Reporter Gu Mengxuan and Xia Xin report from Guangzhou and Beijing

At the beginning of the new year of 2025, DeepSeek-R1, an open-source model of Chinese AI company DeepSeek, has set off huge waves in the capital market. Wind data shows that as of February 12, the Wind Deepseek index has soared 62.85% this year, and it is intensifying. At the same time, a number of AI-themed ETFs have achieved amazing returns during the year, with the highest returns exceeding 500%.

Deepseek was born, which attracted huge attention from the global market and injected a "stimulant" into China's scientific and technological development. Peng Xinchang, manager of the mixed fund of China Commercial Innovative Medical and China Commercial Industrial Upgrading, said in an interview with the reporter of China Business News that the birth of DeepSeek has three important significance. First of all, from the national strategic level, it is a highly original innovation achievement in China. In the context of the current fierce competition in science and technology, it has injected strong confidence into China's scientific and technological development and effectively enhanced national pride and strategic confidence. Second, DeepSeek is completely open source, which allows businesses and developers up and down the chain to benefit from it. Based on this, they can carry out various innovative applications, become active contributors to their ecological construction, and promote the diversification and prosperity of model applications. In addition, it supports fully private deployment, which is a great benefit for enterprises with private data, which can improve production efficiency and optimize business processes with the powerful capabilities of large models while ensuring data security.

A catfish stirring up the global model market

DeepSeek was founded in July 2023 by High-Flyer Quantitative, a well-known quantitative asset management giant, and its head, Liang Wenfeng, is the founder of DeepSeek. At the end of 2024, DeepSeek has released three models: V3, R1, and Janus, which have attracted widespread attention at home and abroad.

According to the data of the AI product list cited by the Zheshang Securities Research Report on February 3, DeepSeek App has more than 20 million daily active users in the 20 days since it was launched. According to Appfigures data (excluding third-party app stores in China), DeepSeekApp topped the global download chart of Apple's AppStore on January 26. According to SensorTower's research, the app topped the Google Play Store's US download charts. According to SensorTower, DeepSeek has accumulated 16 million downloads in the first 18 days of its release.

Liang Wenfeng, the founder of Deepseek, once said in an exclusive interview with Undercurrent that in this wave, our starting point is not to take the opportunity to make a fortune, but to go to the forefront of technology and promote the development of the entire ecosystem.

Talking about the strong market influence generated by Deepseek, Xiao Ruijin, deputy investment director and fund manager of Bosera Fund Equity Investment Department IV, said that the DeepSeek large model is a local scientific research team in the context of limited computing power, using a variety of original technological innovations, to achieve a significant reduction in the cost of training and inference, and at the same time, the comprehensive performance of the overseas cutting-edge large model, has become a new breakthrough in China's independent innovation of science and technology.

Xiao Ruijin pointed out that the emergence of DeepSeek large model technology is of great significance, which shows that the development of China's local artificial intelligence industry has gradually caught up with the level of cutting-edge technology overseas, and has the ability to put forward original and leading technological innovation ideas. From an investment perspective, the DeepSeek model has significantly boosted the confidence of domestic investors and significantly increased the attention of global investors to China's technology sector. DeepSeek is a catfish that stirs up the global model market, bringing a triple impact of performance, price, and open source.

The research report of Zheshang Securities described DeepSeek as a catfish that stirred up the global model market, bringing a triple impact of performance, price, and open source. First, the performance is comparable to the world's top models: The performance of DeepSeekR1 in mathematics, code, natural language reasoning and other tasks is comparable to the official version of the OpenAI1 model. Second, Deepseek subverts the market pattern at low cost. DeepSeekV3 was trained with less than 2.8 million GPU hours, compared to 30.8 million GPU hours for the Llama3405B. The training cost of DeepSeek-V3 is only about $5.576 million, while the training cost of models such as GPT-4 can be in the hundreds of millions of dollars. Third, practice the concept of open source. Both DeepSeek-V3 and the inference model DeepSeek-R1 are open-sourced, and R1 simultaneously open-sources its model weights and allows users to use the model output to train other models through model distillation and other methods. Meta Chief Scientist Likun Yang commented on DeepSeek that "open-source models are surpassing proprietary models".

DeepSeek has promoted the overall prosperity of the AI ecosystem, and all links in the upstream, midstream and downstream industry chain may benefit from it. Specifically, Peng Xinchang analyzed, including upstream computing hardware suppliers, such as enterprises that provide hardware equipment such as chips and servers; A midstream public and private cloud service provider that provides cloud computing resources for model training and applications; Downstream software system suppliers, developing all kinds of applications based on large models, as well as enterprises or individuals from all walks of life as end users.

Wang Peng, an associate researcher at the Beijing Academy of Social Sciences, said that the emergence of Deepseek will benefit from the following industries: First, midstream software services, with the popularization of AI models such as Deepseek, the demand for efficient and intelligent software services will increase significantly, promoting innovation and development in this field. Second, downstream applications, Deepseek's technology will be widely used in many fields such as healthcare, education, entertainment, and intelligent manufacturing, giving birth to new business models and application scenarios. Third, the small model with high inference performance distilled by Deepseek R1 can better adapt to the device, provide technical support for the implementation of AI applications on the device, and promote the R&D and sales of related hardware devices.

AI + Wave of Change

There is no doubt that the emergence of Deepseek has once again ignited the investment enthusiasm of A-share technology stocks. Wind data shows that as of February 12, 23 artificial intelligence-themed ETFs have made amazing returns during the year. The highest return is the Bosera Science and Technology Innovation Board Artificial Intelligence ETF, with a return of 517.49% during the year, the second place Yinhua SSE Science and Technology Innovation Board Artificial Intelligence ETF with a return of 430.46% during the year, and the third place is the GF SSE Science and Technology Innovation Board Artificial Intelligence ETF, with a return of 281.06% during the year. All AI-themed ETFs returned more than 140% during the year, with an average return of 186.89%.

Since 2023, the enthusiasm of public funds to invest in artificial intelligence has never waned. In 2023, ChatGPT will quickly become popular, detonating the artificial intelligence market, and artificial intelligence, especially large language models, will have a huge impact on the labor market and social production and life. AI has set off the fourth industrial revolution, triggering a boom in A-share investment from models to computing power to applications. 2024 is regarded as the first year of AI applications. Overseas OpenAI has been live streaming for 12 consecutive days and throwing heavy "bombs", and domestic ByteDance's bean bags have set off an unprecedented boom, and the attention of the AIAPP market has increased greatly.

According to the data of fund heavy stocks provided by Tianxiang Investment Advisory Fund Evaluation Center, the reporter found that in the fourth quarter of 2024, among the top 50 stocks in the market value of public funds, there are 22 information technology stocks, of which the top five are information technology stocks, which are Cambrian (688256. SH), SMIC (688981. SH), ZTE (000063. SZ), Haiguang Information (688041. SH) and Hengxuan Technology (688608.SH).

Compared with the past two years, what are the different features of Deepseek this time?