ByteDance is hiring and making its foray into embodied intelligence.

Recently, ByteDance has intensively updated its public recruitment information in its Seed Model department. Dozens of professional positions have been released simultaneously, covering a wide range of fields from hardware design, dexterous operation, reducer control to physical simulation and data algorithm engineering.

This sends an extremely clear signal: ByteDance is no longer satisfied with participating in embodied intelligence through financial investment, but is trying to build a full-stack pipeline of "software, hardware + data" for humanoid robots on its own, promoting products to truly move from "production demonstration" to "mass production".

Meanwhile, the macro background is extremely grand: China manufactured more than 80% of the world's humanoid robots in 2025. Morgan Stanley predicts that China's humanoid robot shipments in 2026 and 2027 will be about 50,000 and 100,000 units respectively; data from the Ministry of Industry and Information Technology shows that the output in the first half of 2026 has exceeded 40,000 units, and the full-year output is expected to exceed 100,000 units. At this critical juncture, ByteDance's entry is logical yet full of challenges. When the Internet giant accustomed to "agile iteration" collides with the hardcore manufacturing industry characterized by "heavy assets and zero tolerance for errors", a cruel major test concerning technology, supply chain and business closed loop has just begun.

From Brain to Body

ByteDance's entry into the humanoid robot sector is by no means a random impulse. In June 2025, Jinqiu Fund, a ByteDance-affiliated fund, participated in the Series C financing of Unitree Robotics, a leading domestic robot company. However, the shift from "financial investment" to "personal involvement" means ByteDance has deeper anxiety and desire for the ultimate outcome of AI.

In the era of large models, ByteDance has proved itself at the "brain" (algorithm and model) level with products such as Doubao. But without a physical entity that can move in the physical world, the boundary of AI will always be locked behind the screen. Large models have rich world knowledge, but cannot directly understand friction, inertia and real causal relationships. Turning AI into an interactive physical entity is the only way forward.

Judging from the positions released by ByteDance, its ambition is remarkably huge. The positions covering hardware design, dexterous operation and reducer control indicate that ByteDance is trying to fully grasp the underlying physical structure and motion control of robots; physical simulation and data algorithms rely on ByteDance's extremely powerful computing power reserve and algorithm genes.

In ByteDance's business landscape, embodied intelligence is not an isolated product, but a natural extension of its entire AI ecosystem. It is trying to build a closed loop of "large model inference (Seed) - physical simulation training - hardware entity execution". In this closed loop, ByteDance's biggest advantage lies in software and data. Relying on its huge computing power cluster, ByteDance can build a "simulation world" in the virtual environment, allow robots to "try and make mistakes" in the simulation through reinforcement learning, and then transfer the capabilities to the real world. This is exactly the capability that traditional hardware companies do not possess.

Mass Production Gap

Although ByteDance's algorithm capability is beyond doubt, recent reports from institutions such as Morgan Stanley have poured a basin of cold water on the market. The report points out that the timetable for large-scale deployment, from the "demonstration video" in the laboratory to the actual replacement of manual labor in factories, is being pushed back. Enterprises' willingness to adopt such robots remains, but it will take longer to move from pilot projects to large-scale deployment. There are three core obstacles: data shortage, unreliable hardware, and frequent failures of robots beyond preset tests.

This is a gap that all large Internet companies that cross-border into hardware manufacturing must face.

The logic of the Internet is "small steps, rapid iteration, and tolerance for bugs". But the logic of the manufacturing industry and the physical world is "get it right at the first attempt, safety first, zero tolerance for errors". To make robots tighten screws and carry boxes on the real factory assembly line, you need to feed them massive amounts of real physical interaction data. In reality, the cost of collecting high-quality data is extremely high.

Robots trained in the simulation environment often "crash" immediately once they encounter changes in friction, changes in light, or an irregular object. This is the most famous "Sim-to-Real" (simulation to reality) gap in the industry.

In addition, human fingers have extremely high degrees of freedom and tactile feedback, and the existing dexterous robot hands still face extremely high engineering challenges in terms of service life and force control. The reliability of core components such as motors, reducers and sensors directly determines the service life and cost of robots.

