Riding the wave of the “lobster” craze, Tencent has rushed into the fray.
In the spring of 2026, OpenClaw sparked a nationwide “lobster farming” craze, placing Tencent at the forefront of AI application development for the first time. Within a month, over ten agent products, including WorkBuddy, QClaw, and CodeBuddy, were launched, fully integrating with WeChat, QQ, and WeChat Work.
Behind this spectacle, a deeper strategic outline is emerging: Tencent is not merely chasing the fleeting “lobster” trend, but rather accelerating the implementation of an AI catch-up path centered on a super-entry point and engineering capabilities.
01 Tencent, which was “a step behind” in its early stages, is now accelerating its efforts to “catch up.”
Tencent’s lag in developing its own large-scale models is an open secret.
In early 2025, the explosive popularity of DeepSeek directly impacted Tencent. Within a month of integrating DeepSeek, Yuanbao’s daily active users increased approximately 20-fold. In early 2026, Tencent also spent 1 billion yuan on red envelopes to drive traffic to the Yuanbao app, but the effect was relatively limited.
Some market observers believe that users stayed on the Yuanbao app primarily because of the DeepSeek open-source model and the cash red envelopes, which to some extent reflects that Hunyuan (the platform) has not yet fully met user expectations.
Tencent is not unaware of the severity of the problem. Li Qiang, in an interview with the Science and Technology Innovation Board Daily, also admitted that if the model is well-positioned and performs strongly, the ability to retain customers is relatively strong.
“First, we focus our energy on the development of the ‘engine.’ Of course, we also acknowledge that we still have a long way to go, and we are catching up, but the progress is relatively fast,” Li Qiang emphasized.
In Tencent’s 2025 full-year performance report released in March, the term “AI” was used throughout. Data shows that Tencent’s total capital expenditure reached 79.2 billion yuan in 2025, with R&D investment reaching 85.75 billion yuan, both record highs.
Tencent President Martin Lau also revealed that Tencent expects to increase capital expenditure in 2026 to acquire computing power to support model training and inference, while also having additional computing power available for external leasing and sales. “Last year, Tencent invested 18 billion yuan in new AI products, and this year it will at least double.”
Meanwhile, Tencent is intensively strengthening its talent and organizational structure.
At the end of 2025, Tencent successively brought in Yao Shunyu, a former researcher at OpenAI, as its chief AI scientist, and Pang Tianyu, a former senior research scientist at Singapore’s Sea AI Lab, as the chief scientist of Hunyuan. In March 2026, news broke that Tencent AI Lab was dissolved, with some personnel merging into the Hunyuan team. This also means that Tencent is concentrating all its AI R&D resources on the main line of large-scale models.
However, talent acquisition and organizational adjustments cannot immediately bridge the time gap. Hunyuan 3.0 will not be officially released until April, while ByteDance’s Doubao and Alibaba’s Tongyi have already completed multiple iterations and ecosystem penetration. As we can see, Tencent’s current strategy is to catch up with the underlying model while leveraging the momentum of the “lobster” phenomenon to seize entry points with an application-layer product matrix and amplify the effectiveness of existing models through engineering capabilities.
02 Riding the wave of the “lobster boom”: Seizing the entrance and setting up “scaffolding” projects
As an open-source framework, OpenClaw puts all vendors on a level playing field. Tencent’s advantage lies not in its model, but in the massive, nationwide traffic portal formed by WeChat, QQ, and WeChat Work, and the entire product ecosystem that extends from it.
In this industry-wide “shrimp-raising war,” Tencent’s strategy, seemingly following the trend, is actually aimed at using “lobster” (a metaphor for a competitive market) to drive synergy in its core businesses such as cloud services and SaaS, forming a closed loop of “entry point + infrastructure + application.”
During the earnings call, Ma Huateng (Pony Ma) publicly discussed the “lobster farming” concept for the first time, believing that “lobster” applications can allow AI to be implemented in a variety of rich scenarios, unlike the past focus solely on ChatBots. This would better leverage Tencent’s product matrix and ecosystem advantages, and also provide some inspiration for the WeChat Agent currently under development.
Li Qiang, Vice President of Tencent Group, also believes in an interview with the Science and Technology Innovation Board Daily that the emergence of OpenClaw is a very important watershed moment. In his view, AI has moved from dialogue to execution, making its value more intuitive, and the enrichment of application scenarios can also bring value to enterprises. “Therefore, this is one of the most important reasons why we are following up so quickly.”
According to statistics from the Science and Technology Innovation Board Daily, since March 2026, Tencent’s intensive launch of “Lobster” Agent applications has formed a full-chain matrix covering personal desktops, enterprise deployments, and ecosystem platforms, and has connected with super traffic portals such as WeChat and QQ. As of April 2026, Tencent has launched more than ten AI Agent products and platform services.
