CITIC Securities: How far can the Chinese asset revaluation brought by Deepseek go?

Wallstreetcn
2025.02.18 01:01
portai
I'm PortAI, I can summarize articles.

CITIC Securities analyzes the impact of DeepSeek on the revaluation of Chinese technology assets, believing that although Chinese technology companies have performed outstandingly recently, they are still in the early stages of the AI era. The success of DeepSeek marks a breakthrough in independent innovation, and in the future, Chinese technology enterprises need to continue to work hard in technological innovation and application implementation. Historical experience shows that there will be fluctuations within the technology cycle, and valuation increases need to rely on performance verification. DeepSeek's rapid user growth demonstrates its potential in the global AI market, which may change investors' perceptions of Chinese technology assets

Despite the significant recent gains of Chinese tech companies, looking at the longer historical cycle, we are still in the early stages of the AI era and the beginning of the revaluation of Chinese tech assets. As the allure of AI in the U.S. diminishes, Chinese AI breaks through in the cracks. DeepSeek can be seen as the "Shangganling" of China's tech sector, achieving independent innovation and breakthroughs. However, this "battle" is far from over. From breakthroughs to leadership, from technological innovation to application landing, from changing expectations to creating value, and from local low-quality competition to global high-quality development, Chinese tech companies still have a long way to go. Therefore, we are optimistic about the long-term revaluation and growth potential of Chinese tech and even broader assets, especially after the significant valuation pressure caused by the China-U.S. competition in 2018. In the future, every major technological breakthrough and application landing has the potential to bring about systemic valuation increases. However, at the same time, the experiences of the PC and mobile internet waves tell us that even within a major tech cycle, there will be twists and turns; the experience of AI asset revaluation in the U.S. also indicates that sustained valuation increases require performance to be validated quarter by quarter.

DeepSeek triggers a "butterfly effect" with excellent performance, significantly reduced costs, and an open-source ecosystem.

DeepSeek registered over 100 million users in just 7 days, breaking the record of 2 months set by ChatGPT. Currently, DeepSeek's daily active users have surpassed 30 million, making it the most popular AI large model globally, triggering the revaluation of Chinese AI assets.

Reviewing the revaluation of U.S. tech companies brought about by the launch of ChatGPT: From the beginning of 2023 to now, the Nasdaq has risen by 100%, with leading company NVIDIA's stock price increasing by 8.5 times, entering the top three globally in market capitalization. Companies like Palantir and Applovin have also emerged, rising 17.6 times and 47.4 times, respectively, during the same period, widening the valuation gap between Chinese and U.S. tech stocks. Since January, leading companies such as Alibaba, SMIC, BYD, and Xiaomi Group have seen their stock prices rise by 51%, 43%, 37%, and 30%, respectively, while IDC and cloud infrastructure companies have also experienced significant gains, leading to a narrowing of the valuation gap between Chinese and U.S. tech companiesDeepSeek single-handedly changes the expectation of "America's cliff-like lead in AI," thereby altering global investors' judgment on the value of Chinese tech assets. Founded in July 2023, with just over 100 employees, this startup's breakthrough is a reflection of China's technological innovation power against the backdrop of the prolonged Sino-U.S. competition since 2018. We believe we are currently in the early stages of the AI era and at the beginning of the re-evaluation of Chinese tech assets.

Industry Impact: DeepSeek brings AI equality, changes the global tech landscape, enhances demand for computing power, and ignites expectations for application explosions.

From V3 to R1, DeepSeek boldly innovates algorithm design and optimizes hardware performance to the extreme.

Under limited computing power conditions, DeepSeek first replicated GPT-o1, achieving performance in mathematics, coding, and natural language comparable to o1, and independently open-sourced RL-Scaling technology. The V3 version realizes low-cost and efficient training through innovative architectures such as MoE and MLA, as well as changes in training strategies. According to the DeepSeek-R1 documentation, its pre-training cost is nearly less than $6 million. The R1 version explores the transition from supervised fine-tuning (SFT) to zero-supervised fine-tuning, combining GRPO and an efficient reward system to significantly reduce costs while improving performance. DeepSeek breaks the monopoly advantage of the U.S. in AI competition, and AI equality will change the global tech industry landscape, igniting expectations for application explosions.

Currently, we are on the eve of an AI application explosion, and the demand for computing power will be the first to kick off. Companies like ByteDance's Doubao and Alibaba's Tongyi Qianwen are expected to quickly follow up with innovations, leading to a flourishing of vertical models.

The race among model companies brings a "pulse-like" growth in training computing power demand; the demand for inference computing power and local deployment of large models by big enterprises is also growing rapidly. **We are optimistic about investment opportunities in AI cloud computing, AIDC data centers, AI chips, and high-end manufacturing companies. On the application side, scenarios such as intelligent driving, AI advertising, AI agents, AI smartphones, and embodied intelligence are expected to gradually materialize**

Application Outlook: The future of AI lies not only in pursuing more powerful models but also in solving real-world problems and creating economic value through AI applications.

