The Era of Hardcore Innovation in China has Arrived

Amidst the festive reunion during the 2025 Spring Festival, Deepseek, an "underdog" in the AI sector, has become an overnight sensation across China. This Chinese company, with its cutting-edge products, has made a striking debut and garnered global attention. Its influence has spread from the technological frontier to the capital market, causing significant fluctuations in related stocks in Hong Kong and US markets and prompting international tech giants to urgently adjust their strategies.

Shengma Finance notes that multiple A-share listed companies have successively announced the integration of their products with the DeepSeek large model. Deepseek's popularity underscores the breakthroughs in AI technology and the profound impact on the capital market under the wave of new technologies, heralding a new era in the development of the AI industry.

Deepseek's Global Rise to Fame

On January 27, 2025, DeepSeek, an AI large model company based in Hangzhou, topped the free charts on both the Apple AppStore in China and the United States with its open-source inference large model DeepSeek-R1, released just a week prior. Products like OpenAI's ChatGPT and Google's Gemini followed closely behind. This marked another remarkable achievement for DeepSeek after the release of DeepSeek V3 on December 26, 2024.

One of the key reasons for Deepseek's extensive discussion in overseas markets is the attention from industry leaders. Microsoft CEO Satya Nadella praised DeepSeek's innovations on January 29, and on January 30, NVIDIA announced that its developer website had included the DeepSeek-R1 model in the "Most Popular Models" section.

Subsequently, major domestic internet cloud service providers began competing to integrate the DeepSeek series of models, investing substantial resources to seize opportunities.

Shengma Finance notes that in the capital market, Hong Kong stocks were the first to respond positively to Deepseek's popularity. As of the close on February 4, the Hang Seng Tech Index surged by 5.06%, presenting a booming scene. Hua Hong Semiconductor led the gains with a 12.66% increase, and its technological strength in the semiconductor field synergized positively with the rise of Deepseek. SMIC closed up 8.47%, with its share price hitting a historical high during the session. As a leading enterprise in the semiconductor industry, SMIC's share price performance reflects the market's optimism about the prospects of AI-related industries. Kingsoft Cloud surged 9.72%. In the field of cloud computing, Kingsoft Cloud provides robust technical support for the operation of Deepseek products, and the collaboration between the two has been highly recognized by the market.

On February 5, the first trading day of the Year of the Snake, the DeepSeek concept flourished, leading the gains in both markets. Sub-sectors such as computing power, AI chips, storage chips, and intelligent agents collectively rose, with DeepSeek-related stocks almost all hitting their daily limit during the auction period.

Zhejiang Dongfang had over 12 million sealed orders, and Daily Interactive announced on February 4 that it had not yet provided corpus data to DeepSeek but was still warmly received by funds, achieving a "20CM" daily limit for two consecutive days. Stocks such as Meige Intelligence, Hangjin Technology, and Huajin Capital all recorded straight-line gains.

Numerous users have shared their experiences using Deepseek products on social media, expressing satisfaction with their powerful functions and ease of use, which significantly enhance life and work efficiency. This positive word-of-mouth has further promoted the product's spread and expanded its market influence.

Technological Advantages and Groundbreaking Innovations

In the strategic competition field of AI, highly valued by American capital, China is carving its own path. While the United States has its "ChatGPT moment," China now boasts its "DeepSeek moment."

Compared to ChatGPT's "three high" characteristics—high-end chip clusters, high training investments, and high computing power supply—DeepSeek's most notable feature is its significantly reduced training costs while maintaining excellent performance. The first version of DeepSeek-V3, launched and open-sourced on December 26, 2024, and products like DeepSeek-R1 released this January, have achieved the performance of advanced large models in the United States at astonishingly low costs.

It is particularly noteworthy that the development of DeepSeek did not involve a quick replication of Meta's open-source model Llama but rather numerous innovations in architectural design and algorithm optimization, significantly enhancing inference efficiency and reducing resource consumption.

