After fund managers from the A-share market went to the Hong Kong stock market, $KINGSOFT CLOUD (03896.HK)$ it achieved a sevenfold increase in stock price within five months.

With the continuous amplification of the AI industry model and the technological breakthroughs brought by DeepSeek, the undervaluation and value gap of heavyweight stocks in the Hong Kong market, represented by Kingsoft Cloud, are constantly being filled, allowing Kingsoft Cloud to take this opportunity to achieve over a sevenfold increase in stock price within five months. It is worth mentioning that all Hong Kong-themed fund products under public offering are absent from holding Kingsoft Cloud, and the fund products that currently include Kingsoft Cloud in their top ten holdings list are all from A-share fund products, reflecting the characteristic of being few but elite in the stock selection strategy of relevant A-share fund products.

Public offering accelerates capital allocation, with heavyweight stocks rising sevenfold in five months.

On February 10, Kingsoft Cloud, the heavyweight stock, closed up nearly 10%, with a market cap exceeding 35 billion Hong Kong dollars. Five months ago, Kingsoft Cloud was still a micro-cap company on the Hong Kong stock market with a market cap of 4 billion Hong Kong dollars, which means that the stock achieved a price increase of up to sevenfold in just five months. The latest holdings information disclosed by public funds shows that as of the end of last December, there were 30 public fund products, including Ping An Fund, Penghua Fund, and Golden Eagle Fund, heavily investing in Kingsoft Cloud.

Kingsoft Cloud's accelerated embrace of DeepSeek has obviously brought price benefits to fund managers holding the stock. On February 8, Kingsoft Cloud announced full support for DeepSeek-R1/V3 in public cloud and state-owned cloud/government cloud scenarios, further expanding its service capabilities in high-performance large language model inference and fine-tuning tasks. Kingsoft Cloud has released various mirror services based on the DeepSeek-R1 distilled model, and users can experience it for free on the Kingsoft Cloud official website. Users can set up inference services and conduct parameter tuning verification on public cloud GPU servers and GPU bare metal servers, supporting the operation of DeepSeek series models.

Based on the fact that public fund companies are also embracing and deploying DeepSeek, fund managers have a more direct and convenient way to research and understand Kingsoft Cloud's embrace of DeepSeek. Recently, many leading fund companies in the public offering industry, including China Europe Fund, Huitianfu Fund, Bosera Fund, and Fuguo Fund, have announced the completion of multiple DeepSeek models deployment. At the same time, large models will comprehensively empower the investment research, compliance, operation, and market businesses of fund companies, helping them improve work efficiency and support business innovation. Specifically, in pulling public fund investment stock selection, Nuoan Fund announced the completion of the localized deployment of the DeepSeek financial large model and launched the self-developed "Nuoan AI Assistant" based on mainstream AI open-source frameworks, initiating pilot applications in key business scenarios such as investment research analysis, customer service, and risk control.

Clearly, the fund managers' recognition of DeepSeek as an advanced productivity tool for empowerment, performance enhancement, and efficiency will lead public funds to more profoundly determine the potential of relevant stocks in the Technology sector through their stock selection and heavyweight strategy, especially as heavyweight stocks start to embrace DeepSeek at the business and efficiency level. This also gives the fund managers a more realistic logical basis for their overweight strategy on Kingsoft Cloud.

With the absence of public funds in Hong Kong stocks, A-shares fund managers seize the opportunity.

It is worth noting that while A-share theme funds are profiting from Kingsoft Cloud, public funds' Hong Kong theme fund products are absent from the price feast of Kingsoft Cloud in the Hong Kong stock market.

Interestingly, all public products holding Kingsoft Cloud as a heavyweight stock in Hong Kong come from A-share theme funds, whereas the public funds' Hong Kong theme fund products are entirely absent from this feast. The southbound gold mining by A-share fund managers becomes another reasoning for Kingsoft Cloud's stock price surge. Taking the Ping An Innovation Future Fund as an example, out of the current top ten heavyweight stocks of this A-share fund, eight are allocated to A-share listed companies, with Kingsoft Cloud and Semiconductor Manufacturing International Corporation being the only two Hong Kong stock positions.

Reporters have noticed that many public Hong Kong theme fund products cleared their positions in Kingsoft Cloud as early as the beginning of last year. For quite a long time, Kingsoft Cloud was unable to enter the top ten heavyweight stocks of mainstream public Hong Kong theme products, but while Hong Kong fund products forgotten about Kingsoft Cloud, the public A-share fund products were seizing the opportunities in the Hong Kong market, identifying Kingsoft Cloud as a highly elastic core symbol, leading to the unusual phenomenon where Hong Kong fund managers ignored Kingsoft Cloud while many A-share fund managers regarded it as a core stock.

