Summary of Key Points
At the 2026 WAIC (World Artificial Intelligence Conference), Tencent introduced an "embodied intelligence full-stack solution" in an attempt to continue the "platform logic" of the mobile internet era—avoiding hardware production and focusing solely on serving as the "brain" for robotics manufacturers and a platform provider, generating revenue through its ecosystem. However, this strategy faces three significant challenges: leading robotics companies are developing their own "brains" (unwilling to outsource this task); embodied intelligence data is a scarce and core asset that manufacturers are reluctant to share; and the "brain" is closely integrated with hardware, making it difficult to create a universal platform. Tencent's real opportunity may lie in shifting to providing infrastructure services such as computing power and simulation, rather than aiming to become a "platform dominator."
Detailed Analysis
1. Tencent's Traditional Approach: Trying to Replicate the Mobile Internet Platform Myth?
Tencent's full-stack solution (model + platform + cloud infrastructure + applications) appears comprehensive, but its fundamental approach remains the same: avoiding heavy, low-profit hardware and focusing on the "brain" and platform layer. Just as in the mobile internet era, WeChat made money by connecting users and providing services, and cloud services generated revenue through its ecosystem—initial investments were high, but the cost per user decreased as more people used them. However, embodied intelligence differs from social or e-commerce applications, and this "light-asset platform" logic encounters barriers in the physical world.
2. Lack of Industry Cooperation: Leading Robotics Companies Are Developing Their Own Brains
The industry trend for 2026 is clear: top robotics companies are investing in developing their own large models (the "brains" of robots). For example, Zhiyuan Robot, co-founded by Tencent, not only developed its Qiyuan Large Model but also sold over 10,000 units; Zhifang Technology raised 5 billion yuan, with a focus on its own brain-like system; Yushu Technology also made its model open-source. The reason is simple: without developing its own large model, a company risks being left behind. The capital market agrees with this viewpoint, as companies with self-developed brains are highly sought after, while those without are falling behind. Leading manufacturers are unwilling to hand over their critical capabilities to Tencent.
3. Data Is the Lifeline: The Platform Model Does Not Work Here
Embodied intelligence data is fundamentally different from internet data. Internet data is often free (e.g., articles online), but robot data requires substantial effort to collect—e.g., videos of robots picking up objects or moving materials, which require actual operations or expensive simulations. Moreover, the amount of data collected is minimal; Tencent's model only gathered 10,000 hours of data in a year, compared to the 1.2 billion hours required for large language models. This means that data has become a scarce and competitive asset. Handing over data to Tencent for training would equate to sharing business secrets. The premise for manufacturers to use Tencent's platform for training is thus not viable.
4. The Brain and Body Are Intertwined: Creating a Universal Platform Is Impossible
The "brain" of robots is not a generic software; it must be tailored to the hardware (the robot's body). For instance, the walking posture of bipedal robots and the turning methods of wheeled robots differ significantly, requiring the brain to be optimized for each type of hardware. Creating a universal platform that works with all robots is technically nearly impossible due to the vast differences in kinematics and dynamics. Manufacturers also have no incentive to cooperate, as they can achieve better performance with their own customized solutions.
5. Tencent's Opportunity: Moving From Platform Leader to Infrastructure Provider
Despite the challenges with the platform strategy, Tencent has its strengths—computing power (tens of thousands of GPU cards for training), simulation systems that save manufacturers on real-device testing costs, and data infrastructure (tools for generating synthetic data). These are essential resources that individual robotics companies cannot afford. Tencent could become an infrastructure provider, similar to someone selling shovels near a gold mine, earning revenue from service fees rather than through platform fees. The EaaS (Everything as a Service) service launched at the 2026 WAIC is an attempt in this direction. However, this business requires continuous investment and does not involve passive profit generation, which is fundamentally different from Tencent's traditional platform model.
Conclusion
The success of the internet may not be directly transferable to the hard-tech era. Tencent's approach of connecting everything was effective in social media and e-commerce, but embodied intelligence operates in a physical industry where the brain and body are inseparable, and data is a critical asset for each company. While avoiding hardware production is a wise decision, if Tencent wants to continue relying on platform fees for revenue, it may need to adjust its strategy. In this field, providing essential infrastructure services (the "shovels") is more practical than being a platform dominator, and this business is never easy.