Tencent and JD.com's robot strategy: Don't build hardware, build platforms.

Text by Qiu Xiaofen and Wang Fangyu

Edited by Su Jianxun

At the just - concluded WAIC (World Artificial Intelligence Conference), embodied intelligence was extremely popular.

Among exhibitors, there were only "Eighteen Arhats" (18 humanoid robots) at WAIC last year, while this year more than 90 robots of various forms participated in the exhibition.

In terms of the guest lineup, Sergey, the co - founder of the American star embodied company Physical Intelligence (PI), appeared at the forum of Zilliz Robotics. The "Four from Berkeley" in the field of embodied intelligence—Wu Yi, Gao Yang, Xu Huazhe, and Chen Jianyu—even made a rare joint appearance.

Domestic Internet giants are also eager to enter the field of embodied intelligence.

Tencent rarely invited Zhang Zhengyou, the head of the Robotics X Laboratory, to step forward this time.

Zhang Zhengyou is Tencent's first outstanding scientist at Level 17 (the highest professional rank in Tencent's history). He is also the person who built the Robotics X Laboratory from scratch. He proposed the "Zhang Zhengyou calibration method" in machine vision.

To support its own embodied business, Zhang Zhengyou, who rarely attends media events, announced at the Tencent WAIC forum: "Tencent wants to become a partner of robot manufacturers."

△Zhang Zhengyou, the director of Tencent's Robotics X Laboratory. Source: Authorized by the enterprise

JD, which was little - known in the wave of large AI models, seems eager to prove something. During WAIC, JD not only launched the group - level AI business brand "JoyAI". In the three months before and after WAIC, JD announced five consecutive times that it led the investment in embodied intelligence companies.

According to Intelligent Emergence, to expand the influence of its investment actions, JD also requires the invested enterprises to put the label of "Invested by JD" in their publicity.

"JD requires us to write it in the title of the financing manuscript." A person from an enterprise recently invested by JD told Intelligent Emergence.

At this WAIC, both Tencent and JD announced their strategies for entering the field of embodied intelligence—The commonality of the two giants is that they don't plan to directly manufacture robots in the short term. Instead, they will act as software platforms, integrating their past technological reserves and applying them to robots. And infrastructure such as models and computing power are the fields that everyone is targeting.

This strategy is obviously wise.

Especially when large models drive the explosive growth of cloud - based AI computing power, customers of large models have become a new cake that cloud computing providers and Internet giants are competing for. Embodied intelligence, which is still in its early stage of development, will undoubtedly be a new growth point and development direction for the giants.

On the eve of the explosion of the robot track, being the "water seller" in the industrial chain obviously has a faster commercialization speed than directly manufacturing robots. No matter how the robot industry evolves in the future, the demand of robot manufacturers for models and computing power will always be strong.

The large - scale entry of Tencent and JD may mark that the battle for customers in the field of embodied intelligence has begun.

JD: Not Just a Channel, Wants to Penetrate into Robot R & D

In the field of embodied intelligence, JD doesn't want to just be an e - commerce channel. It wants to go further and penetrate into the in - depth R & D of robots.

The entry point JD has chosen is to help robot manufacturers improve their human - robot interaction capabilities and provide corresponding model services.

He Xiaodong, the deputy dean of JD Group's Exploration Research Institute, told the media including Intelligent Emergence that they have observed that although many current robot manufacturers can do well in control and movement, they rarely develop large models, voice technology, and dialogue agents specifically. For consumers, they are no longer satisfied with just "movable robots". Instead, they need "robots that understand me".

Based on embodied intelligence, JD further proposed the concept of "possessed intelligence" and launched the JoyInside platform. As the name implies, it enables robots to have human - robot interaction capabilities driven by large models. The data used behind this platform includes not only general data but also the data of customer service, digital humans, and shopping guides accumulated by JD over the years.

It is reported that the architecture of JD's JoyInside platform is as follows: at the bottom layer, it builds intelligent computing infrastructure through JD Cloud, integrating multi - modal capabilities such as RAG, TTS, ASR, and LLM, allowing enterprises to use it out - of - the - box.

At the support system layer, the JoyInside platform supports functional modules such as active dialogue, emotion detection, and long - term memory.

