As of February 5, 2025, ahead of the Hong Kong stock market's opening, the Hong Kong Internet ETF (159568) surged 5.24%, closing at 1.47 yuan with a turnover rate of 18.88%, indicating robust market activity.

DeepSeek-R1, which burst onto the scene on the eve of the Spring Festival, remains a key driver of this upward trend. During the holiday, DeepSeek-R1 topped global search trends, continuously capturing market attention and igniting the AI concept in both A-shares and Hong Kong stocks post-festivities. According to Tonghuashun iFinD data, the Hong Kong stock AIGC concept index soared 23.8% in just 9 trading days since January 21. Notably, Kingsoft Cloud (03896.HK) accumulated a 5-day gain of over 70% as of February 4.

Other constituent stocks, including Meitu (01357.HK), Tencent Holdings (00700.HK), Alibaba (09988.HK), and Baidu Group (09888.HK), also witnessed gains, presenting a thriving landscape for the Hong Kong stock AI concept sector.

In recent news, prominent platforms in China such as Baidu Intelligent Cloud, Huawei Cloud, Alibaba Cloud, Tencent Cloud, 360 Digital Security, and ZStack Technology have successively announced the launch of the DeepSeek large model.

A research report by CITIC Securities stated that the release of DeepSeek's new generation of models signifies the gradual popularization of large AI models, accelerating the comprehensive implementation of AI applications.

DeepSeekR1: Domestic AI Makes a Comeback, Igniting the "Hong Kong Stock Market Frenzy"

The DeepSeek concept led the charge into the Spring Festival's stock market frenzy.

On January 20, 2025, QuantInsights' subsidiary DeepSeek released its latest open-source model, DeepSeekR1, which immediately sparked a global craze upon its debut.

Since ChatGPT's emergence, the Scaling Law has become the dominant trend in AI model development. This law underscores the power relationship between model performance and computational power, model parameters, and data size, making model performance heavily reliant on computational power. NVIDIA, a leading computing power chip company, has experienced explosive growth under the Scaling Law's dominance. From 2023 to 2024, NVIDIA's performance consistently exceeded expectations, with its share price soaring nearly 10 times year-on-year.

However, DeepSeekR1 disrupts this inherent pattern of relying on computational power and chips for "miracles through brute force." Instead, it achieves AI equality through algorithm optimization, marking a significant victory for domestic AI and presenting the possibility of reassessing the value of global AI assets.

After analysis, multiple investment institutions believe that DeepSeekR1, a recent global sensation, offers crucial insights for the global AI industry chain with its capabilities and low costs. A research report by Guotai Junan Securities pointed out that the DeepSeekR1 API pricing is as low as 1 yuan per million tokens, with performance comparable to the O1 model. It excels in logic, mathematics, and Chinese tasks, outperforming models like GPT-4o and O1-mini. The R1 model demonstrates the potential to run efficient AI models on low-cost hardware, achieving computational power comparable to that achieved by American giants spending billions of dollars with just a few million dollars.

CITIC Securities stated that DeepSeek-R1, released on January 20, 2025, offers performance comparable to the official version of OpenAI O1, emphasizing a reinforcement learning technology route with strong reasoning capabilities and outstanding cost-effectiveness. The institution believes that the release of DeepSeek's new generation of models is expected to introduce a new Scaling Law, with the model's focus gradually shifting from pre-training to reinforcement learning and reasoning stages, thereby driving continuous growth in computational power demand. It recommends focusing on three major investment themes: tool software and other key industry software, general management software, and AI computational power.

The impact of the DeepSeek model on the past rule of "computational power is king" in the AI field is evident in the performance of US technology stocks. On the eve of the Chinese Spring Festival, NVIDIA's share price plummeted by nearly 17% in a single day, marking its largest single-day market value loss since listing on US stocks. During the Spring Festival, the share prices of the "Tech Seven Sisters" in US stocks also fell sharply, with even AI infrastructure-related hardware suppliers like Siemens and Schneider Electric witnessing declines.

The DeepSeek model's outstanding performance in the AI field has garnered attention and regulation from many Western countries. In the over half a month since DeepSeek's launch, multiple countries, led by the United States, have issued bans, and some technology companies have attempted to prohibit its use. However, this has not dampened investors' optimism about its prospects, and the market still grants it trust.

Post-Spring Festival, the Hong Kong stock market frenzy first erupted in the DeepSeek concept, with related concept stocks continuing to surge significantly. Notably, stocks like Kingsoft Cloud (03896.HK), Meitu (01357.HK), SenseTime-W (00020.HK), and Kingsoft Software (03888.HK) averaged gains of over 20% in four trading days.

When DeepSeek Meets Tech Giants: A Subversion is Underway

In the early stages before the AI application explosion, while the DeepSeek model has temporarily negatively impacted the demand for high-computational power hardware suppliers like NVIDIA, its benefits to the entire industry chain far outweigh its drawbacks.

China Aviation Securities noted that global technology companies are actively exploring the business opportunities that DeepSeek may bring.

During the Spring Festival, many renowned domestic internet platforms announced the launch of the DeepSeek large model, allowing users to access models such as DeepSeek-R1 and DeepSeek-V3 on these platforms.

