Tencent (00700) AI, quietly accelerating.

In the Chinese AI battlefield of 2025, it has evolved from a battle of model parameters to a comprehensive game involving capital efficiency, infrastructure, and traffic entry. According to reports from Wall Street News, a unique three-kingdom strategy has emerged in China. Alibaba has chosen heavy investment in assets, ByteDance has chosen to break through in traffic, and Tencent has chosen a strategy of "internal verification + external empowerment." Tencent completes polishing within its internal ecosystem, then exports its capabilities to users through a plug-in mode. To outsiders, this strategy may seem too restrained or even passive. However, Tencent's AI is quietly accelerating. On December 17, according to reports from Wall Street News, Tencent announced the upgrade of its large model development architecture by officially establishing the AI Infra Department, AI Data Department, and Data Computing Platform Department. The newly established AI Infra Department will serve as the foundation of Tencent's large model system, responsible for core technologies such as distributed training of large models and high-performance inference services. This signifies Tencent's comprehensive strengthening of its large model development system from an organizational structure level. Furthermore, it was announced that Vincesyao has been appointed as the Chief AI Scientist of the CEO/President's Office, reporting directly to Tencent's CEO Liu Chiping; he also serves as the head of the AI Infra Department and the Large Language Model Department, reporting to the President of Tencent's Technology Engineering Group, Lu Shan. This flattening of reporting relationships reflects Tencent's efforts to enhance AI research efficiency and signals its intention to occupy a position in the AI landscape. This move is similar to Google's approach. Previously, Google announced the integration of the AIStudio and Gemini API teams into DeepMind, led by Nobel laureate Hassabis and reporting directly to CEO Pichai. This was done to unify the feedback loop between model training and application development and accelerate iterations. On the same day, Tencent's Mixian World Model 1.5 (WorldPlay) was officially launched. This is not just a video generation tool but the first in China to support "real-time interaction" within a world model. It breaks the traditional limitations of AI generation as users can input text or images to generate a personalized 3D virtual world and freely explore it using a keyboard and mouse like playing a game. Tencent's AI Rhythm Reviewing Tencent's AI strategic path over the past three years can be summarized into three stages: Before 2023, it was a typical "evaluation period." On one hand, Tencent's Chairman and CEO Ma Huateng defined AI as "an opportunity that comes once in a few hundred years." On the other hand, they maintained a defensive posture of "no rush," focusing core resources on fundamental technology research and scene exploration rather than rushing towards commercialization. While the market evaluates the performance of various tech giants' AI based on metrics like daily active users and rankings, Tencent steadily maintained its position among the top tier companies. A turning point appeared in 2024 as breakthroughs in domestic model capabilities like DeepSeek and the initial formation of the market landscape led Tencent to quickly adjust its strategy towards a multi-model approach of "firmly investing in self-developed models + open embrace of advanced open-source models." According to Tencent, whether it is the self-developed Mixian or external DeepSeek, as long as it creates value for users, it is a good tool. After the emergence of DeepSeek last year, Tencent became the most proactive company in embracing DeepSeek, integrating DeepSeek models into several key consumer products and becoming the first top-tier company to introduce DeepSeek into its main consumer products. Entering 2025, Tencent has officially entered the "active offensive phase" through organizational restructuring and ecological penetration. The recruitments of talent from OpenAI and the reorganization of the Mixian team are the most prominent signals of this stage. This proactive move is not a delayed action but a precise plan to minimize risks. Tencent is accustomed to waiting for the market path to be clearly validated before accelerating full force as in the mobile payment battle and the cloud service competition in the past. To support this competition, Tencent is deepening its internal efforts. Over the past year, Tencent's Mixian large model has significantly accelerated its iteration speed. It has iterated over 30 versions in a year, especially in areas such as text-to-image modeling and 3D modeling, maintaining a globally leading position. In the past two months, Tencent has successively launched Mixian 2.0, Mixian 3D 3.0, and Mixian Image 3.0, with Mixian Image 3.0 ranking first in the latest text-to-image leaderboard in the international large model competition, LMArena. What is interesting is Tencent's pragmatism. Unlike other tech giants, Tencent adopts a model of "self-developed + plug-in." While insisting on self-developed