WPAI
WisdomTree Physical AI, Humanoids and Drones UCITS ETF - USD Acc

Veröffentlicht am 28. August 2026
Associate Director, Quantitative Research & Multi Asset Solutions
Associate Director, Quantitative Research
Unitree is one of the best-known humanoid-robot companies, and videos of its robots dancing, boxing and backflipping are widely shared on social media. Its IPO drew the same attention. The retail offering was more than 8,000 times oversubscribed1, and its shares rose 629% intraday before closing 460% higher on their first day of trading. This made Unitree’s price-to-sales ratio soar to more than 200x2. Investors were placing a sizeable bet on the future of humanoid robots.
Although the market was putting a high valuation on the prospect that humanoid robots will eventually be widely used, the harder question is how long it will take for them to be deployed at scale in production environments? Unitree has lowered one barrier by offering relatively affordable, robust models. Its G1 humanoid was launched at around $16k, making it more accessible to researchers than many alternatives from other countries. Unitree reported shipments of more than 5,500 humanoid robots in 2025, making it the world’s largest supplier by shipment volume3.
However, one of the clearest counterpoints came from Unitree itself. Shortly after the listing, founder Wang Xingxing said humanoids’ “ChatGPT moment” could still be 5 to 10 years away4. This contrasts Jensen Huang’s commentary from January, though what they were referring to may differ. Jensen was speaking broadly about Physical AI whereas Wang was solely focused on humanoids. Wang’s definition of the ‘ChatGPT moment’ sets a very high bar: a robot placed in an unfamiliar environment should be able to complete most tasks through simple voice or text instructions. That is a much higher bar than a choreographed performance or a single task in a controlled setting, which are widely circulated on social media.
Unitree’s sales breakdown helps explain the gap. Nearly three quarters of Unitree’s humanoid revenue came from education and research users5. These customers use the robots to test applications, collect data and improve models. They are important early buyers, but their orders are not solid evidence of large-scale deployment in production environments such as factories or warehouses.
To reach that goal, humanoids need to make substantial progress in both locomotion and manipulation. Locomotion includes walking, running, balancing and recovering from disturbance. Manipulation is harder: a robot must reliably grasp, pick up, place and assemble different objects in unstructured environments. This requires more capable hands, but also stronger perception, more training data and more powerful Physical AI models.
The sharp pullback after Unitree’s initial surge suggests that investors are starting to distinguish enthusiasm for the theme from the time required for commercial deployment. The more useful investment question is therefore not simply which humanoid manufacturer will win, but where value will be captured as humanoid robots move towards scalable real-world use.
One answer lies upstream: rather than betting on a single OEM6, investors can gain diversified exposure through suppliers of the motors, actuators, reducers, batteries, power electronics and sensors that ultimately govern humanoids’ cost, dexterity and reliability.
Linear and rotary actuators together account for 51% of a humanoid’s bill of materials, making actuation the largest cost category7 (Figure 1). The chart also assigns 19% to dexterous hands, 10% to control systems, and 4% each to vision and other sensory systems and the energy system. While compute and batteries can leverage existing EV, semiconductor and consumer-electronics capacity, production readiness is weaker in precision actuation and force/tactile sensing as volumes rise.
Figure 1: BOM cost breakdown (by module/system, 2030)

Source : BofA Global Research, Greater China Industrials, “Who Makes the Humanoid Robot (in 2026)? Locomotion first, manipulation follows”, 20 April 2026
The recent World Humanoid Robot Games in Beijing showcased how far Chinese humanoids have progressed. Whether that progress can translate into scalable deployment will depend, in part, on the actuator supply chain’s ability to deliver components at lower cost and with consistent reliability. China’s advantage is the density of an EV-linked electromechanical ecosystem that can be repurposed for humanoids. Motors, power electronics, batteries, bearings and sensors can draw on established suppliers, tooling and production know-how, enabling local sourcing, multisourcing and faster design iteration as volumes rise8. China holds 40% of precision-bearing and encoder/position-sensor capacity, and 30% of driver-board and power-electronics capacity (Figure 2). These inputs are central to actuator systems, the largest cost category in a humanoid’s bill of materials.
Figure 2: Share capacity of rotary actuator components, by region %

