Your Position Home AI Technology

Internet giants under the impact of DeepSeek: Who is rising and who is falling behind

 

Wen| Narrow Broadcast, author| Li Wei (Beijing)

The WeChat search grayscale test connected to the DeepSeek R1 full-blood model, which dropped another blockbuster on the AI industry.

This made Tencent the first Internet giant to integrate the DeepSeek model into its core business, delivering a wave of the most valuable advertisements to WeChat Search, and allowing DeepSeek to truly connect to a national portal that can reach 1 billion users.

This also means that DeepSeek is changing the rules of the game in the AI field. Prior to this, both the AI Six Little Dragons and the Internet giants were “relying on models to respect themselves” and relying on model capabilities to provide centripetal force to build their own AI ecosystem. After that, model capabilities remain important, but the AI ecosystem increasingly relies on huge resource supply and excellent engineering capabilities as support.

Internet giants under the impact of DeepSeek: Who is rising and who is falling behind插图

On the one hand, the giant accepts more spilled traffic by connecting to DeepSeek. Under the temptation of such huge traffic, DeepSeek did not take the initiative to expand its user base, nor did it take measures to alleviate the busy server problems encountered by users during their use. It only made actions such as registering trademarks and adding new scope of Internet information services.

This has resulted in huge C-side traffic being undertaken by more companies with service capabilities. Compared with the AI Six Dragons whose products such as One, MiniMAX and Step Star are still in the user growth stage, the Internet giant has provided users with more mature products such as WeChat Search, Baidu Search, Dingluo, Ali International Station, and Feishu. Connect to DeepSeek, eating up more spilled traffic.

On the other hand, under the trend of equal rights in large-scale model technology, the importance of giants is increasing. If the open source model can maintain its current strong capacity growth, it is likely that the AI field will evolve.“Core model companies + platform giants + small and medium-sized developers”The three-layered structure.

The core model company at the lowest level is responsible for providing the underlying technologies and frameworks for building an AI ecosystem similar to Android. Small and medium-sized developers are responsible for exploring specific opportunities in vertical scenarios. Platform giants rely on resource advantages such as computing power and data to connect the two., driving the backbone of ecological development.

DeepSeek provides cheap and easy-to-use shovels in an open source manner.On the one hand, Internet giants will, like entrepreneurs, expect to use this shovel to dig out gold mines; on the other hand, they will also refer to this shovel to create a more convenient new shovel for themselves.

In this context, there will be giants represented by Byte that will try to use new shovels to span multiple levels to establish a closed loop in the fields of underlying technology, resource supply and vertical scene development, and become Apple in the AI era. But more giants will be like this DeepSeek access wave, breaking the strong binding between large model capabilities and super portals, and using a combination of open source models and self-developed models to serve their own super portals.

From the different attitudes towards DeepSeek, we can more clearly see BAT’s different AI development strategies. The differentiation of competitive strategies will bring about a restructuring of the pattern at this turning stage and will also lead to some people falling behind.

Tencent, looking forward to the old tree sprouting new shoots

Among Internet giants,Tencent’s embrace of DeepSeek was the most decisive and resolute。Over the past weekend, many people around me have experienced the WeChat AI search function connected to DeepSeek R1. From an experience point of view, DeepSeek-R1 is more used to analyze and understand the content of Weixin Official Accounts, and then provide users with search answers. It is a tool to help users tap the treasure of WeChat content.

Prior to this, Tencent’s own big model was not given this role, largely because past big model products could not give WeChat users a significant upgrade to their search experience like DeepSeek-R1. Although Tencent Yuanbao was previously able to realize AI search on public account content, it was unable to achieve stronger question understanding and answer summary like DeepSeek R1 in terms of content summary.