What ByteDance needs to recruit are no longer just geeks who sit in cubicles and write code, but doers who understand mechanical engineering, materials science and supply chain management. Combining the top AI algorithms with the heaviest gears and hydraulic devices, there is no shortcut to "miracles achieved by brute force", and you can only make steady progress step by step through engineering trial and error.

Supply Chain Foundation

In this competition, China's manufacturing industry has demonstrated an extremely strong underlying strength.

According to relevant data, China manufactured more than 80% of the world's humanoid robots in 2025. This means that China is the most fertile soil for global embodied intelligence in terms of the completeness of the supply chain, the room for compression of manufacturing costs, and the trial and error field for product iteration. In the Yangtze River Delta and the Pearl River Delta, you can almost find all the core components needed to manufacture a humanoid robot, with extremely fast response speed.

At the same time, local Chinese unicorns such as Unitree have also released their latest generation of advanced models and products recently. From quadruped robot dogs to humanoid robots, Chinese enterprises have taken the lead in the world in motion control (limbs).

ByteDance's joining at this time will undoubtedly intensify the competition in this track. The entry of large manufacturers can not only bring abundant capital, but also inject the most cutting-edge intelligence of large models into the industry, accelerating the maturity of the entire industry. However, the challenges are equally huge: how to stand out among many local robot enterprises that have been deeply engaged in this field for many years, and how to use their own software advantages to form differentiation, are the urgent problems to be solved for ByteDance.

The Dividend of Shovel Sellers

In the vigorous "human-making" movement of giants, who is most likely to achieve a business closed loop in the short term?

Morgan Stanley gave an extremely cold answer: when tech giants spend billions of dollars chasing sci-fi humanoid robots, the only sure winners at present are traditional suppliers that sell them gears, bearings and hydraulic devices.

This perfectly replicates the "shovel seller" law in the AI large model era. While companies such as ByteDance, Tesla and Figure are fighting fiercely in software and ontology, those core component manufacturers that silently manufacture reducers, servo motors and sensors are steadily reaping the overflowing dividends from the giants. No matter which large manufacturer takes the lead in breaking through mass production, it will inevitably need to purchase core components from these underlying supply chain enterprises. In the early stage of the hardware revolution, selling shovels is always the most certain business.

The Dilemma of Closed Loop

Apart from the technical challenges, the core problem faced by humanoid robots is still the business closed loop.

Large manufacturers invest in R&D to make profits in the end. The cost of humanoid robots is extremely high, and the price of one unit often reaches hundreds of thousands or even millions of yuan, which makes the vast majority of enterprises hesitate. Under the current technology, the things that humanoid robots can do often do not match their high cost.

ByteDance's entry may take a different commercialization path. Relying on its huge content ecosystem and business scenarios, ByteDance may prioritize placing robots in verified links such as internal logistics and warehousing, and then gradually explore directions such as live streaming and content production. It is currently confirmed that ByteDance's logistics robots have been mass-produced over 1,000 units, mainly used for sorting and transportation in Douyin E-commerce warehouses, with external customers such as SF Express and BYD Electronics; the use of humanoid robots in live streaming and content production is still just a vision. It will first reduce costs through internal consumption, and then gradually export products to external enterprises.

But this is still a long way to go. From the demonstration video in the laboratory to mass production and delivery, from the factory assembly line to thousands of households, there are countless sleepless nights concerning gears, motors, data and costs to get through.

Conclusion

ByteDance's full effort to recruit robot engineers and build an embodied intelligence hardware team is a vivid epitome of Chinese tech giants moving into the deep water zone of hard technology.

This road is extremely long, and even more money-burning than large models. It reminds everyone that in the physical world, the Internet-style "magic" can hardly work immediately, and all algorithms must eventually pass the test of gravity.

ByteDance's entry may just be the beginning of this grand transformation. In this competition, the real winner will be the enterprise that takes the lead in finding the business balance point between computing power and gravity.

This article is from the WeChat official account "Tech Business", written by Tech Business, and authorized for release by 36Kr.