Taking the Ping An Advantage Return Fund under Ping An Fund as an example, although this A-share theme product allocated only about 14% to Hong Kong stocks after entering the Hong Kong market, this 14% stock position includes Kingsoft Cloud, a super bull stock, and this A-share fund product positions Kingsoft Cloud as its largest heavyweight stock, thereby achieving an expansion in fund yield elasticity. Additionally, in the Haifutong Technology Innovation Fund under Haifutong Fund, almost all of its top ten heavyweight stocks point towards A-share companies, with Kingsoft Cloud being the only Hong Kong company in the top ten positions, reflecting the selective but refined stock picking characteristics of some A-share fund managers in the Hong Kong market.

It is precisely because the stock picking ability of A-share theme products in the Hong Kong market continues to improve that, so far, the performance comparison between the two types of products featuring the Technology sector shows that the return rates of A-share fund products after the start of the year far exceed those of Hong Kong Technology theme fund products.

As the application end explodes, public funds shift their judgment of AI opportunities from Hardware to Software.

The crazy surge of Kingsoft Cloud, characterized by Software applications, is significantly different from the hardware-centric investment trend in the market in 2024, and highlights the enormous impact of the AI industry on the stock market. Multiple fund managers have determined that the biggest opportunities in the AI sector this year will shift from the hardware side to the Software application side.

Penghua Semiconductor Fund manager Luo Yingyu pointed out that the technological innovation and breakthroughs of DeepSeek have attracted widespread attention from the market and have had a structural impact on the A-share market trend. From the perspectives of Cloud Computing and Big Data, as well as Media, DeepSeek's AI model has significantly influenced market expectations for investments in AI infrastructure, prompting investors to reevaluate the investment logic in the AI hardware market. In the past, investments in the AI industry were mainly concentrated in high-end computing power and Datacenter areas, with a linear relationship between AI performance improvements and computing power investments. However, the emergence of DeepSeek indicates that it is also possible to achieve this through innovative technological paths and business models. DeepSeek may reduce reliance on high-end hardware, as the low-cost AI model it launched has achieved performance comparable to industry giants with only $5.6 million in training costs.

These fund managers have determined that this breakthrough has led the market to reevaluate the demand for expensive hardware, especially for companies like NVIDIA, whose stock price has adjusted due to DeepSeek's success, reflecting investor concerns about potentially weakening demand for high-end AI Chips. Correspondingly, in the domestic market, related stocks along the NVIDIA chain have also shown significant adjustments during the day, such as notable adjustments in optical communication modules.

DeepSeek will accelerate the commercialization of AI applications, especially in terms of cost-effectiveness. Luo Yingyu emphasized that DeepSeek's R1 model performs comparably to OpenAI's model, but its training and running costs are only 3%-5% of that. This innovation enables more companies to develop and deploy AI applications at a lower cost, fostering commercial opportunities in Education, Entertainment, and e-commerce sectors. Furthermore, DeepSeek's open-source nature also promotes the proliferation of AI technology, which is expected to accelerate the development and innovation of the entire industry.

Morgan Stanley's digital economy fund manager Lei Zhiyong believes that the domestic large model DeepSeek has recently become the global focus, officially launched in January 2025, and has since stood out among similar products globally due to its remarkable performance and low-cost advantages. After 20 days of launch, it surpassed 20 million daily active users globally, making it the fastest-growing AI application in the world. In the context of limited computing power, DeepSeek-R1 has significantly reduced AI computing cost through engineering optimization, while its open-source feature has leveled the playing field for models and narrowed the technological gap between open-source models and closed-source models, as well as between AI technologies in China and the USA, bringing competition in the AI field into a new phase.

We believe that the parameters and training tokens of the DeepSeek model itself are being magnified, and it is expected to develop towards inference models and multi-modal integration in the future, with accelerated iterations of the inference models. Lei Zhiyong emphasized that DeepSeek is likely to accelerate the realization of AI applications, driving an explosive growth in demand for computing power in subsequent training and inference. In the medium to long term, according to 'Jevons Paradox', technological advancements in AI will reduce the cost of computing resources, thus leading to a rapid increase in demand, ultimately resulting in sustained growth in computing power demand.

Looking ahead at the development of the entire AI field, Lei Zhiyong stated that with the emergence of DeepSeek as a representative, the future development of the global AI field may be more diverse, as reducing costs with large models will inevitably accelerate the formation of a commercial closed loop for AI applications, while open-source and innovation will also accelerate the explosion of diversity in AI applications.

These fund managers expect that in 2025, the AI industry chain may still maintain high prosperity, and companies involved in the inference side, endpoint computing power chips, and leading Software application companies in the A-share market, as well as leading enterprises in Hong Kong that actively layout AI, are likely to benefit from the domestic AI wave brought about by DeepSeek. Whether this round of market can continue will depend on the subsequent commercialization process of domestic AI applications, such as software applications for the ToB side or Saas companies, and whether endpoint AI + Hardware can gradually transition to a commercial closed loop.

Editor/new