GuShiio.com Wen| Silicon-based laboratory, author| kiki
A month has passed since DeepSeek exploded, and the impact caused by this start-up is still spreading. In particular, China Internet companies represented by BAT and Byte are deeply embracing DeepSeek’s “circle of friends.”
In addition to binding their own cloud services to DeepSeek, on the more competitive application side, major manufacturers have changed their previous defensive stance and are quite active in connecting DeepSeek in their main positions.
Recently, WeChat was exposed to the new “AI Search” function after grayscale testing was connected to the DeepSeek R1 model. Baidu also quickly announced on the same day that it would fully access the latest deep search functions of DeepSeek and Wenxin Model in its search business.
WeChat is Tencent’s basic social ecosystem, and search is Baidu’s core business.This also means that DeepSeek is reaching deep into the business hinterland of major manufacturers and bringing a new round of AI restructuring.
However, active embrace is just “face”, and achieving large-scale implementation of applications is the “essence”. In 2025, technological innovations such as new algorithms and architectures around computing power and models will promote the long-term decline in the cost of large models, and the application of large models will be more critical. Silicon Valley investor Zhang Lu said:“Everyone’s attention has gradually shifted from the rapid growth of AI models to practical applications, that is, how to quickly implement AI technology and achieve large-scale industrial applications. rdquo;
In other words, DeepSeek is more like the best card played by major manufacturers at the beginning of the year. They all want to take advantage of DeepSeek and win again. nbsp;
Tencent and Baidu are both planning “new people’s designs”
Facing the “catfish” DeepSeek, there are two most hotly debated and proactive among China’s major Internet companies:Baidu and Tencent.
After the release of DeepSeek, Baidu not only announced that it would abandon the closed-source route and be completely free of charge, but Baidu founder Robin Li also responded to DeepSeek’s explosion at the 2025 Dubai World Government Summit: 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. rdquo;
Faced with DeepSeek’s proactive attitude, Baidu is obviously using DeepSeek to take the lead in shaping the mentality of “open source while free” among users and developers, using new people to increase their bargaining chips. nbsp;
Tencent has sought stability in its AI layout, especially the “AI transformation” of WeChat has always adhered to Zhang Xiaolong’s restraint. However, within less than a month of the release of DeepSeek R1, WeChat quickly integrated and opened up internal applications, creating a new AI search portal in a relatively closed ecosystem, which is very rare compared to WeChat’s slow feature updates in the past. nbsp;
But in fact, before WeChat connected to DeepSeek, Tencent began to apply it to applications such as to to B and to C as early as February 1, such as knowledge base products ima and Tencent Yuanbao. An industry source commented:“WeChat +Deepseek, although it is still in grey testing, has already reflected Tencent’s change in attitude.” rdquo;
Ma Huateng, Chairman and CEO of Tencent’s Board of Directors and CEO, also previously made a forecast on Tencent’s AI exploration at the 2024 annual meeting. On the one hand, Tencent will continue to invest resources to reserve computing power. On the other hand, products such as WeChat, QQ, and Input Methods will also launch AI agents. Products such as browsers, and games, WeChat reading, and Tencent videos will also do more AI exploration based on mixed elements. nbsp;
DeepSeek is the face, and “application scale implementation” is the essence
On the issue of “whether to accept DeepSeek”, the big companies want to use DeepSeek to promote the large-scale implementation of large-scale model applications rather than rudely think that this is a “defensive action” when facing competition. nbsp;
Marc Andreessen, co-founder of Andreessen Horowitz, mentioned when talking about the landscape changes brought by DeepSeek: For winners and losers at a point in time, the winners are all users, and the losers are companies with proprietary models.” nbsp;
Baidu and Tencent have collectively tried to connect DeepSeek model capabilities to their hinterland because they saw the possibility of application scale brought about by the combination of better model capabilities and their own products.——On the C-end, better models, lower prices, more friendly interaction forms and experience will most directly bring more user scale and activity, and indirectly increase users ‘paying propensity and conversion opportunities; on the B-end, which is already heavily paid and focuses on ROI, the improvement of model capabilities will drive the increase in token usage.A cloud vendor salesperson mentioned that benefiting from DeepSeek’s “computing power affirmative effect” has spurred the consulting craze since the beginning of the New Year. nbsp;
However, for large manufacturers, the cloud business focuses on capital expenditures and scale. Internet cloud vendors have also launched many rounds of price wars before. This round of borrowing DeepSeek also uses low-cost APIs to roll services and prices, and will not change the past cloud vendors. MaaS model. nbsp;
Therefore, compared with the impact of DeepSeek on the B-side, the more intense battlefield may still be on the C-side.