For example, it introduces a novel MLA mechanism (Multi-head Latent Attention), which, compared to the traditional MHA mechanism (Multi-head Attention), reduces memory usage to only 5%-13% of the original. Specifically, the MLA mechanism introduces the concept of latentization, performing a certain degree of preprocessing on input information to selectively compress and simplify it, thereby significantly reducing the number of parameters and lowering computational complexity when generating queries (Query), keys (Key), and values (Value).

Taking DeepSeek-V2 as an example, it adopts the Transformer architecture, leveraging the self-attention mechanism to accurately capture key information in sequence data, thereby better processing various types of data such as language and images. This greatly enhances the model's ability to understand contextual information and is one of the core technologies in the current AI field.

Shengma Finance understands that the MLA architecture introduced in DeepSeek-V2 is a highlight of its technological innovation. The MLA architecture further optimizes the attention mechanism and significantly reduces computational volume and inference memory by efficiently allocating computing resources. This makes the model more efficient in processing large-scale data and capable of completing complex tasks in a shorter time. At the same time, its performance reaches GPT-4 levels, yet it is open-source and can be used commercially for free. This advantage provides numerous enterprises and developers with low-cost, high-performance AI solutions.

In the field of image recognition, image analysis tools developed based on DeepSeek-V2 can accurately identify objects, scenes, and other information in various images, providing strong support for the development of related industries. This technological innovation not only improves the model's operational efficiency but also reduces hardware costs, laying a solid foundation for the popularization and application of AI technology. In practical applications, some small enterprises have quickly built their own intelligent customer service systems leveraging the open-source advantages of DeepSeek-V2, which not only reduces development costs but also improves customer service efficiency.

Redefining the AI Market Landscape

Shengma Finance learned from multiple industry analysts that the rise of Deepseek is akin to a bombshell that is highly likely to completely disrupt the existing AI market landscape.

For a long time, international technology giants such as NVIDIA have dominated the AI field, leveraging their strong technological strength and market share to build formidable competitive barriers. However, Deepseek's low-cost advantage poses a serious threat to their market share and profitability.

From January 24 to January 31, NVIDIA's share price fell by over 18%, with a single-day drop of over 17%. This significant share price fluctuation fully reflects the market's concerns about its future development.

In response to the impact of Deepseek, international technology giants have adopted countermeasures. On January 29, Microsoft, Amazon Web Services, NVIDIA, and other international technology giants announced their integration with Deepseek, attempting to leverage Deepseek's technological advantages while mitigating its impact on their market share through collaboration.

Huawei Cloud also announced on February 1 that it has partnered with Silicon-based Flow to launch and go live with the Deepseekr1/v3 inference service based on Huawei Cloud Shengteng services, further promoting the application and promotion of Deepseek technology.

Meanwhile, domestic AI-related enterprises have ushered in unprecedented development opportunities. Securities traders were actively engaged during the Spring Festival holidays, conducting roadshows and issuing research reports even on New Year's Eve and Spring Festival, with over 60 events in just 8 days. Multiple securities traders such as Huatai Securities, Guotai Junan, and China Galaxy issued research reports on Deepseek, focusing on topics such as its triggering of model cost reduction and opening up the upward channel for the AI application industry.

From the perspective of industry development trends, Deepseek's popularity heralds a new phase for the AI industry. Its "low-cost + high-performance + open-source" model provides a novel approach for the popularization and application of AI technology, breaking the constraints of traditional AI development models.

Shengma Finance believes that for investors, while the AI field will be a popular investment direction, it is necessary to rationally view market fluctuations and pay attention to enterprises' technological strength and innovation capabilities. On the global AI competition stage, Deepseek has taken a solid step forward. It remains to be seen whether China's AI industry can seize this opportunity to occupy a larger share in the international market.

In the future, with the advancement of AI technology and the expansion of application scenarios, it is expected that more enterprises like Deepseek will emerge, pushing the domestic AI industry to new heights and injecting new vitality into global technological development and economic growth.

END

Follow us for more exciting content

Shengma Finance's sincere original creation, reproduction is prohibited without authorization