Hong Kong's hard-tech sector extended its multi-day rally on Thursday (August 27), with semiconductor and large language model stocks advancing in tandem. Among the constituent stocks of the Hang Seng Tech Information ETF (159131) tracking the Hong Kong hard-tech benchmark, Z.AI, dubbed the "world's first major AI model stock," surged over 12% to lead the gains. Other top-weighted stocks also rallied, with Kingboard Laminates Holdings climbing more than 10%, Kingboard Holdings rising over 7%, and Hua Hong Semiconductor gaining over 4% for a third consecutive session of strength. Semiconductor Manufacturing International Corp and AAC Technologies each advanced more than 2%.
On the news front, on the evening of August 26, Z.AI announced the launch and open-sourcing of its GLM-5.3-Flash model, officially confirming it as the anonymous model "Ox-Alpha" that had dominated overseas developer communities over the past week—dubbed "Niu Lai" (the bull is coming) by Chinese netizens. Notably, the new model's pricing represents a sharp drop compared to its predecessor, while its benchmark scores match Anthropic's Claude Opus 4.8—yet it is priced at just 1/40th of Opus 4.8. Additionally, Z.AI revealed that all online traffic for GLM-5.3-Flash is powered by 100,000 domestically produced chips, marking the first large-scale collective "export" of China's domestic computing chips.
According to CSC Financial, GLM-5.3 has reclaimed the top spot among open-source models in code and agent benchmarks, with post-training serving as the core driver of this capability leap. GLM-5.3 achieved a 31.4% task accuracy on the Z.ai Code Bench High tier, surpassing Claude Opus 4.8's 29.5%, while average output tokens dropped from approximately 120,000 to 50,000. Leading domestic models have essentially reached the level of Opus 4.8 in code and agent capabilities, with the gap to the latest overseas closed-source models potentially narrowing to just a minor version. Given that domestic models typically adopt open weights and low-cost APIs, they will continue to pressure the pricing and developer ecosystems of mid-tier overseas closed-source models.
Shenwan Hongyuan Securities pointed out that semiconductor industry validation is now spreading from foundry operations to the full chain, including equipment and packaging/testing, reinforcing the logic of volume and price increases driven by AI computing demand. As domestic interim earnings reports enter their final push week, three verifications are converging: equipment order upgrades, optical module results exceeding expectations, and packaging/testing structural improvements. Meanwhile, overseas memory giants are shifting real capital toward shareholder returns, confirming the sustainability of cycle profits at the cash flow level. Supported by national policy and AI infrastructure development, the semiconductor industry continues to enjoy high prosperity, with price hikes sweeping the entire supply chain—including silicon wafers, electronic specialty gases, targets, wet electronic chemicals, and packaging materials. Investors may consider positioning around themes such as the domestic memory supply chain, computing ecosystem, and AI application chain.
For investors seeking scarce "pure-blood" hard-tech exposure in Hong Kong with T+0 trading support, the Hang Seng Tech Information ETF (159131) tracks the Hang Seng InfoTech Index, comprising approximately "80% hardware + 20% software." The index heavily weights Hong Kong-listed "semiconductor + electronics + computer software" names, covering 67 hard-tech companies, including the "foundry duo" Semiconductor Manufacturing International Corp and Hua Hong Semiconductor, "domestic AI PC leader" Lenovo Group, and the "domestic large model duo" Z.AI and MiniMax. Notably, the index excludes large-cap internet firms such as Alibaba, Tencent, and Meituan, making it more effective at capturing Hong Kong's AI hard-tech momentum. Additionally, investors without a securities account can access this theme via the off-exchange feeder fund (026755) for one-click exposure.
According to the fund manager's assessment, the Hang Seng Tech Information ETF carries a risk rating of R4 (medium-high risk), suitable for aggressive (C4) and above investors; suitability matching opinions should refer to sales institutions. Recent market volatility may be significant, and short-term gains or losses do not predict future performance—fund investments may incur losses. Investors should make rational decisions based on their own capital status and risk tolerance, with close attention to position and risk management. Data sources include the China Securities Index Company and exchanges; reference institutional views are drawn from CSC Financial's August 18, 2026 report "August AI Monthly" and Shenwan Hongyuan Securities' August 26, 2026 "Morning News." For fund fees, subscription and redemption agents may charge commissions up to 0.5% for the ETF, while on-exchange trading fees follow actual broker charges with no sales service fee. For the feeder fund, subscription fees range from 0.30% below 1 million yuan, 0.20% from 1 to 2 million yuan, and 1,000 yuan per transaction above 2 million yuan. Redemption fees for individual investors are 1.50% within 7 days and 0.00% after 7 days or more; for institutional investors, fees are 1.50% within 7 days, 1.00% from 7 to 30 days, 0.50% from 30 to 180 days, and 0.00% after 180 days, with no sales service fees. Index constituent stocks are shown for demonstration purposes only; individual stock descriptions do not constitute investment advice of any form, nor do they represent the holdings or trading activities of any fund under the manager. As of August 17, 2026, the mentioned index constituent weights are: Lenovo Group 16.625%, Semiconductor Manufacturing International Corp 15.178%, Xiaomi Corporation 13.851%, Hua Hong Semiconductor 7.598%, Kingboard Laminates Holdings 2.755%, AAC Technologies 2.438%, Kingboard Holdings 2.358%, Z.AI 1.204%, and MINIMAX-W 0.242%. Risk disclosure: The Hang Seng Tech Information ETF passively tracks the CSI Hang Seng Information Technology Composite Index, with a base date of November 14, 2014, and a release date of June 23, 2017. Constituent stocks adjust according to index compilation rules, and backtested historical performance does not predict future index performance. Any information appearing in this article (including but not limited to individual stocks, commentary, forecasts, charts, indicators, theories, and any form of expression) is for reference only. Investors are solely responsible for their own investment decisions. Furthermore, any views, analyses, or forecasts in this article do not constitute investment advice to readers of any form, nor do they assume responsibility for direct or indirect losses arising from the use of this content. Fund investments carry risks; past performance does not represent future results, and the performance of other funds managed by the same manager does not guarantee the fund's performance. Please invest with caution.
Comments