Huatai Securities: DeepSeek-V4's long context reduces costs and expands demand.

Huatai Securities research report states that on April 24th, DeepSeek released the V4 series model and paper, with the core change being the low-cost practicality around the 1M token context, systematically upgrading the model architecture, training process, and infrastructure. Huatai Securities believes that the market is likely to understand V4 as "lowering costs and reducing computing power and storage requirements", but the more important marginal change is that after the cost of long context decreases, the usability of complex agents, multi-document analysis, long-term tasks, online learning, and other scenarios can be improved, and the volume of inference calls and storage access frequency is expected to expand. Potential catalysts include domestic computing power Day0 adaptation, follow-up volume release of the Ascend 950 super node, advancement of DeepSeek's multimodal and Engram routes, and industrial development brought by the interoperability of domestic open-source model algorithms.