The preview version of DeepSeek's new series model, DeepSeek-V4, was officially launched and simultaneously open-sourced on April 24. The new version achieves leading performance domestically and within the open-source field in Agent capabilities, world knowledge, and logical reasoning. The model adopts a Mixture of Experts (MoE) architecture, supports a native context of 1 million tokens, increases inference speed by 35 times compared to V3, and reduces usage costs significantly to just 1/70th of GPT-4's.
This model iteration creates a two-way industrial catalyst. On the computing power side, DeepSeek-V4 has completed deep adaptation with domestic chips, driving procurement of domestic AI chips and boosting upstream segments like wafer foundry and advanced packaging. On the application side, the substantial decrease in DeepSeek-V4's inference cost promotes the commercial implementation of scenarios such as cross-document analysis and agent workflows, unlocking growth potential for cloud computing and enterprise-level AI services.
Against this backdrop, two ETFs from E Fund are worth particular attention. The E Fund AI ETF (03489) closely tracks the FTSE Custom Global Artificial Intelligence Select Index, which gained 13.25% in April. It covers local computing power leaders like SMIC and Hua Hong Semiconductor, allocates to core AI application players such as Tencent Holdings and Alibaba, and includes overseas tech giants like NVIDIA, Microsoft, and Broadcom.
The E Fund Asia Semiconductor ETF (03486) focuses on core semiconductor leaders in Asia, encompassing industry benchmarks like TSMC and Hua Hong Semiconductor. Its tracked index surged 26.05% in April, fully benefiting from the dual tailwinds of exploding domestic computing power demand and the ongoing global semiconductor cycle upturn.
Comments