The "lobster" AI application that has exploded across the Internet recently has become the focus of Tencent CEO Ma Huateng.
"After the 'lobster' type application was launched at the beginning of this year, the industry quickly realized that in addition to the two major directions of ToC-side chat robots and ToB-side AI programming, AI has opened up a new office scene track. When the office scene is deeply integrated with various terminal hardware, a new decentralized entrance will be created, and the corresponding AI use demand will be generated for ordinary consumers and office groups at the same time." On March 18, at the performance exchange meeting after the release of the annual report, Ma Huateng said.
In Ma Huateng's view, "lobster" applications do not require users to stay in front of the screen waiting for feedback results. At the same time, they have continuous memory capabilities and distinctive personalized features. They are more like a real assistant and have a vivid "feeling of a living person". The new AI implementation scenarios opened up by "Lobster" have also opened up more business fronts for Tencent to explore.
On March 18, Tencent Holdings (0700.HK) officially released its fourth quarter and full-year performance report for 2025. Data show that in the fourth quarter of 2025, Tencent achieved revenue of 194.37 billion yuan, a year-on-year increase of 13%; adjusted operating profit was 69.52 billion yuan, a year-on-year increase of 17%. From a full-year perspective, Tencent's total revenue in 2025 will reach 751.77 billion yuan, a year-on-year increase of 14%; adjusted net profit will be 259.626 billion yuan, a year-on-year increase of 17%.
This year, the core focus of the industry is almost entirely in the field of AI. Regarding the layout and planning of this track, Ma Huateng and Tencent executives gave detailed responses at the performance meeting.
What has the "lobster" craze brought to Tencent?
Ma Huateng said that in the past, most of the core entrances to AI applications were chatbots, and many underlying capabilities could not be fully released. However, after new scenarios emerge, Tencent can reintegrate the PC-side product matrix and Tencent Cloud's various capabilities to give full play to its own product matrix and ecological advantages. At the same time, "lobster" applications also provide new ideas for the WeChat AI functions that Tencent is planning and developing. In the future, every small program can be intelligent and "lobster-like" transformed, and can be called through API interfaces or command lines.
The previously launched Yuanbaopai product also allowed Tencent to see that there is still room for improvement in the context understanding efficiency of current agents in processing complex tasks, especially in multi-person scenarios. It needs to complement the file system, local database and memory parameter capabilities of "lobster" applications. Tencent's technical team will also fully absorb new technologies and achievements in the industry and continue to optimize products.
Talking about the relationship between the decentralized nature of "lobster" applications and the centralized nature of WeChat itself, Ma Huateng said frankly that WeChat itself is a centralized App, but its internal ecosystem is completely decentralized and currently accommodates hundreds of thousands of active mini program merchants. How to integrate the decentralized concept of "lobster" applications into the WeChat ecosystem in the future is a proposition that the team needs to focus on.
"All ecological service providers are worried about being 'short-circuited' and 'channelized' by AI agents – if the agents only call the underlying atomized services and no longer guide users to enter merchants' stores and mini-programs, service providers will inevitably be unable to accept such a model. They would rather retain their own traffic and entrances. Therefore, we must fully consider this core appeal when doing ecological design." Ma Huateng made it clear that the layout of centralization and decentralization needs to be promoted simultaneously in the future.
Tencent President Liu Chiping also disclosed at the meeting that Tencent has formulated a clear AI strategic plan and will implement a number of key actions in the next two to three quarters, and will also release quantifiable business progress in the future. At the large model level, Tencent has completed intensive team adjustments and upgrades in the past few months, reconstructing the organizational structure and work processes to make the organizational operation efficiency more suitable for the pace of AI R&D; at the same time, it has rebuilt the pre-training and reinforcement learning infrastructure to comprehensively improve data quality. Based on these layouts, Tencent is confident in creating large models with stronger intelligence in the future and further speeding up the overall iteration speed. In addition, Tencent will continue to deepen its efforts in the fields of video models and world models.