Currently, JD has quickly implemented this set of capabilities in the products of more than a dozen robot manufacturers.

For example, Fuzozo, an AI trendy toy, achieved emotion perception and anthropomorphic feedback by accessing JD's JoyAI large model, with long - term memory and highly consistent words and deeds. This also made it one of the popular products in the AI toy category recently, known as the "AI version of Labubu".

△The robot dog of a partner at JD's booth. Source: Authorized by the enterprise

At WAIC 2025, JD invited a group of manufacturers to support its robot "friend circle", including Fuzozo, the most popular companion robot recently, Yuanluobo, an educational robot, and embodied intelligence manufacturers such as Zhongqing, Yun Shenchu, and Magic Atom.

Of course, in addition to providing model support, JD also provides channel support for a group of robot manufacturers. Sun Zhaozhi, the founder of Luobo Intelligence, told the media including Intelligent Emergence that Fuzozo was launched on JD for the first time. "It received a lot of support from JD, and now the product is out of stock."

As JD's previous data showed, during the "618" period, the self - operated sales of JD's intelligent robots increased by 3 times year - on - year, and the sales of embodied intelligent robots increased by 17 times.

Tencent: Offering a "Tairos" to Robot Manufacturers

Tencent doesn't completely avoid hardware.

In the seven years since the establishment of the Robotics X Laboratory, they have developed a self - balancing bicycle (in 2018), a quadruped robot dog with integrated legs and wheels (in 2021), a dexterous hand TRX - Hand (in 2023), and a pension robot named Xiaowu (in 2024).

However, these products are mainly for internal research and are not intended to be launched on the market. Compared with hardware, Tencent values the robot software platform more.

At the WAIC forum, Tencent launched the Tairos Embodied Intelligence Open Platform, whose Chinese name is "Tairos"—as the name suggests, it mainly provides software development capabilities for robot body manufacturers and application developers.

According to Zhang Zhengyou, the "Tairos" platform is mainly divided into two parts: model algorithms and cloud services. Among them, Tencent calls the model part the "SLAP3" system, which includes three types of large models—

Firstly, the planning large model (right - brain), which mainly helps robots understand complex goals and break them down into executable steps;

Secondly, the perception large model (left - brain), which is used to help robots understand the situation of the outside world and the state information of their own bodies;

Thirdly, the perception - action joint large model (cerebellum), which helps robots transform the collected information into executable action instructions from "seeing" to "doing".

In the cloud service part, Tencent hopes that through this development platform, robot manufacturers can complete the entire process of simulation, training, and data management and flexibly connect to hardware through SDK or API.

△Tencent's Tairos platform. Source: Authorized by the enterprise

Regarding the entry from the aspects of models and cloud services, Zhang Zhengyou said that it is mainly because Tencent has observed the pain points in the implementation of embodied intelligence. "There are 'pits' hidden in every link from the basic model to the deployment of real - world machines."

For example, in terms of basic models, the model capabilities of start - up companies are generally weak, and model training requires a huge amount of resource investment.

In the aspect of data collection for robots, the data efficiency of supervised fine - tuning is low, the cost is high, and it is difficult to scale up. A more neutral and complete platform is needed.

Regarding Tencent's cooperation with robot companies through strategic investment, Zhang Zhengyou told Intelligent Emergence that he is not responsible for investment, but it is most important to develop the entire ecosystem through investment.

Although playing the role of a software platform, Tencent doesn't plan to take on everything but provides modular services.

Zhang Zhengyou said that each manufacturer can select the desired modules according to its own technical strengths and weaknesses. "For example, if a manufacturer has strong perception (left - brain) technology, it can only use Tencent's planning (right - brain) or behavior (cerebellum) models."

However, Intelligent Emergence learned that the unified end - to - end VLA large model (Vision - Language - Action Model) has become the mainstream technical route in the field of embodied models, and some model manufacturers may have difficulty accepting the modular approach.

Zhang Zhengyou even said directly to Intelligent Emergence, "If a robot enterprise can independently complete the R & D of an end - to - end model, it is not a potential partner of Tencent."

To maintain the neutrality of the platform, Zhang Zhengyou repeatedly emphasized that the goal of the Tairos platform is not to make money. "We hope to develop the entire robot industry as quickly as possible."

end

end