests that while chemical leaders' earnings floors at cycle bottoms are already exceptional with ample value, near-term risks remain from high Q2 comparables and weak Q3 restocking leading to sequential earnings declines. Recommended focus areas include overseas refined oil products maintaining high prosperity, fluorine chemical refrigerants seeing both volume and price gains, potash fertilizers with new capacity ramping up, and tires benefiting from significantly lower input costs.
The software development sector staged a collective recovery with Kirin Xinan and Dingjie Digital Intelligence both hitting 20% limit-ups, while Topsec, Shuìyǒu Shares, and Jiuer Software all reached 10% caps. Foxit Software and 360 touched intraday limit-up levels before closing up 7.06% and 6.65% respectively. The underlying index of the Software Development ETF Huabao (159036), encompassing 107 software development stocks, briefly rose over 3% in early trading before closing 0.69% higher. Strong earnings from Salesforce, CrowdStrike, and other international giants significantly eased concerns about AI disrupting traditional software, driving the global AI software sector higher and opening a valuation repair window for A-share software stocks.
Within AI application subsectors, AI office productivity is poised to become the next deterministic token consumption direction following code generation. Overseas, Anthropic launched Claude Cowork, while OpenAI integrated Codex with ChatGPT expanding into office scenarios. Domestically, Tencent, Alibaba, DeepSeek, and other leading vendors are advancing AI agent products like WorkBuddy. The AI office scenario's vast potential could bring AI to mass markets, making office productivity a platform-level product and new traffic gateway for the next B-end cycle. China Securities noted that AI industry investment logic is shifting from model capability and capex competition toward commercialization validation, with domestic models rapidly narrowing capability gaps in Coding and Agent tasks, and application flywheels closing better than expected—recommending attention to B-end AI application vendors with scenario, data, and enterprise delivery capabilities.
Cybersecurity has drawn market attention amid rapid AI agent development. CrowdStrike's robust growth performance further confirms the "AI threats force security spending" logic. On August 27, OpenAI published an open letter titled "A Collective Call to Action for Cyber Defense," joining over 100 organizations including Anthropic, AWS, Google, and Microsoft in warning that the window for strengthening cyber defenses is limited, recommending all organizations elevate network defense to top leadership priority. Zheshang Securities believes AI will not replace cybersecurity vendors but will instead amplify the rigid demand for security protection, with leading companies possessing full-stack AI security capabilities continuing to widen their technical moats. Overseas leading AI firms are accelerating development of specialized tools for cybersecurity scenarios, while AI-enabled automated cyberattack methods iterate rapidly, forcing government and enterprise organizations to continuously increase security procurement budgets. Domestic cybersecurity firms with AI security products and full-stack protection capabilities will benefit from both AI security incremental growth and domestic substitution dividends.
On valuation, the software development index's price-to-earnings ratio (TTM) stood at 164.78x as of August 27, lower than 78% of its historical trading range since listing, presenting notable valuation cost-effectiveness and safety margins.
Throughout historical technology revolutions, profits have consistently flowed from hardware to applications. Early this year, the "large models devour software" narrative disrupted valuations with AI application directions experiencing significant discounts, making the software development sector a "water level depression" in the AI industry chain. The software sector is currently in an upward cycle, though uncertainty remains over which sub-segments or individual stocks will outperform. The Software Development ETF Huabao (159036) offers full-industry exposure with its 107 constituent stocks comprehensively covering AI+finance, AI+healthcare, AI+office, AI+education, AI+information security, and AI+government affairs.
US stocks gave back intraday gains as AI applications showed active trading while internet leaders mostly strengthened. The CSI Hong Kong Stock Connect Internet Index closed up 0.6%, with Tencent Holdings and Xiaomi Group-W both gaining over 1%. Tencent's Hunyuan released and open-sourced its next-generation large language model Hy4 preview today, featuring significant expansion in model size, context length, and data scale, maintaining its position in the open-source model first tier. Last month, Tencent released Hunyuan Hy3 official version, and WorkBuddy's monthly active users have surpassed 20 million, with base model iteration and application commercialization progress notably accelerating. Xiaomi unveiled three self-developed chips in a single launch—Xuanjie O3 (AI flagship SoC), Xuanjie O100 (high-bandwidth AI acceleration chip), and Xuanjie D100 (autonomous driving high-computing AI chip)—breaking beyond single smartphone chip boundaries to build an AI computing foundation supporting the "human-vehicle-home" ecosystem.
With interim report disclosures, AI application commercialization validation continues boosting sector sentiment. Meitu rose over 7% post-earnings with adjusted net profit attributable to parent of 520 million yuan in H1, up 39.5% year-over-year, while AI productivity application annual recurring revenue reached approximately 620 million yuan. Bilibili-W closed up 6% post-earnings with Q2 adjusted net profit up 25% year-over-year, with advertising serving as the revenue growth engine and AI ad revenue surging over 100%. Meituan-W is scheduled to report earnings after market close, with multiple institutions noting that food delivery subsidy tapering is exceeding expectations, potentially establishing a competitive inflection point and driving accelerated earnings recovery. Additionally, SaaS concept stocks moved higher with Kingdee International up over 6%, as Salesforce's strong results shattered the "AI eats software" narrative, demonstrating software companies can be beneficiaries rather than victims in the AI wave.
National Data Administration figures show domestic daily token calls exceeding 140 trillion, a more than 1000-fold increase from the 100 billion at the start of 2024 and up over 40% from 100 trillion at the end of 2025. Zhongyuan Securities stated that the substantial increase in daily token calls indicates China's AI applications have entered a rapid growth phase. Model vendors continue intensive iteration updates, enhancing capabilities in long-text processing, inference performance, agent functionality, code generation, and multimodal applications, accelerating AI marketing, AI multimodal content, and AI e-commerce applications while advancing commercialization.
Southwest Securities believes the AI hardware segment has entered an earnings validation phase, with application ends gradually becoming a key market focus. As model capabilities improve and enterprise acceptance of AI tools rises, office, industrial, and software service sectors are likely to see increasing commercialization opportunities. The Hong Kong Internet ETF Huabao (513770) passively tracks the CSI Hong Kong Stock Connect Internet Index, with heavyweight holdings including Alibaba-W, Tencent Holdings, and other tech giants alongside AI application companies across various sectors. The top ten constituents account for over 80% of total weight, demonstrating significant leader advantages, supporting intraday T+0 trading with excellent liquidity.
Comments