Liu Chiping said that the promotion effect of Yuanbao during the Spring Festival far exceeded expectations, and it achieved considerable new installations. Next, the team will improve user retention and activity through continuous product iteration, focusing on optimizing core capabilities such as voice dialogue. The current data has clearly shown that the upgrade of Yuanbao's capabilities will directly drive positive growth in user activity and retention rates. In addition to the existing functions of Yuanbao Pai, more differentiated features will be launched for Yuanbao in the future. In fact, WeChat has already implemented a number of AI-related layouts: search scenarios have been intelligently upgraded, AI technology has been deeply used in content recommendation and product recommendation, mini program developers can use AI tools to complete programming work, Yuanbao Dialogue Assistant has also accumulated a stable user group in WeChat, and many AI functions have completed preliminary testing in the WeChat ecosystem.
Liu Chiping made it clear that one of Tencent’s next core plans is to create an exclusive AI agent within WeChat. This agent will rely on the WeChat ecosystem to form a deep connection with users, opening up full-link capabilities such as mini programs, social networking, and payment. It can not only achieve natural dialogue with users, but also complete various complex scenario-based tasks.
AI investment will double in 2026
Liu Chiping revealed that Tencent will continue to increase investment in the AI field: Tencent's investment in the research and development of new AI products will reach 18 billion yuan in 2025, and this number will at least double in 2026.
The steady growth of Tencent's core business provides sufficient financial support for increased investment in AI. Liu Chiping said that due to the procurement gap of computing hardware such as GPUs in 2025, Tencent will make greater efforts in stock repurchases; in 2026, if the procurement of computing hardware can be successfully completed and capital expenditures meet the company's expectations, the scale of repurchases may be moderately reduced. Overall, the repurchase budget will maintain flexible room for adjustment. As long as there are sufficient resources to support the long-term development of the AI business, Tencent will prioritize investment, which will also create more considerable long-term returns for shareholders in the future.
Talking about the supply of chips and related hardware, Tencent executives said that in addition to GPUs, core hardware such as CPUs and storage are currently facing tight supply. Suppliers usually give priority to ensuring the supply of large-scale procurement partners. Tencent Cloud has previously signed multiple long-term procurement contracts. Leading cloud vendors have a greater probability of obtaining scarce hardware resources and further subletting them to customers. Because of this, domestic cloud service prices have continued to decline for a long time. It was not until the industry-wide supply shortage occurred that cloud service prices began to rise. If the supply chain tension continues, cloud service prices are likely to undergo further adjustments in the future.
It is worth noting that Tencent also revealed at the media communication meeting after the release of the financial report that the HY3.0 version of its Hunyuan large model is currently in the internal business testing stage and is planned to be officially released in April this year.
It is understood that HY3.0 is a major upgrade of Tencent's Hunyuan large model. Compared with the HY2.0 version, the overall effect has greatly improved, and the reasoning ability and agent capabilities have been significantly improved. Since the second half of 2025, Tencent Hunyuan team has completed a comprehensive reorganization of the team and R&D process, focusing on optimizing data quality, while rebuilding the core infrastructure of pre-training and reinforcement learning. These layouts will greatly improve the efficiency of model iteration and help create a base model with a higher level of intelligence.
In the future, AI will not be “one size fits all”
Talking about the fierce competition faced by Hunyuan, Tencent management said frankly that the future AI world will not be a one-size-fits-all monopoly, but a diversified world with multiple models and professional division of labor. Just as the evolution from PC Internet to mobile Internet has given rise to countless applications, the AI era will also move towards distributed and decentralized technology. Different models will have their own strengths in different areas such as code, long text, multi-modality, etc.
Under this pattern, Tencent's core advantage lies in its ability to reach users across PC, mobile, and cloud, as well as its in-depth coverage in diverse scenarios such as social networking, games, and content. The hybrid model will be closely integrated with decentralized ecosystems such as mini programs, allowing agents to have the ability to autonomously perform complex tasks, thereby providing unique value to users.
Tencent management said that with its strong ecosystem and application scenarios, Hunyuan can become one of the excellent models that users and enterprises can choose based on cost-effectiveness and performance needs.
On the day of the results release, topics related to Tencent employee salary also appeared on the hot search list.
Financial report data shows that as of December 31, 2025, Tencent's total number of global employees reached 115,849, compared with 110,558 employees at the end of 2024, with 5,291 new employees throughout the year. At the same time, the financial report disclosed that Tencent's total salary cost in 2025 will be 130.7 billion yuan, a year-on-year increase of 15.87%; based on this calculation, the per capita annual salary cost of Tencent employees in 2025 will be approximately 1.128 million yuan.
