ByteDance makes its first investment in four years, finally entering the robotics arena

ByteDance, which has kept a low profile in the field of embodied intelligence for a long time, has finally made its first direct investment in this track.

On January 12th, Independent Variable Robotics completed a Series A++ financing round of 1 billion yuan. It is an embodied intelligence enterprise founded in 2023 and had previously attracted investments from Meituan and Alibaba. In this financing round, ByteDance unexpectedly appeared.

Since 2025, the popularity of the embodied intelligence track has been on the rise. Internet giants have been taking action one after another. Tencent, JD.com, and Meituan have been making high-profile layouts, while ByteDance has remained extremely low-key.

Data from Qichacha shows that the peak period of ByteDance's previous wave of robot investments was from 2019 to 2021. In this wave of the embodied intelligence boom, ByteDance has been relatively restrained. Especially among the embodied intelligence companies founded after 2023, Independent Variable Robotics is ByteDance's only investment so far.

This investment may be related to the continuous attention of the founder, Zhang Yiming.

According to a report by LatePost, in 2025, Zhang Yiming visited some early-stage embodied intelligence startups and companies in the industrial chain. Some investors even commented that among entrepreneurs at his level, many would no longer meet such early-stage entrepreneurs. However, Zhang Yiming still maintains a strong sense of curiosity and a willingness to learn new things.

When Internet giants invest in robots, it is often closely related to their own business strategies, which is called "occupying ecological niches". Independent Variable Robotics can handle the "last 100 meters of delivery". From this perspective, it is not difficult to understand why Meituan and Alibaba invested in it, as they both have delivery needs. But what is the strategic intention behind ByteDance's investment?

01 The WALL - A Model: Independent Variable Robotics Completes the Last 100 Meters of Food Delivery

Independent Variable Robotics was founded in December 2023. Its founder, Wang Qian, graduated from Tsinghua University. During his Ph.D., he participated in multiple robot learning research projects at top robot laboratories in the United States.

However, he did not continue in the robotics industry but founded a quantitative fund company. This is something that Wang Qian regrets, as he wishes he had continued his robotics business.

Later, Wang Qian came into contact with OpenAI's GPT - 3 model at work and decided to return to China to start a business, founding Independent Variable Robotics.

It must be said that quantitative funds and AI are a natural fit. Liang Wenfeng, who is engaged in quantitative funds, launched DeepSeek. Similarly, Wang Qian, who is also in the quantitative field, founded Independent Variable Robotics, and both have ventured into AI - related industries.

From the very beginning, Independent Variable Robotics has focused on the R & D of the "general embodied large model". Wang Qian chose the technical route of the "end - to - end unified VLA large model". At that time, it was the only company in China that chose the end - to - end unified large - model technology from the start.

The end - to - end VLA (Vision - Language - Action) large model is the mainstream path in the current industry. Google's Gemini robotics, the US embodied intelligence enterprise PI, and Skild AI have all chosen this technical route.

Based on this, the self - developed WALL - A model of Independent Variable Robotics pioneered a system paradigm that deeply integrates VLA and the world model, enabling dual - arm robots to perform multiple complex and delicate operations such as clothing handling, storage organization, and wire harness arrangement.

The value of this model technology lies in the fact that the robot has demonstrated zero - sample generalization ability in some new, unseen task scenarios, which is one of the key indicators for realizing general robots.

In March this year, when being interviewed by Intelligent Emergence, Wang Qian stated that the model level of Independent Variable Robotics is basically on par with that of PI and Google. In contrast, domestic manufacturers have generally just started in this direction and lag far behind in terms of progress.

Some investors also said that Unitree and Zhipu are in the first echelon, with financing amounts exceeding 1.5 billion yuan each. Independent Variable Robotics, with a financing amount exceeding 1 billion yuan, has entered the quasi - first - tier from the second - tier.

The technological leadership has made Independent Variable Robotics favored by the capital market. Information from Qichacha shows that Independent Variable Robotics has received a total of 9 rounds of financing and has reached the Series A++ stage. Among them, there was a Series A financing of hundreds of millions of yuan exclusively invested by Meituan in May 2025, and Alibaba Cloud participated in the financing in September.

Source: Qichacha

With the maturity of technology, Independent Variable Robotics has also completed the product closed - loop from software to hardware. It has not only achieved full R & D of core components such as robotic arms, joint modules, and power drives and algorithm adaptation but also successively released two robot bodies, "Quantum One" and "Quantum Two".

In August 2025, the fully self - developed wheeled dual - arm humanoid robot "Quantum 2 (Quanta X2)" released by Independent Variable Robotics can meet the needs of complex scenarios such as intelligent warehousing (express sorting), services, and households.

Source: Independent Variable Robotics WeChat official account

What truly demonstrates its technical strength is a real - shot video recently made public by Independent Variable Robotics: The "Quantum One" robot equipped with the WALL - A model completed the task of the "last 100 meters" of food delivery independently in a real open environment without any human intervention.

Independent Variable Robotics claims that this is the world's first mobile operation example of fully autonomous food delivery in a real open scenario based on the VLA end - to - end model.

Source: Independent Variable Robotics WeChat official account

The video includes three major parts: "picking up the package, unpacking, and recycling", "long - distance autonomous delivery", and "precise delivery by taking the elevator". There are many technical difficulties involved. For example, when the robot takes out the food from the takeaway carton with Velcro, it is a test of the robot's dual - arm coordination and force control ability; when the robot folds the takeaway carton and stuffs it into the narrow recycling opening, it demonstrates the end - to - end "force - position" control skills.

This demonstration also allows the outside world to see the potential of Independent Variable Robotics' robots to "think with the brain and work with the hands" in a complex real environment. It is also a breakthrough in the integration of hardware, technology, and scenarios.

