Ma Yun's appearance once again sparks discussion! DeepSeek becomes popular, can Alibaba (09988) leverage AI?
12/02/2025
GMT Eight
Yesterday, Alibaba founder Jack Ma appeared again in the Alibaba campus. According to the Sci-tech Innovation Board Daily, Ma appeared in the Alibaba Intelligent Information Business Group where Quark is located, and interacted with employees.
After the opening of the Hong Kong stock market yesterday, Alibaba's stock price rose to 108.6 Hong Kong dollars, then fell and closed down by 0.57%. From the beginning of the year to now, Alibaba's stock price has risen by about 27% overall. In 2024, after Alibaba completed a leadership change and invested a total of 16 billion US dollars (about 116.8 billion RMB) in share buybacks within the year, Alibaba's stock price increased by 10%, while in 2021, 2022, and 2023, the stock price fell by about 48%, 19%, and 11% respectively.
Some industry analysts believe that DeepSeek has significantly ignited the interest of international investors in Chinese technology again. The new AI models released by Alibaba recently and the integration of DeepSeek models into Alibaba Cloud have all contributed to the previous rise in Alibaba's Hong Kong stock.
American hedge fund giant continues to increase holdings
Alex Au, CEO of Alphalex Capital Management in Hong Kong, stated that DeepSeek has once again significantly ignited the interest of international investors in Chinese technology and is expected to help narrow the valuation gap. Alibaba announced its latest AI model shortly after DeepSeek was released, claiming better results. Alibaba Cloud integrating DeepSeek models also contributed to the rise of Alibaba's Hong Kong stocks. Au emphasized that cloud service providers like Alibaba will be one of the major beneficiaries of this DeepSeek trend.
On Monday this week, American hedge fund giant and billionaire investor David Tepper's Appaloosa LP submitted a 13-F filing to the US SEC, showing that in the fourth quarter of last year, the hedge fund significantly increased its holdings of Chinese stocks and Chinese stock ETFs.
Among them, Appaloosa's holdings of Alibaba increased by 18% to 11.8 million shares, currently valued at over 1.2 billion US dollars. As of the end of last year, Alibaba remained its largest holding, accounting for 15.4% of its 6.7 billion US dollar investment portfolio.
In addition, the holdings of JD.com increased from approximately 7.3 million shares to approximately 10.5 million shares, a 43% increase. Currently valued at over 400 million US dollars. The holdings of Pinduoduo increased by approximately 55,000 shares to 5.4 million shares, an increase of about 1%. As of the end of last year, Pinduoduo was the third largest holding in Appaloosa's portfolio, accounting for about 8.04%. The holdings of Beike Holdings increased from approximately 2.18 million shares to approximately 2.57 million shares.
In September last year, Tepper called for "buying everything" related to China, causing a sensation. The latest report shows that he has followed through on his commitment.
Jack Ma: AI will change everything
On December 8, 2024, Jack Ma appeared at the 20th anniversary event of Ant Group and emphasized AI development. He stated that from today's perspective, the changes that the AI era of the next 20 years can bring will exceed everyone's imagination, because AI will be a greater era. "AI will change everything, but that does not mean AI can decide everything. Technology is important, but what will truly determine victory or defeat in the future is what valuable and unique things we do for this upcoming era today," Ma added.
Today, Jack Ma appeared in the Alibaba campus, and according to the Sci-tech Innovation Board Daily, he visited the Alibaba Intelligent Information Business Group where Quark is located and interacted with employees.
Recently, Alibaba's AI to C business has been constantly expanding, and the Alibaba Intelligent Information Business Group has taken the responsibility of Alibaba's AI applications on the consumer end. Previously, Alibaba's AI application "Tongyi" became part of the Alibaba Intelligent Information Business Group, becoming one of the two consumer brands developing alongside Quark. This month, a top global AI scientist and former Vice President of Salesforce, Xu Zhuhong, was appointed as the Vice President of Alibaba Group, responsible for the basic research and application solutions related to AI to C business.
In terms of large-scale models, the technical progress of Alibaba's Tongyi large model has attracted attention in the market. On the first day of the Lunar New Year, Alibaba Cloud released the flagship version of the Thousand Questions Tongyi model Qwen2.5-Max. In the Chatbot Arena LLM Leaderboard, Qwen2.5-Max ranked seventh globally with 1332 points, surpassing models like DeepSeek V3, o1-mini, and Claude-3.5-Sonnet. At the same time, Qwen2.5-Max ranked first in mathematics and programming, and second in Hard prompts.
Recently, research teams from Stanford and UC Berkeley have used Alibaba's Qwen2.5 large model for fine-tuning to reproduce the DeepSeek-R1 model at a low cost. For example, Stanford University professor Fei-Fei Li's team used Alibaba's open-source Qwen2.5 model to train a new reasoning model for less than $50 in cloud computing costs. The team led by Pan Jiayi at UC Berkeley also replicated the core functions of DeepSeek R1-Zero at a cost of $30.
It is worth noting that DeepSeek also selected the Thousand Questions Tongyi as the basic model. DeepSeek distilled DeepSeek-R1 into six open-source models, four of which are based on the Qwen series. In particular, the model distilled based on Qwen-32B has achieved performance comparable to OpenAI o1-mini.
DeepSeek has also brought new opportunities for cloud infrastructure to Alibaba Cloud. Alibaba Cloud has fully launched six full-size DeepSeek models on its Hundred Refinements platform.
Large model development may see a variety of breakthroughs
According to Shi Mao, partner at Changlei Capital, the breakthrough of DeepSeek has opened a window for the world. The eruption of DeepSeek has confirmed the arrival of the era of AI-native applications. Previously, when mentioning model-as-application, many people thought it was at the basic.Based on the model, some applications can be made. And DeepSeek has proven one thing: there may be various model fine-tuning breakthroughs coming up next.In other words, future entrepreneurs, even application entrepreneurs, must have the ability to model. "By solving a certain bottleneck in model technology, a model with its own unique effectiveness in different scenarios is created. Just like DeepSeek, although its inference ability is good and can even rival with OpenAI O1 model, it is still OpenAI that leads in generalization. Therefore, in the future, there may be a situation where various models will flourish on the model side, rather than the oligopoly era of only a few large model bases as previously speculated."
This article is reprinted from "Cailian Press", edited by GMTEight: Liu Jiayin.