Internet giants under the impact of DeepSeek: Who is rising and who is falling behind插图1

For WeChat, which has always been cautious about product feature updates,This high-profile operation demonstrates Tencent’s strategy of actively embracing AI applications。As a company with strong product genes, Tencent adheres to the principle of giving priority to application experience. Before a sufficiently excellent large model experience appears, it does not easily put in its own chips.

Once the experience exceeds the baseline, Tencent quickly connects products such as WeChat Search, Tencent Yuanbao, Tencent Maps, QQ Music, and Tencent Documents to DeepSeek.In one fell swoop, it reversed the outside world’s judgment that his actions were slow and his strategy was not clear enough in the field of large models, from a follower of a large model development to an important player with a latecomer advantage.

Ma Huateng once judged: “In an industrial revolution like AI, which is unprecedented in hundreds of years, it is not so important to take out the light bulb a month earlier.” This also shows that Tencent wants to seize the core industry opportunities that have emerged in this industrial revolution. In this process, it is not important whether we rely on self-developed large models and whether it is an AI-native product.Tencent will give priority to how to build user minds in the AI era in existing core scenarios.

This kind of thinking is also in line with the logic of “old trees sprout new shoots” that Ma Huateng emphasized at Tencent Group’s earnings conferences in the past two years. Developing new shoots means that old products can have the vitality of sustainable growth through the innovation of products and functions. Take WeChat as an example. Video numbers, Mini programs, Mini games and search are all new shoots that grow from it. andAI is bringing new opportunities to more Tencent products.

The combination of AI and search is a reflection. Search is a super entry that Tencent has never been able to obtain in the era of mobile Internet, and it is also a new sprout that WeChat has been cultivating and growing. When DeepSeek broke the existing balance of AI search and took the user experience a big step forward, Tencent did not hesitate to play the trump card of WeChat AI search, causing a sensation in the industry.

For Tencent,This is also an opportunity to leapfrog growth in the original search mindThrough this wave of operations, more people have begun to have a more specific understanding of WeChat search. At the same time, shortly after Tencent connected to DeepSeek on WeChat, it launched its self-developed Mixed Origin T1 deep thinking model on Tencent Yuanbao to create independent model capabilities around AI search scenarios.

The combination of AI and games is also an opportunity that Tencent wants to seize to develop new shoots. Some developers told us that Tencent is now anxious about how to apply AI to the game business as soon as possible. In the creation process, Tencent recently launched the open source Mixed 3D Generation Model 2.0, and simultaneously launched the Mixed 3D AI Creation Engine, which can shorten the production time of some 3D assets from 5 – 10 days to minutes.

In terms of operations,AI is also expected to improve players ‘gaming experience。Through the development of intelligent NPC and intelligent man-machine technology, AI can realize in-depth interaction between players and NPCs, driving the growth of time, payment and ARPPU. In August 2024, Tencent Rubik’s Cube Studio teamed up with the Arena Breakout project team to launch voice-commanded FPS AI, allowing AI teammates to adapt to more props, weapons, scenes and more complex battles.

Compared with core business brewing new shoots through AI,Tencent is more restrained about new AI products such as Yuanbao and ima, and does not invest much promotion resources.。Also a positioning assistant, Yuanbao neither conducts large-scale investment like Doubao and Kimi, nor attempts to attract users through continuous functional innovation like Tongyi. ima, which is also connected to DeepSeek, is also a product that passes through the public’s eye and does not pursue explosive growth.

Ali, a high-profile and all-round attack

Ali is one of the biggest beneficiaries of DeepSeek’s popularity. The news of Ali’s investment in DeepSeek has attracted more people’s attention to Ali’s investment and achievements in AI, establishing Ali’s good reputation in the AI field. Although Ali subsequently denied its willingness to invest, it is still considered to be a target company capable of providing computing support and conducting in-depth cooperation to DeepSeek.