From technological innovation to application landing, from changing expectations to creating value, from local low-quality competition to global high-quality development, Chinese technology companies still have a long way to go. Currently, American large model companies remain significantly ahead, with strong advantages in cutting-edge fields such as multimodal, world models, and reinforcement learning. Looking ahead, the United States will continue to raise the technical ceiling of AI, while China's greater opportunity lies in integrating AI technology with traditional industrial foundations to solve practical problems and create economic value.

We judge that applications in areas such as intelligent driving, AI advertising, and AI agents will land first, contributing to revenue growth and changing market dynamics starting in 2025. The innovations of Apple's iPhone 17 in the third quarter are noteworthy, as AI phones may bring a new wave of edge-side intelligent opportunities, with local large models in China, such as Alibaba's Tongyi Qianwen, expected to participate. Embodied intelligence benefits from China's solid manufacturing foundation and is expected to become another advantageous industry for China. In terms of industry applications, AI + Internet, AI + Finance, AI + Healthcare, AI + Industry, and AI + Transportation are expected to land first. From the business chain perspective, AI + Auditing, AI + Regulation, and AI + Research are expected to significantly enhance operational management efficiency. In the more distant future, the deep connection and mutual influence between the physical and digital worlds may accelerate the arrival of the metaverse.

Revaluation Logic: R&D Spending in the Technology Sector from "Negative Valuation" to "Positive Valuation."

In the post-pandemic era, Chinese technology assets remain undervalued, mainly due to: 1) Slowing profit growth, leading to insufficient investor confidence in core profit growth; 2) Chinese AI being perceived as "generationally behind," with new business "R&D investment" understood as "losses" that erode valuations. In recent years, Chinese tech giants have often been valued using a PE ratio based on the overall method, and the losses from new business R&D have thus been reflected in the total market value as negative valuation. The pressure on core profit growth has exacerbated the valuation divergence between Chinese and American tech giants. The revaluation brought by DeepSeek, on the surface, appears to have increased the PE valuation multiples of leading companies, but at a deeper level, it signifies the shift of Chinese AI assets from negative to positive valuation. In the future, if there is an improvement in economic expectations, there is room for an increase in core valuation multiples; once AI businesses achieve significant revenue contributions, a shift in PS segment valuation is also expected. We judge that the revaluation of Chinese tech assets is still in its early stages. In the future, every major technological breakthrough and application landing has the potential to bring about systemic valuation increases. However, the experiences from the PC and mobile internet waves remind us that even within a major tech cycle, there will be twists and turns; the revaluation experience of AI assets in the US stock market also indicates that continuous valuation increases require performance verification quarter by quarter.

Investment Opportunities: Chinese AI Assets Primarily in the Internet Sector, Focusing on Computing Power, Domestic Chip Manufacturing Processes, Intelligent Driving, and AI Agents.

Chinese AI is accelerating its iteration, and domestic large models and intelligent cloud companies are expected to benefit significantly. The internet is not only an important landing scenario for AI, such as WeChat integrating DeepSeek, the acceleration of AI advertising landing, and the increasing demand for AI content generation; it is also a provider of intelligent cloud and domestic large models. Under the grand narrative of Chinese AI, the revaluation of internet companies will continue, but the sustainability of this revaluation depends on the quarterly verification of leading companies' performance. In terms of computing power, similar to the position of NVIDIA in the revaluation of US assets over the past two years, SMIC will play a crucial role in the revaluation of Chinese tech assets in the next 1-2 years. SMIC's pricing power, revenue growth rate, and profitability in advanced processes are expected to become important directional guidance for the market. On the application side, intelligent driving is expected to achieve a doubling of penetration rates by 2025, as the Chinese automotive industry begins to explore globalization. Traditional enterprises are starting to attempt local deployment of DeepSeek, accelerating the application landing of AI large models, with areas such as AI + finance, AI + healthcare, AI + industry, and AI + transportation worth paying attention to; AI agents are expected to become an important form of application landing. Additionally, AI is expected to feed back into the digital upgrade of industries, bringing performance opportunities for sensor, PC, server, and IT service companies.

Investment Strategy:

We maintain the judgment in the report "Technology Industry January 2025 Investment Strategy - Focus on Domestic AI Models and Computing Power Processes, Intelligent Driving Entering an Acceleration Development Period" (2025-1-14), the attractiveness of Chinese AI assets is gradually becoming apparent, and leading AI companies are expected to gain performance-driven alpha investment opportunitiesLeading AI companies have a significant comparative advantage in valuation relative to comparable companies in the United States. Once the performance and valuation pressures due to slowing CAPEX growth emerge in U.S. tech stocks, the attractiveness of Chinese assets is expected to further manifest. Key investment opportunities to focus on include domestic large models and intelligent cloud, China's AIDC, domestic computing power processes, intelligent driving, AI agents, and embodied intelligence.

Authors of this article: Xu Yingbo S1010510120041, Xu Tao S1010517080003, etc., Source: CITIC Securities Research, Original title: "CITIC Securities | How far can the revaluation of Chinese assets brought by Deepseek go?"

Risk Warning and Disclaimer

The market has risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at one's own risk