On February 1, Huawei Cloud announced the joint launch and go-live of the DeepSeekR1/V3 inference service based on Huawei Cloud's Shengteng Cloud Service after days of collaboration between the silicon-based flow and Huawei Cloud teams. On the same day, Tencent Cloud announced the one-click deployment of the DeepSeek-R1 large model on Tencent Cloud's "HAI," enabling developers to access and call it within just 3 minutes. 360 Digital Security also stated that its security large model had recently been officially connected to DeepSeek. ZStack Technology announced that its AI Infra platform, ZStack Zhita, fully supports enterprises in privately deploying the three models of DeepSeek V3/R1/JanusPro.

On the evening of February 3, Baidu Intelligent Cloud announced that its Qianfan platform had officially launched the DeepSeek-R1 and DeepSeek-V3 models, introducing ultra-low-price plans and limited-time free services. Users can experience them by logging into Baidu Intelligent Cloud Qianfan ModelBuilder. On the same day, Alibaba Cloud announced that its PAI ModelGallery supports one-click deployment of DeepSeek-V3 and DeepSeek-R1 on the cloud.

Not only in the domestic market but also overseas technology vendors such as Amazon AWS, Microsoft Azure, and NVIDIA have recently announced the integration of the DeepSeek model. Amazon stated that the DeepSeek-R1 model is now available on Amazon Web Services, with CEO Andy Jassy encouraging users to utilize it boldly.

Notably, Microsoft, as the largest shareholder of OpenAI, which once aimed to create non-profit AI for humanity's benefit, has long funded OpenAI and deeply integrated its AI technology. However, Microsoft recently announced that it would officially incorporate DeepSeek-R1 into Azure AI Foundry, making it part of the enterprise-level AI service platform.

NVIDIA's website shows that the DeepSeek-R1 model has been released as an NVIDIA NIM microservice preview on its developer-oriented website and included in the "Most Popular Models" section. The website states that the DeepSeek-R1 model is an advanced and efficient large language model that excels in reasoning, mathematics, and coding.

Clearly, DeepSeek has attracted global technology giants with its innovative technology route, indicating that domestic technology has strong competitiveness in an open and innovative open-source environment and is poised to emerge with more universal technologies. The value of domestic local technology enterprises should be reassessed.

Consequently, technology enterprises in both A-shares and Hong Kong stocks are becoming highlands for investors to deeply explore new opportunities. AI applications are likely to be the core theme throughout the 2025 investment market.

DeepSeek Breaks the Deadlock: AI Industry Enters an "Acceleration Phase"

CITIC Securities' analysis points out that DeepSeek has achieved a significant breakthrough at the ecological level and is expected to become a powerful engine driving the rapid development of the domestic AI industry. Its open-source characteristics and low-cost advantages can provide robust support for AI application vendors and accelerate the implementation of edge-side AI.

As early as last year's end, rumors circulated that Lei Jun had poached Luo Fuli, a key DeepSeek developer, with an annual salary in the tens of millions. Luo Fuli will directly lead Xiaomi's large model team. Xiaomi expects her to optimize and enhance Xiaomi's AI large model's efficiency and user experience based on her valuable experience in developing DeepSeek-V2, injecting new vitality into Xiaomi's AI field development.

Everbright Securities believes that as the only strategic cloud platform within Xiaomi and Kingsoft's ecosystem, Kingsoft Cloud will continue to provide cloud services to Kingsoft Software and Xiaomi Group over the next three years. Kingsoft Software's office business, WPS AI, continues to advance its commercialization, while cloud game distribution and operations gradually expand. Xiaomi Group's research and development in intelligent driving training for Xiaomi cars, data storage for autonomous driving, and the continuous development of the mobile phone × AIoT segment will all drive steady growth in Kingsoft Cloud's revenue from the ecosystem, helping it secure a more favorable position in market competition.

For AI vertical application enterprises, DeepSeek, with small parameters, low inference costs, and high performance, brings significant benefits. Taking Meitu as an example, its cumulative gain has exceeded 46% in the past five days. Despite slight fluctuations in its share price during this period, the market generally remains optimistic about its development prospects with AI technology's support, believing that Meitu is poised to achieve greater breakthroughs in the image application field with this technological advantage.

DeepSeek's influence also extends to Hong Kong stock giants such as SMIC, Tencent Holdings, Alibaba, and Baidu Group. These companies' share prices have shown a fluctuating upward trend amid the recent AI concept boom. After Baidu Intelligent Cloud announced the launch of the DeepSeek large model, the market has reconsidered Baidu's competitive landscape and future development direction in the AI field. Although Baidu itself has a robust AI technology system, the industry changes triggered by DeepSeek have prompted Baidu to reconsider its technology path and market strategy to adapt to the evolving market environment.

Although DeepSeek primarily focuses on the level of AI software models, the AI industry's development cannot be separated from the support of chip computational power. As a leading enterprise in the field of integrated circuit manufacturing, SMIC provides manufacturing services to numerous global chip design companies and serves as a crucial link in the semiconductor industry chain. With the continuous growth in demand for computational power from AI applications, SMIC will undoubtedly continue to benefit, further consolidating its industry position.

In the Chinese market, companies akin to "DeepSeek" that seek development through innovation and breakthroughs are still continuously emerging. According to reports, relying on its independently developed AI large model and Agent technology, Salesforce has created the AI-Agentforce intelligent agent platform. This platform covers various products and services such as marketing intelligent agents, sales intelligent agents, business decision-making intelligent agents, and search intelligent agents, providing strong support for the digital and intelligent transformation of multiple industries. So far, Salesforce has successfully tailored digital and intelligent transformation solutions for over 200,000 customers, demonstrating robust technical strength and market competitiveness.