The investment market in 2025 is like a lively carnival, and DeepSeek concept stocks are undoubtedly the brightest stars among them.

On February 7, the DeepSeek concept rose sharply for the fourth consecutive day, and as of the close, many stocks such as Daily Interactive, Capital Online, and Montnets Technology rose to the limit.

Some friends who don't understand may want to ask, what is so good about DeepSeek? To put it simply, DeepSeek has three major core technology breakthroughs:

The Hybrid Expert Model (MoE) trains a set of expert models for different tasks through the task allocation mechanism, each expert is responsible for a specific domain, and only some of the expert models are activated during task inference, which greatly reduces the cost of AI computing.

Through reinforcement learning (RLHF) + self-developed data generation mechanism, the dependence on external data is bypassed.

Computing power optimization: Breaking through the key technology of chip control, through fine-grained control of the computing process, from stacking high-performance GPUs to optimizing existing computing power.

Driven by the positive benefits of the DeepSeek industry chain, the market showed two rounds of upward movement. The first wave of gains focused on the core links directly related to DeepSeek, including its equity partners, hardware supply chain synergies, etc.; The second wave of the market will spread to the application side, and the market hotspots will gradually differentiate into related enterprises in downstream scenarios.

With the acceleration of the commercialization process of the DeepSeek large model, a number of listed companies have recently disclosed their business progress intensively, claiming that they have realized the integration of their products with the large model, which further confirms the rapid expansion of the application-side ecosystem. Next, we will focus on conceptual stagflation stocks.

Wanma Technology (300698. SZ）

Youka Technology, a wholly-owned subsidiary of Wanma Technology, has fully carried out the adaptation of the full range of DeepSeek models, including R1, and can now experience DeepSeek locally with one click. Youka Changzhou Intelligent Computing Center has deployed a dedicated autonomous driving and large model development tool chain for car companies that meet compliance requirements, and deeply integrated NVIDIA NVAIE products, providing an efficient environment for large model deployment, and DeepSeek has excellent running performance. At present, the relevant functions of DeepSeek have been opened to customers, and a leading car company customer is trying it out.

Wanma Technology said that the DeepSeek large model reduces the training cost of autonomous driving models by 75% through MOE architecture and PTX programming technology, and only needs 2,000 GPUs (1/8 of the same computing power) and 6 million US dollars to complete the training. Combined with knowledge distillation technology, it reduces the cost of urban NOA hardware to 3,000 yuan, promotes the popularization of intelligent driving of 10-150,000 models, and simultaneously reduces inference energy consumption by 80% and responds in milliseconds.

Wondershare Technology(300624. SZ）

Wondershare Technology, a digital creative software company, announced on February 4 that a number of products in its video creativity, drawing creativity and document creative software businesses have completed in-depth adaptation with DeepSeek-R1. Related products include Wondershare Filmora, Edraw Icon, Edraw MindMaster, Wondershare PDF, etc.

Wondershare Technology said that the combination of digital creative software and domestic reasoning models will open up new ideas and directions for the development of creative software. Through deep integration, creative software can make full use of the powerful natural language processing, basic data analysis and logical reasoning capabilities of the new generation of large models, improve the accuracy of Chinese comprehension, further enhance the innovation of creative software in terms of functions, user experience and intelligence, and lead the industry to a more intelligent and efficient future.

Montnets Technology Co., Ltd. (002123. SZ）

Montnets has deeply integrated the DeepSeek large model into the multi-source AI scheduling engine "Tianhui Zhihuitai 2.0", promoting the upgrade of messaging services from basic connection to intelligent decision-making, and opening up a new path for the industry to reduce costs and increase efficiency. After the integration of this large model capability, the intelligent interaction level of Tianhui Zhihui Platform 2.0 will be greatly improved, and breakthroughs will be achieved in various links such as intelligent interaction, data analysis and decision support, multi-modal content generation, and trusted AI system.

Visual China (000681. SZ）

Visual China has recently completed the integration and localization deployment of DeepSeek-R1 and applied its capabilities in multiple products, marking a further upgrade in the application of AI technology in the field of visual content services. By combining the core advantages of "visual data and application scenarios", Visual China is comprehensively promoting the implementation of the AI strategy of "visual data + AI technology + application scenarios", and promoting its innovation and development in multiple application scenarios such as image digital copyright trading, audio and video digital copyright trading, and creative communities.

Dongfang Guoxin (300166. SZ）

At present, Dongfang Guoxin has completed the full integration of DeepSeek and the intelligent digital products of the staff. After testing, the overall performance improvement of the Staff's Intelligent Data platform has been achieved by about 15% after connecting to DeepSeek. At the same time, thanks to DeepSeek's excellent inference capabilities in application scenarios, the work efficiency and user experience of data personnel have also been greatly improved.

Qi Anxin(688561. SH）

Cybersecurity company Qianxin has announced that it has completed a full integration with DeepSeek. Qianxin introduces DeepSeek-R1 into scenarios such as threat analysis, security operations, penetration testing and vulnerability management, identity and access management, phishing protection, malware and ransomware protection, data breach prevention, security training, and supply chain security.

Investment outlook

BOCOM International believes that in the long run, lowering the model threshold or accelerating the deployment of generative AI on the device side and accelerating the upgrading of consumer electronics products is generally good for the end chip and equipment OEMs. The prospect of deploying AI on the device side is more optimistic than before, although the specific number of demand growth and the timing of replacement are uncertain, but software companies that provide services for vertical industries by providing generative AI may be the most direct beneficiaries of the successful development of efficient open source large models.

The emergence of DeepSeek has brought new development opportunities to the AI industry chain, but it is also accompanied by risks such as technology iteration and market competition. Investors should focus on the company's technical strength, market competitiveness and the depth of cooperation with DeepSeek when deploying DeepSeek concept stocks.

At the moment when the wave of AI is surging, high-quality enterprises in the DeepSeek industry chain will undoubtedly become the darlings of the market. Let's pay attention to these good targets, especially those high-quality enterprises that have not yet been fully launched and have low growth rates, seize the investment opportunities and share the dividends of AI development!