With the upcoming release of the DeepSeek R2 model, its impact on the industry and the future layout of major technology companies are full of uncertainty, but one thing is certain: China’s AI field is ushering in a new era of competition and innovation.
It is difficult to say whether the upcoming DeepSeek R2 model will bring another impact to the large model industry, but what is certain is that in the 100 days since the launch of DeepSeek R1, the original normal rhythm of competition in China’s AI field has been completely reconstructed.
Since the release of DeepSeek R1 on January 20, it has only taken more than 3 months for DeepSeek to create an industry myth, and as of March, QuestMobile data shows that its APP of the same name has become the AI native application with the largest number of users in China with more than 190 million monthly active users.
The instantaneous explosion of DeepSeek’s user scale has made technology giants face up to the possibility of AI to C for the first time. Tencent and Baidu have begun to fully access DeepSeek. With the assistance of DeepSeek traffic, Tencent Yuanbao once surpassed Doubao and topped the Apple free app download list.
Resisting the temptation of Alibaba and Byte, which are tempted to access DeepSeek, one chooses to rely on self-developed models to increase the C-end and try to invest in Tongyi APP, and the other chooses to link more closely with Douyin and Doubao to try to further do large-scale effects.
However, as the DeepSeek boom weakens, looking back at the 100-day node, the competitive situation of major technology companies has shown signs of returning to the starting point: Doubao has returned to the first position in the number of downloads of AI native applications, Yuanbao has been weak, falling out of the top 20, Alibaba’s emphasis on the C-end has not yet been released, and Baidu still insists on not investing in the flow and continues to develop slowly.
B-end product manager’s ability model and learning improvement
The first challenge faced by B-end product managers is how to correctly analyze and diagnose business problems. This is also the most difficult part, product design knowledge is basically not helpful for this part of the work, if you want to do a good job in business analysis and diagnosis, you must have a solid …
View details >
Although the C-end application download ranking has changed several times in the past 100 days, there is no doubt that many technology giants have become the biggest beneficiaries of the DeepSeek phenomenon, regardless of whether they connect to DeepSeek on the main C-end application or not.
On the one hand, the explosion of DeepSeek has driven the capital market to refocus its attention on domestic technology giants, and thus ushered in a new wave of revaluation opportunities; On the other hand, under the impact of DeepSeek, the former “AI Six Tigers” have almost no financing progress (the only exception is Zhipu), which has also made the outside world realize again that this battle for large models has entered the stage of giant games.
DeepSeek, which is like a mirror, not only allows the outside world to see clearly the competition elements of this new stage of the AI war, but also allows technology giants to see their respective focuses: Byte continues to do soybean bag scale, Alibaba pursues the leading position of open source models, and Tencent reveals a new strategy to continue to activate WeChat traffic diversion AI.
01
The unexpected rise of DeepSeek has allowed the outside world to see the possibility of AI assistants running through the scale effect, and no major manufacturer dares to risk losing the first-mover advantage and continue to wait. For the first time, major technology companies have launched all-round competition for AI assistants.
In this new competition, Tencent has become the most radical. Among the self-developed large model manufacturers, Tencent is the first company to connect DeepSeek to its C-end main product.
Prior to this, many technology giants chose to only access the DeepSeek model on their own cloud service platform for external developers to call, deliberately avoiding the situation of fighting with their own AI assistant products on the C-end.
In addition to binding the DeepSeek traffic password, Tencent Yuanbao has also taken a multi-pronged approach, not only getting the diversion of Tencent products such as WeChat and QQ, but also surpassing Doubao and Kimi to become the new investment king. According to AppGrowing data, from January 1 to March 13, 2025, the investment amounts of Yuanbao, Kimi, and Doubao were 710 million yuan, 150 million yuan, and 90 million yuan respectively, and the changes mainly occurred after February after accessing DeepSeek.
In the face of the current situation that has lagged behind Doubao and Wen Xiaoyan, Tencent, which has taken the lead, relied on the spillover DeepSeek traffic on one hand and the investment regardless of cost on the other, and gathered a large number of users for Yuanbao in a short period of time, thus becoming the biggest beneficiary of DeepSeek’s traffic dividend: On February 13, as the first Yuanbao to access the DeepSeek R1 full-blood version in the C-end main product, it once took advantage of the rise of DeepSeek Dongfeng, until March 3, It beat DeepSeek and Doubao to top the domestic Apple free app download list for the first time.