Since last year, major manufacturers have been looking for “super applications” in the AI era through internal horse racing or external investment. Among them, Byte’s bean buns, Baidu’s library, Ali’s quark, etc. are all AI products that have surfaced, but DeepSeek’s growth rate on the C-end still exceeds people’s expectations. nbsp;
According to the data of the AI product list as of January this year, domestic AI products currently rank among the “Ten Million Monthly Live Club” are: Doubao (78.61 million MAUs), DeepSeek (33.7 million MAUs), Kimi (19.43 million MAUs), Wen Xiaoyan (13.05 million MAUs).In other words, DeepSeek completed the KPI for the big factory for more than a year in 20 days.
According to “Tiger Sniff” report, DeepSeek is also forcing Ali to accelerate the overall progress of “AI to C”. Relevant sources mentioned that DeepSeek’s popularity in the C-end market has made Quark’s team feel both hope and anxiety. On the one hand, they realize that “AI to C is far from the end,” but on the other hand, competition for C-end users is intensifying. nbsp;
At Byte’s latest staff meeting, Byte CEO Liang Rupo also mentioned that in 2025, AI will focus on strengthening scale effects. AI products should be used by more people, the better they can be used. Now this feature on bean buns is not clear. rdquo; Liang Rupo said. nbsp;
Judging from the trend of C-side applications, Agents will be a key position for competition among large manufacturers. Let alone Baidu, which already regards Agents as the best form of application, judging from the basic form of WeChat’s implementation of DeepSeek, it may hide WeChat Agents. ambition——Use the WeChat portal and ecosystem, as well as other products such as Tencent ima knowledge base to connect content, services and search, making WeChat Agent a “new atomic component” in personal or corporate workflows. AI Agents and others will be the key direction for Quark to make efforts in 2025.
It can be said that in 2025, the large-scale implementation of applications, especially Agents, will be the main direction of major manufacturers. nbsp;
It’s time to compete in innovation efficiency
Wang Xiaochuan once commented on this round of big model competition: big innovations rely on small factories, and small innovations rely on big factories. However, whether innovation subverts or not, one point people easily overlook is that effective innovation relies on technological idealism or some luck, but It is also a science of efficiency, which has been highlighted in DeepSeek. OpenAI founder Sam Altman also predicted:
“In 2025, we are playing an incredible new computational efficiency game.& rdquo;
This also means that big manufacturers will shift from the past open-ended and open-ended to focusing on efficiency and results in terms of talents, products and organizational forms. In the words of Zhang Yiming, founder of ByteDance, this is a kind of “pragmatic romance.” nbsp;
“Innovation efficiency” is mainly reflected in streamlining goals, streamlining models and streamlining organizations.& nbsp;
In terms of goal streamlining, large factories will pay attention to the business income and profits of the C-terminal on the one hand, and internal horse racing will continue.”Photon Planet” once disclosed Baidu’s OKR in 2025. Yanhong Li has clear requirements for Wenxinyan’s revenue and profits., and there are specific scale requirements for API calls. On the other hand, in measuring user indicators on the C-side, in addition to user growth, retention, AI penetration, payment scale, etc. will become new focuses. nbsp;
In terms of model streamlining, DeepSeek asked major manufacturers to collectively reflect on the past model of “vigorously creating miracles” of stacking resources. Major manufacturers began to return to first-principles thinking that “a good product should be used by more people, the easier it will be to use” and see the users behind the product.”Real needs”. From this point of view, in 2025, Dachang will still carry out more exploration of software and hardware on the C-side, and will pay more attention to seizing access and interactive value.——On the hardware side, through the collaboration of cloud, end, and edge models, we find opportunities in mobile phones, glasses, etc. The cooperation between Apple and Ali has given new highlights to the competition for AI mobile phones in 2025; on the software side, the focus is on Multimodal Progress in areas such as Agents, and AI programming.
In terms of organizational streamlining, bidding farewell to dazzling technological research and development and embracing efficient innovation has become a new consensus among major manufacturers. For example, Quark has clear requirements for “simplicity, efficiency, and directness”. Baidu’s OKR has also repeatedly mentioned rejuvenation and organizational collaboration.& nbsp;
However, advocating efficiency is one thing, but specific implementation is another. Big factories always know it is easier than it is to do it, but they want to use AI to achieve reassessment. How to seize the window opportunities brought by DeepSeek within the same consensus on application scale? To a certain extent, big factories are at a new starting line. nbsp;
References:
1. Tiger Sniff: Ali hid a trump card
2. Photon Planet: 2025 Baidu OKR update: reduce costs and increase efficiency, develop KPIs, and encourage horse racing
3. Late: Byte all-staff meeting: Revisiting pragmatic romance and becoming a technology company that can invent new technologies