Among them, personal desktop applications include QClaw, WorkBuddy, and skill-based tools such as meeting/document tools; enterprise-level applications include Enjoy 2.0, Qidian Marketing Cloud MAGIC Agent 2.0, TC Data Agent, and ClawPro; R&D and ecosystem platforms include CodeBuddy, the ADP intelligent agent development platform, as well as the security tool “Lobster Butler,” the SkillHub skills community, and more.
It is worth mentioning that at the Tencent Cloud Summit at the end of March, Tencent Group Senior Executive Vice President Tang Daosheng gave a more systematic explanation of the development of AI Agents from a technical perspective, proposing the “Harness” theory.
He believes the capability gap between mainstream large-scale models is gradually narrowing. Enterprises’ core need is no longer simply possessing the best models, but rather how to maximize model capabilities without altering the model architecture and parameters through model harness—including tool usage, layered context engineering, long memory management, and workflow design—and other system engineering techniques.
This approach is already reflected in Tencent’s AI product strategy. Tencent Cloud’s Intelligent Agent Development Platform (ADP) connects agents to a “library” through capabilities like RAG and knowledge bases. Claw, acting as the neural hub, runs within the Agent Runtime security sandbox, discovering and invoking skills from the skill library to trigger actions. The Agent Runtime sandbox solution can also be used for verifying program results in large-scale model reinforcement learning, improving training efficiency.
It’s important to note that this engineering capability is not unique to Tencent; competitors like Alibaba and ByteDance also possess it, and the open-source community is rapidly catching up. Therefore, supplementing the underlying model capabilities is crucial for Tencent. When the model capabilities are comparable, the entry point and Harness act as a moat; once the competitor’s model is significantly stronger, this system will become a “more efficient funnel,” and users who come in and find that the experience is not as good as the competitors will turn around and leave.
03 After the “lobster” boom, model capabilities and ecological closed-loop systems are key to competition.
“The lobster craze will eventually pass,” Huang Guangmin, product director of Tencent Cloud AI Intelligent Agent, told the Science and Technology Innovation Board Daily.
In fact, this craze came and went quickly. WeChat Index shows that the popularity of the term “lobster” surged on March 6, peaked around March 11, and then quickly declined. Clément Delangue, CEO of the AI community Huggingface, also predicted as early as March 13 that the OpenClaw craze would fade in six weeks.
As mentioned above, Tencent intends to use this “lobster craze” to build a closed loop of “entry point + infrastructure + application”.
However, each link in this closed loop is surrounded by strong competitors. At the bottom layer, there is competition for computing power from Alibaba and ByteDance; at the application layer, there is competition for office scenarios from Lark and DingTalk; and at the model layer, there is an open-source encirclement from DeepSeek and Zhipu. Whether Tencent’s “full-stack” approach can translate into “complete victory” remains to be seen.
It is worth noting that along with the “lobster” wave, token consumption has become a focus of the industry. Many cloud service providers have launched price subsidies and aggressive performance indicators, plunging the industry into fierce competition. In contrast, Tencent has adopted a restrained approach.
Li Qiang told a reporter from the Science and Technology Innovation Board Daily, “Tokens are an important indicator for evaluating AI products, but they are by no means the only one.” He used an analogy: “Imagine tokens as fuel consumption in a car. If you only focus on fuel consumption and not on the engine’s economy and output capacity, customers will eventually abandon it. You still need to focus more on the product itself. If there’s a user-friendly AI product that customers can derive value from, then real token consumption will naturally occur.”
As the market frenzy subsides, and “shrimp farming” (a metaphor for artificial intelligence) becomes routine, the real competition is just beginning. Tencent’s trump cards are clear: the super-entry point of WeChat, the synergistic capabilities of its product matrix, and a pragmatic focus on “engineering” rather than “model parameters.”
However, the shortcomings of its underlying model remain prominent. Whether Tencent can truly close this generational gap remains to be seen until the official release of Hunyuan 3.0.
According to Tang Daosheng, Hunyuan 3.0 features significantly reduced activation parameters for a superior user experience, and marked improvements across multiple dimensions, including complex reasoning, long memory, long text processing, multi-round questioning, and agent capabilities. “Experimental tests in Yuanbao (the game’s AI platform) showed very clear positive benefits.”
This industry-wide “lobster craze” has become a crucial opportunity for Tencent to fill gaps in the AI race. As the hype subsides, AI competition will ultimately return to the core battle of model strength and ecosystem closed-loop systems. Whether Tencent’s full-stack strategy can truly gain a foothold remains to be seen and will be tested by the market and time.