The technology sector staged a strong rebound today (August 4), with the HUABAO SHANGHAI SCIENCE AND TECHNOLOGY INNOVATION BOARD ARTIFICIAL INTELLIGENCE TRADING OPEN ENDED INDEX SEC (589520), which focuses on the domestic AI industry chain, continuing its upward momentum in the afternoon. Its intraday gain once reached 4.96% and is now up 4.79%, recovering above the 10-day moving average. Data shows the ETF attracted net inflows of 8.88 million yuan on a single day yesterday, and over the past 20 trading days, cumulative net inflows have reached 70.6 million yuan.
Among its constituent stocks, VeriSilicon and UCloud-W led the gains, surging over 12%, while Tianzhun Technology and Juchen Technology rose more than 10%. Other stocks like StarRing-U, Anlu Technology, and Kaipuyun also posted significant gains. Among heavyweight stocks, Cambricon and Haiguang Information rose over 5%, and Montage Technology gained over 6%.
The End of the Price War?
On the news front, on August 3, Alibaba launched its Qwen 3.8-MAX model (featuring 2.4 trillion parameters), marking the first time a MAX-level model has been fully open-sourced. This completes the flagship iteration cycle for all five major domestic large language models within a period of less than 100 days. Benchmark data indicates that the performance gap between models has narrowed to a single point, shifting the competitive focus toward building complete ecosystem loops.
The release of Qwen 3.8-MAX sends a clear signal: the price war may be over. It has abandoned per-token billing in favor of a Token Plan subscription model. Personal plans are offered in three tiers (39, 139, and 499 yuan per month), with usage measured uniformly in Credits. This is coupled with tiered discounts of 0.2x at night and 1x during the day, meaning the actual cost for high-frequency developers is significantly lower than the listed price suggests.
This establishes three distinct pricing strategies for domestic large models: Kimi K3 (priced at 20/100 yuan per million tokens) pursues a high-end capability pricing model; Qwen 3.8-MAX adopts a subscription model, targeting high-frequency, deep-user developers; and DeepSeek V4 Flash (priced at 1/2 yuan) focuses on extreme cost-effectiveness, using peak-valley pricing combined with a 98% cache discount to push single-task costs to the industry's lowest. These three models are no longer directly competing on a per-million-token price basis.
A Historic Summer of Launches
From June to August 2026, this summer witnessed the completion of flagship iterations for all five major domestic large language models. MiniMax M3 kicked off on June 1, followed by Zhipu GLM 5.2 on June 17. Kimi K3 set a global record for the largest open-source weights on July 16, DeepSeek V4 Flash entered the market with extreme cost-efficiency on July 31, and Qwen 3.8-MAX concluded the lineup on August 3. This represents the most concentrated launch window for flagship AI models in the history of China's domestic AI industry.
Value Migration to Infrastructure
With model-level performance converging and the trend toward open-sourcing becoming clear, industry value is migrating from the "model layer" to the "computing infrastructure layer." Regardless of which model ultimately prevails, massive inference computing power will be a rigid requirement. Kimi K3 suspended new user subscriptions on its launch day due to insufficient computing power, and the 2.4-trillion-parameter inference of Qwen 3.8-MAX also places immense pressure on the computing supply chain. As domestic semiconductor production capacity continues to expand and the penetration rate of general-purpose chips increases, there remains considerable room for reducing AI inference costs. The computing infrastructure segment (including AI servers, AI chips, and data centers) is likely the most certain beneficiary direction within the current industry chain.
Domestic AI Champion ETF
The HUABAO SHANGHAI SCIENCE AND TECHNOLOGY INNOVATION BOARD ARTIFICIAL INTELLIGENCE TRADING OPEN ENDED INDEX SEC (589520), along with its feeder funds (A-share: 024560, C-share: 024561), strategically focuses on the domestic AI industry chain. Its constituent stocks include 30 larger market-capitalization companies listed on the STAR Market that provide fundamental resources, technology, and application support for AI. The semiconductor sector accounts for 70.4% of the weight, giving it strong offensive characteristics. The weight of GPU concept stocks and AI application concept stocks is 41.98% and 23.19%, respectively. With a 20% price fluctuation limit, the ETF provides a low-barrier entry point for investors seeking exposure to the breakthroughs in the sci-tech innovation board. Furthermore, this ETF is a margin trading and short-selling target, serving as an efficient tool for a one-click investment in domestic computing power.
Disclaimer: Recent market volatility may be high, and short-term gains or losses do not predict future performance. Investors must make rational investment decisions based on their own capital situation and risk tolerance, paying close attention to position and risk management. The HUABAO SHANGHAI SCIENCE AND TECHNOLOGY INNOVATION BOARD ARTIFICIAL INTELLIGENCE TRADING OPEN ENDED INDEX SEC passively tracks the Shanghai STAR AI Index. The securities mentioned in this article are for display purposes only and do not constitute investment advice. Investing in funds involves risks, and past performance does not guarantee future results.
Comments