At the same time, Apple and Ali cooperated, and Ali developed artificial intelligence functions for the China version of the iPhone, further providing key customer endorsement for Ali’s AI capabilities. Cai Chongxin said in response that picky Apple had talked with many companies in China and finally chose Ali as a localized AI partner. For Ali,Stabilizing Apple as a major customer means the access of hundreds of millions of hardware users and the opening of AI hardware doors.

Ali, which has shifted its innovation breakthrough from new retail to AI, seems to have regained some of the ideals it once had, and devoted more technical vision and longer-term vision to specific business operations. Under the first two benefits,Ali’s business logic is being restructured: from the logic of e-commerce + cloud services to the logic of e-commerce +AI infrastructure providers.

First of all, through integration and with the support of AI technology, AliCreating the prototype of a smooth global e-commerce networkLet brands, sellers and supply chains do their jobs well in this network, obtain more certain benefits, and at the same time meet diversified consumer needs. Among them, Ali Mom, 1688, Taotian, and Xian Yu are all rapidly exerting the effectiveness of AI in their respective business areas.

Ali’s mother is using AI technology to give merchants more accurate understanding of consumers, and using AI to help merchants achieve more stable and convenient creative material generation and marketing material delivery based on insight into trends; after 1688 connected to DeepSeek, it tried to build a business model; idle fish positioned itself as an AI technology testing ground within the Ali system.Realizing AI is not only a kind of political correctness, but also carries the real needs of business development.

Secondly, unlike Tencent, while Ali expects AI to make “old trees sprout new shoots”,The layout of the AI field will be more high-profile and comprehensive。In terms of big models, cloud platforms, to B and to C applications, Ali has shown a relatively clear approach.

In terms of large models, data shows that Ali’s Tongyi Thousand Questions Model has become one of the largest open source model groups in the world, with the number of derivative models exceeding 90,000. The newly released Qwen2.5 Max model has performed well in the fields of mathematics and programming. The above-mentioned developer told us that his company currently only opened the APIs of Tongyi Qianwen and DeepSeek.”Before DeepSeek came out, Tongyi Qianwen really did a good job.”

Internet giants under the impact of DeepSeek: Who is rising and who is falling behind插图2

Qwen2.5-Max benchmark scores

In terms of cloud platforms, Alibaba Cloud is the world’s fourth largest cloud computing service provider after Amazon, Microsoft, and Google. It covers more than 200 countries around the world and has the ability to provide API interfaces for large model manufacturers and provide external reasoning services. At the same time, through the large model application development platform Alibaba Cloud Bailian, it helps developers create exclusive AI and large model applications.The combination of cloud platform and large model capabilities together constitute the AI base for Ali to support applications.

At the application level, Ali has products such as Nail, Quark, and Tongyi. Among them, Dinghao is the first product that Alibaba has access to DeepSeek, allowing users to build DeepSeek-based AI assistants themselves. At the same time, Dinghao also plays the role of providing AI toolkits to B-end users, helping Ali link the AI base with the organizational management, data precipitation, and business development needs of B-end users.

Ali also showed new determination to explore in the direction of AI to C。On the one hand, Ali classified Tongyi App into the intelligent information business group, and the Tmall Elf team was also managed by Wu Jia, president of the intelligent information business group. On the other hand, Ali continues to recruit talents for the AI to C business. Just last week, artificial intelligence scientist Xu Zhuhong joined Ali, responsible for the multimodal basic model of the AI To C business and basic research and application solutions related to Agents.

In this way, Ali once again integrated its AI to C resources with Wu Jiahe’s intelligent information business group as the core. Among them,Quark, who has the largest user size and has been updated as an AI all-round assistant, is most likely to become the core of Ali’s AI to C businessIntegrating its own search accumulation, Tongyi App’s AI function research and development capabilities and Tmall Elf’s hardware capabilities, Ali has built a super portal spanning multiple ends in the AI era.

Cai Chongxin pointed out at the World Governments Summit 2025 that the explosion of DeepSeek has stopped people from investing tens of billions of dollars in computing infrastructure, but instead focusing on actual problems in the world and using AI to solve them. More small models may emerge in the future, and many people are beginning to invest resources in application areas.

