At the recently concluded WAIC (World Artificial Intelligence Conference), embodied intelligence was undoubtedly the hottest topic. At the exhibitor level, while last year's WAIC featured only the "Eighteen Arhats" (18 humanoid robots), this year saw over 90 robots of various forms participating in the exhibition. In terms of guest lineup, Sergey, co-founder of American star embodied intelligence company Physical Intelligence (PI), appeared at the forum of Zhiyuan Robotics, while the "Berkeley Four" in the embodied intelligence field – Wu Yi, Gao Yang, Xu Huazhe, and Chen Jianyu – made a rare joint appearance.

In entering the embodied intelligence space, domestic internet giants are not to be outdone. Tencent made the rare move of bringing out Zhang Zhengyou, head of the Robotics X Laboratory. Zhang Zhengyou is Tencent's first Level 17 (the highest professional rank in Tencent's history) Distinguished Scientist and the architect who built Robotics X from scratch. He previously proposed the "Zhang Zhengyou calibration method" in machine vision.

To champion the company's embodied intelligence business, Zhang Zhengyou, who rarely attends media events, announced at Tencent's WAIC forum: "Tencent aims to become a partner to robot manufacturers."

JD.com, which remained relatively quiet during the AI large model wave, seems eager to prove something. During WAIC, JD.com not only released its group-level AI business brand "JoyAI" but also made five consecutive official announcements about leading investments in embodied intelligence companies over the three months surrounding WAIC.

It's understood that to amplify the impact of its investment activities, JD.com also requires its investee companies to include the "JD.com Investment" label in their promotional materials. "JD.com requires us to definitely include it in the title of our financing announcements," a person from a company recently invested by JD.com told the media.

At this WAIC, both Tencent and JD.com revealed their strategies for entering embodied intelligence – both giants share the commonality that they do not intend to directly manufacture robots in the short term, but rather serve as software platforms, integrating their past technological reserves and transplanting them into robots. Models and computing power are the foundational infrastructure areas that everyone is targeting.

This strategy is obviously wise, especially as large models drive explosive growth in cloud AI computing power, making large model customers the new prize that cloud computing providers and internet giants are competing for. Embodied intelligence, still in its early development stage, will undoubtedly be a new growth driver and direction for these giants.

On the eve of the robot industry's explosion, being the "water seller" in the industrial chain is obviously faster to commercialize than directly entering robot manufacturing. Regardless of how the robot endgame ultimately converges, robot manufacturers will always have strong demand for models and computing power. The major investments by Tencent and JD.com may signal that the battle for embodied intelligence customers has begun.

**JD.com: Not Just a Channel, Wants to Penetrate Robot R&D**

In the embodied intelligence field, JD.com doesn't want to be just an e-commerce channel but wants to go further, penetrating into the deep-level R&D of robots. JD.com's chosen entry point is helping robot manufacturers improve human-robot interaction capabilities and providing corresponding model services.

He Xiaodong, Deputy Director of JD.com Group's Exploration Research Institute, told media that they observed that while many robot manufacturers can handle control and movement well, they rarely develop large models, voice technology, and conversational AI agents specifically. For consumers, people are no longer satisfied with just "robots that can move" but need "robots that understand me."

To this end, JD.com further proposed the concept of "possessed intelligence" based on embodied intelligence, releasing the JoyInside platform. As the name suggests, it enables robots to have human-robot interaction capabilities driven by large models.

The data behind this platform includes not only general data but also JD.com's years of accumulated customer service, digital human, and shopping guide data. According to the introduction, the architecture of JD.com's JoyInside platform involves building intelligent computing infrastructure through JD Cloud at the bottom layer, integrating multimodal capabilities such as RAG, TTS, ASR, and LLM for enterprises to use out-of-the-box.

At the support system layer, the JoyInside platform supports functional modules such as proactive dialogue, emotion detection, and long-term memory. Currently, JD.com has quickly deployed these capabilities to the products of more than a dozen robot manufacturers.

For example, AI toy Fuzozo achieved emotion perception and anthropomorphic feedback by integrating JD.com's JoyAI large model, possessing long-term memory and highly consistent words and actions. This made it one of the explosive products in the recent AI toy category, being called the "AI version of Labubu."

At WAIC 2025, JD.com specifically brought numerous manufacturers to showcase their robot "circle of friends" – including the recently popular companion robot Fuzozo, educational robot Yuanluobo, and embodied intelligence manufacturers such as Zhongqing, Cloudwalk, and Magic Atom.

Of course, besides providing model support, JD.com also provides channel support to numerous robot manufacturers. Sun Zhaozhi, founder of Luoboshi Intelligence, told media that Fuzozo launched exclusively on JD.com and "received tremendous support from JD.com, and the product is now sold out."

As JD.com's previous data showed, during "618," JD.com's self-operated intelligent robot sales increased 3 times year-over-year, while embodied intelligence robot sales increased 17 times.

**Tencent: Offering Robot Manufacturers a "Titanium Screw"**

Tencent doesn't completely avoid hardware. In the seven years since the establishment of Robotics X Laboratory, they have successively created balance bicycles (2018), leg-wheel integrated quadruped robot dogs (2021), dexterous hand TRX-Hand (2023), and elderly care robot Xiao Wu (2024).

However, these products are mainly used for internal research and are not intended for market release. Compared to hardware, Tencent values robot software platforms more. At the WAIC forum, Tencent released the Tairos embodied intelligence open platform, with the Chinese name "Titanium Screw" – true to its name, mainly providing software development capabilities to robot body manufacturers and application developers.

According to Zhang Zhengyou, the "Titanium Screw" platform mainly consists of two parts: model algorithms and cloud services. Among these, Tencent calls the model part the "SLAP3" system, encompassing three types of large models – first, planning large models (right brain), mainly helping robots understand what complex goals are and break them down into executable steps; second, perception large models (left brain), used to help robots understand the external world and the robot's own state information; third, joint perception-action large models (cerebellum), used to help robots go from "seeing" to "doing," converting collected information into actionable instructions.

For the cloud services part, Tencent hopes that through this development platform, robot manufacturers can complete the entire process of simulation, training, data management, and flexibly interface with hardware through SDKs or APIs.

The reason for entering through models and cloud services, Zhang Zhengyou explained, is mainly because Tencent observed the pain points in embodied intelligence implementation: "From basic models to real machine deployment, every link hides 'pitfalls.'"

For example, in terms of basic models, startup companies generally have weak model capabilities, while model training requires massive resource investment. In robot data collection, supervised fine-tuning has low data efficiency, high costs, and is difficult to scale, requiring a more neutral and complete platform.

Regarding Tencent cooperating with robot companies through strategic investments, Zhang Zhengyou told media that while he doesn't handle investments, developing the entire ecosystem through investment is most important.

Although playing the role of a software platform, Tencent doesn't intend to take on everything but provides modular services. Zhang Zhengyou stated that each manufacturer can select the modules they want based on their own technological strengths and weaknesses. "For example, if a manufacturer has strong perception (left brain) technology, they can only use Tencent's planning (right brain) or behavioral (cerebellum) models."

However, it's understood that unified end-to-end VLA large models (Vision-Language-Action Models) have become the mainstream technical route in the embodied model field, and some model manufacturers may find it difficult to accept modular approaches.

Zhang Zhengyou even directly stated that "if robot companies can independently complete end-to-end model R&D, they are not Tencent's potential partners."

To maintain platform neutrality, Zhang Zhengyou repeatedly emphasized that the Tairos platform's goal is not to make money: "We hope to develop the entire robot industry at the fastest speed possible."