Computing Power Cost Reduction and Intelligent Leap: How does Gemma 4 catalyze the AI industry chain? The Growth Enterprise Market Artificial Intelligence ETF (159243) rises over 2%

On April 3, ChiNext Artificial Intelligence (159243) opened higher in the morning session. As of the latest report, it was up 2.19%. Component stocks saw mixed gains and losses. Guangku Technology rose, Tianfu Communication rose by more than 4%. Multiple stocks such as Zhongji Xuchuang, Changliang Technology, and Guanghuan Xinwang led the gains, while Tongniu Information, Yihualu, and Kunlun Wanfang were among the top decliners.

On the news front, Google officially launched Gemma 4 large models, including four variants of Gemma 4 general-purpose models: the efficient 2 billion-parameter version (E2B), the efficient 4 billion-parameter version (E4B), the 26 billion mixture of experts model (MoE), and the 31 billion dense model (31B). Gemma 4 is built specifically for advanced reasoning and agent workflow, achieving an unprecedented “unit-parameter intelligence level.” It can handle complex logic and agent workflows.

Among them, large-parameter models achieve top performance at the same scale: the 31B model is currently ranked third globally among open-source models on the industry-standard Arena AI text leaderboard, while the 26B model ranks sixth. Gemma 4’s performance on the leaderboard even surpasses models that are 20 times larger in its own scale. For developers, this new unit-parameter intelligence level means being able to achieve cutting-edge AI capabilities with only a significantly lower hardware cost.

CITIC Construction Investment believes that, in the AI application industry, investors may focus on investment opportunities under industrial trends. This week, OpenRouter token calls are expected to increase by 11% month over month. Amid rapid growth, Agents’ task delivery capabilities are becoming increasingly strong. In addition to improvements in model capabilities, Agent Harness (structuring model capabilities for application) is an important driver. In 2026, Agents are expected to be quickly rolled out; attention can be paid to Agents’ self-iterative evolution and advancement.

According to information, the “optical module” content within the ChiNext Artificial Intelligence Index exceeds 50%. There are strategic layouts for leading companies such as Yizhongtian, Xin Yisheng, Zhongji Xuchuang, and Tianfu Communication, and compared with similar indices, it demonstrates relatively strong profit elasticity. In the full year of 2025, the index’s overall gain reached 106.35%, significantly outperforming similar indices such as Kechuang AI and CS Artificial Intelligence. Meanwhile, the expected growth rate of net profit attributable to the parent company in 2025 is as high as 127.36%, far exceeding the overall level of the ChiNext Index. The industry conditions and growth potential are expected to form a dual resonance.

Risk warning: Funds involve risks; invest with caution.