Judging from the current layout, Ali is more likely to become the infrastructure contractor behind this trend. At the same time, there is also a certain probability of having a super portal for the C side in the AI era.

The old king is in danger, but the new king rises

Baidu and Byte both maintained a certain degree of restraint in the face of DeepSeek, but Baidu was forced by the situation and chose to access DeepSeek for search, while Byte was reflecting on why he didn’t make R1. The different attitudes reveal the different development paths of the two AI businesses. Some have seized new opportunities, while others have begun to face more severe challenges.

With the rapid rise of bytes in the mobile Internet era, many people began to replace the B in BAT from Baidu to bytes. However, the prosperity of the big model has caused a new deflection in the direction of B. Baidu took the lead in keeping up with the pace of OpenAI, released Wenxinyan very early, and even became Apple’s first selected partner. However, Byte was slow to respond to the big model and became a chaser.

More than a year later, the situation reversed again, and the leading Baidu was gradually catching up with bytes. Whether in terms of the volume of the large model or the user scale of the assistant product, the new king byte has once again surpassed the old king Baidu. Data from the AI product list shows that Wen Xiaoyan’s MAUs in January 2025 were 13.05 million, far lower than the 78.61 million MAUs of Doubao, ranking fourth.

Especially after the outbreak of DeepSeek, Li Yanhong’s previous assertion that “closed-source models must be more powerful than open source models and have lower reasoning costs” was overturned; Baidu’s business model of large-model applications built based on its membership business also faces great challenges.Baidu is forced to embrace free and open source in the face of reality, and under the pressure of WeChat Souyisou, it followed up and announced Baidu Search’s plan to access DeepSeek R1 and Wenxinyiyan in-depth search functions.

Internet giants under the impact of DeepSeek: Who is rising and who is falling behind插图3

At the same time, the cooperation between Apple and Ali was confirmed, which further weakened Baidu’s accumulated reputation in the AI field.Even with relatively successful AI-based products such as Baidu Library as support, Baidu’s share price in the capital market still fell。Yanhong Li also began to admit that innovation cannot be planned.”You don’t know when and where innovation will come. All you can do is create an environment conducive to innovation.”

As a byte of the new king, after an initial slow response, it quickly caught up with the development of the big model by investing heavily in resources and clarified the three AI business development paths of basic big models, general assistants and AI applications. But in the face of DeepSeek’s onslaught,Byte CEO Liang Rupo started the reflection mode again at the new All Hands meeting:”If we take the lead on major issues at the beginning, we have a chance to achieve them earlier.”

When struggling with why he was not the first to achieve innovation based on the long-chain thinking model,Byte also shows to a certain extent its willingness to become a leader in AI innovation and even build a complete closed-loop ecosystem from large models to cloud services, from applications to hardware.

At present, Byte’s Volcano Engine has already supported DeepSeek model products, and Feishu has also quickly completed its connection to DeepSeek. But unlike Tencent Yuanbao’s simultaneous access, Byte is likely to not rely on the open source DeepSeek to achieve the deep thinking capabilities of bean buns, but will accelerate the launch of its own deep thinking model.

Liang Rupo also further clarified the strength of Byte in the AI business at the all-staff meeting: First, we must get rid of the pursuit of specific product DAU, regard intelligence itself as the most important goal, stimulate more attempts and not ignore key technology nodes; Secondly, explore wearable, more portable, and more natural hardware interactions; Finally, strengthen the scale effect that AI products are easier to use as more people use them.

As Professor Li Feifei said, historians will definitely call this period the “real first AI era” in the future. At this point in time, DeepSeek’s impact accelerated the division and change of giants in the old era. BAT’s various responses are part of the narrative of the change of the times and a piece of the jigsaw puzzle that makes up the new era.

Popular Articles