02 The Dual - Line Strategy Emerges: ByteDance's Ambitions in Robotics

From the general logic of large - scale Internet companies' investments in embodied intelligence, it is often closely related to their own business scenarios.

For example, Meituan's investment in embodied intelligence tends to expand its unmanned service capabilities. By cooperating with Galaxy General, the two parties have created a 24 - hour "unmanned" pharmacy operated by humanoid robots.

So, what is the logic behind ByteDance's investment in Independent Variable Robotics?

This has to be viewed from ByteDance's robot strategy.

ByteDance's layout in robotics largely stems from Zhang Yiming. As early as 2020, Zhang Yiming showed interest in robots and would participate in robot project discussions irregularly. However, at that time, ByteDance's investment was still relatively restrained.

The turning point was in 2023. According to a report by LatePost, in mid - 2023, Zhang Yiming, ByteDance CEO Liang Rubo, and other ByteDance executives had a meeting with the AI Lab robot team to discuss the direction of ByteDance's robots. Eventually, the goals were divided into two parts: One is to produce robots to serve ByteDance's e - commerce fulfillment needs; the other is to focus on technology and apply the capabilities of large AI models to robots.

What is meant by serving e - commerce fulfillment is that the robots can pick, transport, and pack goods in the warehouse. Douyin has its own e - commerce warehouses. Using robots can relieve the pressure of picking and transporting caused by the sudden increase in orders during major promotion periods.

At that time, Zhang Yiming also encouraged the team to set bigger goals. In that year, ByteDance completed the mass production of 200 units. Later, LatePost reported in July 2025 that the total mass production of robots developed by ByteDance exceeded 1,000 units, and they are wheeled logistics robots.

When Zimu AI verified the above information with ByteDance, the other party said they would not respond.

In addition, in terms of mass production, in 2025, ByteDance also cooperated with Seres to jointly promote the mass production and application of humanoid robots.

Besides hardware mass production, ByteDance is also promoting the progress of embodied intelligence technology, especially accelerating significantly since 2025.

In July 2025, ByteDance's Seed team launched the VLA (Vision - Language - Action model) model - Seed GR - 3.

The Seed team is the core R & D team of ByteDance in the field of artificial intelligence, founded in 2023, with the field of embodied intelligence among its research directions. Since 2023, the AI Lab team has been gradually incorporated into the Seed team.

According to the introduction of the Seed team, the GR - 3 model can understand abstract instructions, manipulate flexible objects, and has good generalization ability, enabling it to effectively handle long - range tasks and perform highly dexterous operations.

For this purpose, ByteDance has also simultaneously developed the general dual - arm mobile robot ByteMini, equipped with the GR - 3 model, forming a complete solution.

Source: ByteDance official website

In December, the Seed team further released the GR - RL framework, enabling robots to stably complete multi - step, high - precision operation tasks in real scenarios.

On one hand, there is technological innovation; on the other hand, there is talent reserve.

In June 2025, the Seed team was reported to be recruiting several leaders for robot products, engineering, and embodied intelligence large models. A few months later, Volcengine, a subsidiary of ByteDance, also launched a recruitment drive, such as for a "senior expert in operation algorithms (embodied intelligence)" to lead the R & D of operation algorithms for humanoid embodied robots and VLA models.

It can be seen that in the embodied intelligence track, ByteDance is also taking a two - pronged approach, adopting a strategy of "combining software and hardware, and combining self - research and investment".

Although ByteDance has been relatively conservative in its investments in embodied intelligence, Jinqiu Fund, which has a deep connection with ByteDance, has made multiple investments. According to a report by National Business Daily, ByteDance is one of the many limited partners of Jinqiu Fund. In 2025, Jinqiu Fund invested in several embodied intelligence enterprises such as Stardust Intelligence, Inkers, and Unitree Technology.

It is not difficult to understand why ByteDance invested in Independent Variable Robotics, as the two sides have strategic synergies. In terms of software, Independent Variable Robotics' end - to - end VLA model, WALL - A, has multi - modal thinking chains and zero - sample generalization ability; at the hardware level, the intelligent warehousing capabilities demonstrated by the Quantum One are also complementary to ByteDance's current business.

As Jiang Yi, the founding partner of Hengye Capital, said, the value of Independent Variable Robotics far exceeds the surface of "last - mile delivery". It is building the fundamental model of the physical world, which is exactly the strategic asset that ByteDance values. The end - to - end VLA technical route of Independent Variable Robotics complements ByteDance's large - model genes. The multi - modal thinking chain ability of the WALL - A model may provide underlying support for ByteDance's AR/VR, smart hardware, and other businesses.

In addition, Jiang Yi also said that although ByteDance does not directly engage in delivery, its e - commerce and local life services are accelerating their layouts. The flexible operation ability and generalization deployment experience of Independent Variable Robotics may become the key technological springboard for ByteDance to break through in offline scenario intelligence.

References:

Wang Qian of Independent Variable Robotics: The Large Model of Embodied Intelligence Can't Copy Foreign Examples. Intelligent Emergence

This Company Has Received 8 Rounds of Financing in Two Years, with Heavy Bets from Meituan and Alibaba Cloud. China Entrepreneur Magazine

ByteDance Plans to Build Robots, and the Team Aims to Expand to 100 People. LatePost

ByteDance Has Built Over a Thousand Robots in Two and a Half Years, with the Long - Term Goal of Embodied Intelligence. LatePost

Independent Variable Robotics Receives 1 Billion Yuan in Investment from ByteDance, Sequoia, etc., Building the Fundamental Model of the Physical World. Independent Variable Robotics

This article is from the WeChat official account "Zimu AI", author: Xue Yaping, published by 36Kr with authorization.