Physical AI in China: three companies, three paths

Key Takeaways

During CES 2026, Jensen Huang stated that ‘we’re entering the era of Physical AI’. The remark was a strong reminder that artificial intelligence (AI) applications are expanding from the digital world into the physical world. As AI becomes increasingly integrated into robotics, autonomous vehicles and smart factories, the AI megatrend is entering a new phase.

On the other side of the Pacific, Physical AI is also developing rapidly. Supported by a deep hardware supply chain, large-scale manufacturing capacity, competitive technology firms and industrial policy support, China is playing an increasingly important role in the development of Physical AI.

My recent trip to China made the theme feel even more tangible. Watching humanoid robot demonstrations during onsite visits to robotics companies, taking robotaxi trips, or simply ordering bubble tea delivered by drone all reinforced the same point: Physical AI is beginning to move beyond research labs into real commercial settings. Against this backdrop, the three Chinese companies discussed below illustrate distinct stages in the commercialisation of Physical AI.

Three Companies, three paths to Physical AI

UBTech: scaling humanoid robotics

UBTech is now closely associated with humanoid robotics, but the company’s origins are more conventional. Its earlier businesses in education, cleaning and logistics robots provided engineering experience, manufacturing know-how and relationships with industrial customers before humanoids became commercially relevant. These legacy businesses also provided a more resilient revenue base while humanoid robotics remained at an early stage.

What structurally changed in 2025 was UBTech’s revenue contribution. According to its earnings for FY2025, revenue from full-size humanoid robots rose sharply to 41% of group revenue, compared with only 3% in 2024. More than 1,000 humanoid units were shipped during the year, suggesting the industry is moving toward early deployment.

UBTech’s investor relations team appeared more convinced, when I visited them, that the near-term use cases for humanoids are in factories rather than households. The Walker S series is being tested in manufacturing and logistics scenarios, including Foxconn's 3C electronics factories, for tasks including handling, sorting and quality inspection. This may reflect a broader trend: factory environments are comparatively structured and repetitive, making them a more realistic starting point for humanoid adoption.

The next challenge is scale rather than proof of concept. Management is targeting 5,000 humanoid shipments in 2026, with planned production capacity of 10,000 units. The latest Walker S2 can also autonomously swap its own batteries, helping improve uptime in industrial settings.

UBTech’s strategy is not only about hardware. The company has developed its own Physical AI model, Thinker, while also building a broader automation ecosystem combining humanoid robots and autonomous logistics vehicles. On manufacturing, UBTech mainly designs key humanoid components while relying on suppliers for production, but its recent acquisition of Fenglong Electric suggests a move toward greater supply-chain integration.

UBTech’s humanoid strategy builds on capabilities developed through its earlier robotics businesses, but applied to a far more ambitious and still early-stage market.

Pony AI: robotaxis beyond pilot programmes

Pony AI represents the autonomous mobility side of Physical AI. Founded by former Baidu autonomous-driving engineers, the company develops software for robotaxis, robotrucks and autonomous-driving licensing, with robotaxis increasingly becoming the spotlight of the investment case.

Pony AI's autonomous vehicles are now operating in designated areas across multiple Chinese cities. Pony AI has also started expanding internationally, launching commercial robotaxi services in Croatia, while pursuing deployments and partnerships in the United Arab Emirates and Singapore. Autonomous driving is not only a technology challenge, but also a regulatory and local operating capability. Expanding across different road systems and regulatory frameworks is an important test of scalability.

The attraction of robotaxis is straightforward. Unlike many AI applications that remain digital, autonomous vehicles interact directly with the physical world. That creates a much higher technical and regulatory hurdle, but also a clearer commercial use case if deployment costs continue to fall.

Pony AI’s recent results showed strong growth in robotaxi revenue from a small base, although robotruck and licensing revenue still account for a meaningful share of the business. The debate now is less about whether robotaxis can technically operate, and more about whether utilisation rates, regulation and fleet economics can support sustainable returns on capital.

Compared with humanoid robotics, autonomous mobility appears further along in commercialisation, with operating services already providing a clearer indication of potential adoption.

Shenzhen Dobot: the more tangible side of Physical AI

Shenzhen Dobot represents the more tangible side of Physical AI. Its core business is collaborative robots, used in manufacturing, education and commercial settings. These robotic arms are already deployed for tasks such as assembly, inspection, sorting and material handling, particularly in automotive, electronics and semiconductor production.

Unlike many humanoid robotics companies, Dobot already generates revenue from established factory automation products. The company is not relying entirely on future humanoid adoption, but is also gaining exposure to current industrial automation demand and factory spending cycles.

Traditional industrial robots are often expensive to install and designed for narrowly defined tasks. Cobots are easier to integrate into existing production lines and can adapt more quickly to changes in workflow, making adoption more realistic beyond large-scale factories. Dobot already has an installed base of more than 100,000 cumulative shipments across multiple countries. Existing customer relationships may also help the company introduce more advanced embodied-AI products over time.

The company is also exploring broader embodied-AI applications. Management describes a ‘one brain, multiple bodies’ approach, using shared software architecture across robotic arms, humanoids and mobile robots. In 2025, Dobot launched products including a humanoid, but the revenue contribution remains small.

Compared with humanoid robotics, Dobot offers more direct exposure to current automation spending and factory upgrades, where commercial adoption is already taking place.

Conclusion

Together, UBTech, Pony AI and Dobot show that Physical AI is not a single product category, but a transition across different machine types and end markets. UBTech brings AI into humanoid robots, Pony AI brings AI into vehicles and Dobot brings AI into factory automation.

The common thread is embodiment: AI is connected to hardware that can interact with the physical world. These companies also illustrate why China is becoming increasingly important to the theme. China’s manufacturing ecosystem, fast product iteration and domestic automation demand make it a key market for early Physical AI deployment.

Investing in the opportunity with WisdomTree

WisdomTree Physical AI Humanoids and Drones UCITS ETF (WPAI) provides targeted exposure to the Physical AI transformation by investing across the ecosystem, spanning humanoid robotics, drones, autonomous mobility, smart manufacturing, logistics automation, and emerging real-world AI applications. All three companies in this blog are also included in WPAI¹.

Physical AI remains an emerging and rapidly evolving theme. Companies operating in robotics, autonomous mobility and AI-enabled automation may face significant risks including technological hurdles, regulation, supply-chain disruption, competitive pressures and slower-than-expected commercial adoption. Many companies in the sector remain exposed to high research and development spending and may experience elevated share-price volatility.

¹ WisdomTree, as of 11 May 2026.