Subsequently, Baidu also connected DeepSeek in its own C-end main applications Baidu APP and Wen Xiaoyan. But as the second technology giant to choose to access DeepSeek, Baidu’s DeepSeek dividends are far inferior to Tencent’s.
According to QuestMobile data, as of March, Wen Xiaoyan had 8.87 million monthly active users. Compared with 12.24 million in December last year, the number of monthly active users of Wen Xiaoyan did not increase but decreased. It is worth mentioning that this is also related to Baidu’s insistence on not investing.
Among the technology giants, only Alibaba and Byte have resisted the temptation of DeepSeek’s dividends: in addition to accessing DeepSeek on their own B-end and niche C-end applications, whether it is Douyin, Doubao, Taobao, Tongyi and Quark, they have never followed up on DeepSeek.
This is directly related to the way the two companies play AI strategies. Before DeepSeek became popular, Byte’s AI assistant Doubao had become the AI conversation application with the largest number of monthly active users in China.
After DeepSeek became popular, Byte CEO Liang Rubo set a goal for the team for 2025, one of which is to require the team to strengthen the scale effect, and not by accessing DeepSeek.
To this end, Doubao has also begun a new path of urgent model functions: on February 25, a small-scale test of the self-developed deep thinking reasoning model, and on March 18, the deep thinking function was fully launched.
In addition to aligning with DeepSeek in terms of deep thinking functions, Doubao also continues to expand the linkage with the Douyin content ecosystem to enhance user stickiness and as one of the important means to compete against Tencent Yuanbao + DeepSeek competition.
The latest move is that Byte has expanded the scope of internal testing of implanting beanbags as contacts in Douyin message lists, guiding more Douyin users to download or use the Doubao APP.
Prior to this, Doubao had supported providing users with Douyin short video feature content in the generated answers, and tested a new feature in mid-March, that is, under the original dialogue content, users continued to swipe up the screen, and a content stream based on user interests and conversation recommendations would appear, which was comparable to creating a miniature version of “Douyin” in Doubao.
Different from Byte’s priority pursuit of scale effects, Alibaba cares more about the competition of model performance, and does not care about the traffic gains and losses of one city and one place under the long-termism of pursuing AGI.
Although according to AppGrowing data, since the end of February, Tongyi Qianwen has gradually increased its investment in advertising, Alibaba’s overall investment budget is still limited. This can also get a glimpse of the intuitive changes in monthly active users.
According to QuestMobile data, as of March, Tongyi had 3.82 million monthly active users. For comparison, in December last year, Tongyi was 2.91 million, and the growth rate of monthly active users was far less than that of Yuanbao and Doubao.
In a recent interview with “Late Dialogue”, when asked how he thought of the outside world ridiculing Alibaba as the “Wang Feng” in the field of large models, Zhou Jingren, CTO of Alibaba Cloud and head of Tongyi Laboratory, responded, “The traffic of a certain day is actually not that important.” ”
02
Regardless of whether they accept DeepSeek or not, large manufacturers have become the ultimate winners under the DeepSeek phenomenon.
Since entering 2025, with the explosion of AI startup DeepSeek, driven by it, Chinese technology stocks have rebounded, and the capital market has also re-examined the AI value of technology giants.
In the Hong Kong stock market year-to-date, Alibaba’s stock price has risen 47%, Tencent has risen 19%, and Baidu has risen 3%.
Byte, which has not yet been listed, has also returned to a high level. At the end of February, Bloomberg quoted people familiar with the matter as saying that Byte’s valuation had exceeded $400 billion, and at least three major investors had revalued the company, including SoftBank, Fidelity Investments and Puxin Group. The growth potential of Byte’s AI business “Doubao” is an important factor for investment institutions to be willing to push up Byte’s valuation.
Another positive factor for technology giants under the new opportunities of AI comes from Apple’s AI cooperation order.
In the latest revelations from foreign media The Information, Alibaba has been added to the cooperation list of the National Bank version of Apple’s AI. Prior to this, the names of Baidu, Tencent, and Byte appeared in this list.
The reason why almost all major manufacturers, even the popular DeepSeek, are difficult to shortlist, is because Apple’s requirements for service stability make it difficult for ordinary AI companies to support the manpower and experience required by large customers like Apple.
The Information revealed that Apple’s goal is to find a Chinese partner that can understand Chinese users’ requests based on their personal data. Under this premise, technology giants are undoubtedly Apple’s most ideal AI cooperation partners.
More importantly, under the competitive pressure of DeepSeek, the outside world has also seen the nature of the capital game of the current large model competition.