models, Tencent provides more open choices for users and customers. The underlying support for these choices is based on Tencent's engineering capabilities that create user-friendly products. For example, in core products such as WeChat and Yuanbao, Tencent not only integrates its self-developed Mixian but also actively introduces external open-source models like DeepSeek. This strategy has sparked two different opinions in the industry: some believe it reflects Tencent's open mindset, emphasizing efficiency; others worry whether the traffic king can counterbalance the technological gap caused by rapid iterations. In the deep waters of AI, excessive reliance on open source or external technology may lead to a lack of underlying competitive advantages. In response to this, Tencent seems to focus on first getting AI to work internally, following the principle of "eating its own dog food." Before outputting its large model capabilities externally, Tencent aims to make AI an "engine" that improves its own business efficiency. It is reported that Tencent's AI has been fully internalized, with the Mixian large model deployed in over 900 scenarios and applications internally. Particularly in core business scenarios such as advertising and gaming, AI has significantly increased efficiency for Tencent. Ma Huateng pointed out that "Tencent's strategic investment in AI not only benefits Tencent in fields such as precise ad targeting and gaming engagement but also improves efficiency in programming, gaming, and video production." The capital market also recognizes Tencent's high-certainty participation in AI. According to a report by Sealand, based on the SOTP valuation method, the total target market value of Tencent's various businesses in 2026 is estimated to be RMB 6.5 trillion / HKD 7.1 trillion, with a target price of HKD 773, maintaining a "buy" rating. Emphasizing applications, with a focus on light rather than heavy If Alibaba is betting on the thickness of "computing power," Tencent is playing the "super-connection" eco-card. With over a billion users in its ecosystem's social, office, and consumption behavior data deeply integrated into platforms like WeChat and QQ, Tencent has created a closed-loop of "data feeding model optimization, model enhancing user experience," which is difficult for companies reliant solely on technology or traffic to emulate. As an analyst from Guosen mentioned, Tencent's natural advantage of users and scenes in its ecosystem makes it one of the best-positioned companies in the AI era. In Tencent's view, the final competition of the AI era is not simply about entry points but about the penetration rate of scenarios. Many AI Chatbots on the market face challenges of high try-out rates and low retention, mainly due to a lack of real scenarios. This is where Tencent's biggest moat lies it is building a "social service AI" closed-loop system, making AI ubiquitous yet invisible like the elephant in the room. For example, on the consumer end, Yuanbao is no longer just an app but an intelligent central hub for this "connection" card. On one hand, Tencent continues to optimize the user experience of Yuanbao, even maintaining a daily iteration frequency in the initial stages, quickly gaining user recognition. Yuanbao has become one of the top three AI-native apps in China, becoming one of the most popular AI-native apps among new middle-class consumers. On the other hand, the strategic intent of Yuanbao is not merely to create a new super entry point but to become the "connector" of Tencent's ecosystem. Earlier this year, Tencent's President Liu Chiping clearly stated that Yuanbao's future does not rely solely on pouring in traffic but on organic integration with Tencent's existing products. It is evident that Yuanbao is transitioning from a solo breakthrough to deep integration with the ecosystem: For example, connecting social activities by becoming part of WeChat contacts list or integrating with QQ "chat summaries"; connecting content by linking with QQ Music's "search songs, listen instantly" feature or integrating with Tencent Video to build a complete viewing experience; and further connecting services by entering public accounts and video comment sections, allowing users to directly interact with Yuanbao. Users using super apps like WeChat or QQ Music may not even realize they are "using AI" because AI capabilities are integrated into these products like organs in a body. If Yuanbao and other AI-native products on the consumer end showcase Tencent's AI capabilities in the front, Tencent Cloud serves as the backbone behind the scenes. For most enterprises, training a large model from scratch is neither practical nor necessary they truly need intelligent tools that can be quickly generated, deployed with low barriers, and directly address business pain points. Therefore, in the middle of this year, Tencent Cloud launched the Intelligent Body Development Platform (ADP). This platform's strategic positioning is clear: it greatly lowers the barrier for businesses to adopt AI through BC linkage. Leveraging the underlying capabilities of Tencent's Mixian large model, enterprises do not need to assemble a large algorithm team. Using a drag-and-drop "low-code" or even "no-code" approach, businesses can build custom customer service assistants, data analysts, or code writing assistants in minutes. This Lego-like development experience turns AI from a lofty tech concept into a readily accessible production tool for enterprises. Furthermore, it is a continuation of Tencent's "connection" gene on the business end. Unlike other cloud service providers, Tencent Cloud's Intelligent Body Development Platform has a unique advantage integration with the WeChat ecosystem. Smart agents or entities developed by enterprises on this platform can be seamlessly deployed to WeChat public accounts and mini-programs with a single click. This means that a "smart guide" developed by a travel company can directly answer user inquiries within WeChat, and a "financial assistant" from a bank can directly serve VIP customers on the enterprise WeChat platform. This integration of B-end development with C-end engagement creates a "C2B" link that is the most lethal part of Tencent's "social service AI" iron triangle strategy. Lastly, Tencent Cloud's deep dive into industrial scenarios is not stuck at the conceptual verification stage but has packaged AI capabilities into concrete industry solutions like code assistants, intelligent customer service, and meeting summaries. For example, in over 30 typical scenarios in finance, government affairs, and travel, Tencent Cloud's AI code assistant has assisted in creating 50% of new code within Tencent's internal business lines. This industrial capability, validated through a vast amount of internal business scenarios, is now spilling over to external industries through the cloud platform. In summary, Tencent's "super-connection card" strategy is clear: using products like Yuanbao to internalize AI into the daily lives of billions of users and the cloud platform to externalize enterprise AI into the Tencent ecosystem. From embedding AI into daily life on the consumer end to empowering businesses on the enterprise end, Tencent is not fighting for a singular victory but is attempting to reconnect everything with AI. Is Tencent Replicating a "Google-style" Counterattack in China? Overseas, Google's strong ecosystem integration capabilities have led Gemini to have a ubiquitous presence through Google Search, Cloud Services, and Android systems on a global scale. As analyst MoffettNathanson mentioned: "Google has finally played its hand well; search, cloud, hardware, and AI have formed a closed loop, an advantage that OpenAI will never be able to replicate." Tencent seems to be replicating this path in the Chinese market. With the richest internet product matrix in all of China, Tencent's AI strategy is increasingly showing a sense of "endless weight." As technological dividends are exponentially unleashed through platforms like WeChat and QQ, the resulting network effects may create a "data-scenario-technology" flywheel that other companies find difficult to replicate. The core DRIVE of this flywheel lies in Tencent's "dual-track Agent" strategy. Tencent's management has revealed that future intelligent bodies will be divided into two categories: one is the General Agent (such as Yuanbao, IMA), responsible for solving general problems in an open world; the other is the Ecosystem Agent, perhaps this is Tencent's true "nuclear weapon." As Tencent President Liu Chiping mentioned at a recent financial report meeting, WeChat will eventually launch an AI intelligent body based on its unique ecosystem. It will rely on five major ecosystems communication, social, content, payment, and commerce to truly understand user intent and execute tasks in a closed loop. This smart entity, deeply embedded in WeChat, is the true "ideal assistant for users" and a differentiation barrier that other companies cannot replicate. Of course, the prosperity of an ecosystem cannot hide the underlying challenges. The speed of iteration in the AI era far surpasses that of the mobile internet, and the flourishing of applications must be built on a solid foundation in the model layer. While Tencent's current strategy may reduce short-term risks, in the long run, its ability to maintain a leading or even advanced position in underlying technology will determine the depth of its moat. The ecosystem is the moat, but technology is the foundation. In this context, Tencent is doubling down on AI investments, bringing in top research talent, and adapting to a flatter, more efficient organizational structure. Ma Huateng once said during the big model boom: "It doesn't matter if you take out the light bulb a month early; the key is to do the underlying algorithms, computing power, and data well, and most importantly, landing scenarios." In this AI marathon, the starting gun has long been fired. Some are sprinting to claim their positions, while others are adjusting their breath. Tencent, with the power to connect hundreds of millions of users, clearly belongs to the latter. Today's Tencent no longer hides its ambitions it does not want to be the fastest runner but aims to be the last one standing at the table. Perhaps we should never underestimate the strategic resolve of an enterprise that continuously navigates through technological cycles.