In January, Zero One Everything, one of the “Six Little Tigers of AI”, officially adjusted its strategy and became the first large model manufacturer to give up the pursuit of AGI.
In the face of the impact of DeepSeek, Kai-Fu Lee said in an interview in March that the pre-training of super-large models in China and the United States is gradually becoming oligopoly, and predicts that the future pattern of China’s large model field will converge to DeepSeek, Alibaba, and ByteDance.
Reality also verifies Kai-Fu Lee’s assertion of “oligarchization” to some extent. After the release of DeepSeek R1, the update of the entire large model field has almost become a one-man show for major manufacturers, including Byte, Tencent, Alibaba, and Baidu have released new models one after another, but AI startups have almost stopped competing in model updates.
Greater pressure is also reflected in financing. After entering 2025, foreign OpenAI, xAI and Anthropic have completed and are about to complete billions or even tens of billions of dollars in new financing.
DeepSeek has not changed the pursuit of computing power demand in the field of large models, but the competing players have further shrunk to the category of technology giants.
03
But the enthusiasm for DeepSeek has also begun to fade.
Radical like Tencent no longer pursues higher download rankings. The latest situation is that as of May 8, Doubao once again surpassed DeepSeek in the Apple App Store free APP download ranking in China, followed by DeepSeek was squeezed to the fifth position, and Tencent Yuanbao slipped to 25th place.
The change in download rankings is enough to illustrate a truth: in the battle of AI applications, DeepSeek can only save the moment, but it cannot be the silver bullet that seals the victory.
After all, under the current free strategy, ordinary users have not yet developed loyalty to large model products, and users’ minds are influenced by the quality of the experience. As various AI applications have successively added the chain of thought display brought about by the deep thinking function, users’ curiosity about accessing DeepSeek products has begun to fade, and the retention rate of these trial users who come out of curiosity is often low.
Greater fading, appeared on the model. While accessing DeepSeek, major technology companies are also starting to develop their own models and begin to distance themselves from DeepSeek step by step.
Even, some major technology companies have begun to turn their heads and complain about DeepSeek to stand for their own models. In a recent speech, Robin Li, the founder of Baidu, named DeepSeek, saying that it is not yet able to understand and generate multimedia content such as pictures, audio, and video, and the hallucination rate is high, and the bigger problem of using DeepSeek is slow and expensive, “The price of the vast majority of large model API calls in the Chinese market is lower than that of the full-blooded version of DeepSeek, and the speed is also faster.” ”
Robin Li’s previous bytes, and Alibaba after him, both focus on cost-effective advantages in their new models. But cost-effectiveness alone is obviously not enough to open up the market.
The reason why DeepSeek can get out of the circle, in addition to its low cost, is that one of its important premises is that the model ability has been innovated, and the wonderful existence of the chain of thought has been shown to the outside world for free for the first time.
When commenting on the explosion of DeepSeek, Altman also believes that the ability of the model is not really the key to DeepSeek’s virality, the key is that DeepSeek opens up OpenAI’s hidden chain of thought for free.
However, looking at the current number of new model products, in addition to comparing or catching up with DeepSeek and even OpenAI in various data, they have not yet innovated in product experience and functions.
In order to pursue innovative breakthroughs like DeepSeek, technology giants have begun to seek opportunities from both organizational and talent adjustments in addition to launching new self-developed models.
Byte poached Wu Yonghui, vice president of research at Google DeepMind, and recently further expanded the scope of the Seed department, and launched the Top Seed program for large model talents; Alibaba recruited Xu Zhuhong, the world’s top AI scientist, recalled former DingTalk founder Chen Hang, and continued to implement A Star and Al Clouder talent projects, launching a large-scale investment of 380 billion yuan in the next three years to build cloud and AI hardware infrastructure; Tencent launched the “Qingyun Project” for top AI talents and recently established two new departments: the Large Language Model Department and the Multimodal Model Department, which are responsible for exploring cutting-edge technologies for large language models and multimodal large models, respectively.
If in the past few years of large model melee, technology giants still grabbed their noses and eyebrows, now, with DeepSeek as a reference, they have gradually found their own main directions: Byte continues to move along the path of making soybean buns, Alibaba strives to maintain its leading position in the field of open source models, and Tencent shows a new momentum of relying on WeChat to seize the entrance to AI applications.
However, the current pace of AI competition among major technology companies is still facing the impact of a new variable – DeepSeek R2. With the R2 model just around the corner, it’s a big test not only for DeepSeek, but also for tech giants.