Source : McKinsey & Co – “Turning humanoid supply chain constraints into billion-dollar wins” https://www.mckinsey.com/industries/industrials/our-insights/turning-humanoid-supply-chain-constraints-into-billion-dollar-wins
That manufacturing base is reinforced by listed Chinese specialists across the stack: Leaderdrive in reducers, Shenzhen Inovance in motors, Ningbo Tuopu produces linear and rotary actuators, hand motors and sensors, while Ningbo Zhongda Leader supplies precision reducers and motion-control components used in robotic joints. In 2025, Unitree, AgiBot and UBTECH together represented over 60% of global humanoid shipments, indicating that China’s component depth is already translating into production scale (Figure 3).
Figure 3: 2025 humanoid robot market share (by shipment volume)

Source: BofA Global Research, Greater China Industrials, “Who Makes the Humanoid Robot (in 2026)? Locomotion first, manipulation follows”, 20 April 2026
Hardware is necessary, but it is not sufficient to enable physical AI. Humanoids also require a software stack that links perception and task planning to safe, real-time motion, learning, and developer tools.
China is increasingly assembling this stack: Unitree has open-sourced VLA and world-model-action systems, while AgiBot combines real-robot data, simulation, GO-2, and Genie Envisioner. UBTECH, Alibaba, Galbot, Xiaomi, Tencent, XPeng, and Horizon add industrial models, fleet control, chips, and sim-to-real tooling9. By leveraging lower-cost, open-source platforms, Chinese companies can fine-tune Qwen-based10 models for humanoids and iterate quickly.
Data is equally critical for training these models, and China has begun establishing dedicated humanoid-robot data-collection centres across Beijing, Shanghai, Zhejiang, and other regions. These “data factories” use remote human demonstrations to generate large embodied-AI datasets for VLA training, world models, and sim-to-real validation11; It is estimated that a single action may require more than 600 training hours and 2,000 data samples10.
China’s lower-cost open platforms, factory access, data infrastructure, and rapid deployment could create a reinforcing data flywheel, positioning it as a major software as well as hardware player in physical AI.
Bringing humanoids into wider real-world use will require coordinated progress across software, components and manufacturing. The companies providing perception, actuation and motion-control technologies may therefore play an important role regardless of which finished robot brands ultimately lead.
The industry’s development will be shaped not only by the robots attracting today’s headlines, but by the full ecosystem required to make them commercially viable.
1 Reuters: Unitree's Shanghai IPO more than 8,000 times oversubscribed by retail investors
2 Bloomberg. P/S ratio is based on FY 2025 revenue.
3 Company filings, Unitree’s IPO Prospectus.
4 CNBC: Humanoid robots’ ‘ChatGPT moment’ could be 10 years away, Unitree founder says
5 Financial Times: Who is really buying China’s humanoid robots?
6 OEM : An OEM is the company that designs, builds, and brands the final physical hardware. Examples include robotics companies like Unitree, UBTech and XPeng Motors
7 BofA Global Research, Greater China Industrials, “Who Makes the Humanoid Robot (in 2026)? Locomotion first, manipulation follows”, 20 April 2026
8 McKinsey & Co – “Turning humanoid supply chain constraints into billion-dollar wins” https://www.mckinsey.com/industries/industrials/our-insights/turning-humanoid-supply-chain-constraints-into-billion-dollar-wins
9 BofA Global Research, Greater China Industrials, “Who Makes the Humanoid Robot (in 2026)? Locomotion first, manipulation follows”, 20 April 2026
10 Seeking Alpha, July 23, 2026 https://seekingalpha.com/article/4916847-alibaba-stock-group-qwen-robot-triggers-full-stack-physical-ai-inflection-point-reiterate-buy
11 Xinhua, 16 March 2026 https://english.news.cn/20260316/9a6dffb3472c4ea8a02b58af155d8d00/c.html
WisdomTree Physical AI, Humanoids and Drones UCITS ETF - USD Acc

Associate Director, Quantitative Research & Multi Asset Solutions
Baoqi Zhu joined WisdomTree in 2023 as a Senior Associate on the Research team. Baoqi focuses on quantitative research on thematic equity indices and portfolio solutions. Prior to WisdomTree, Baoqi spent over two years at Ernst & Young (EY) in their Quantitative Advisory Services, where he was involved in the research and development of quantitative risk models. Earlier in his career, Baoqi served as a quantitative analyst within a multi-asset structuring team at Maven Global for more than three years. His responsibilities included designing and optimising bespoke hedging strategies based on derivatives. Baoqi holds a MSc in Financial Engineering & Risk Management from Imperial College London and a BSc in Actuarial Science from Nankai University, China. He is also a certified Financial Risk Manager (FRM).

Associate Director, Quantitative Research
Blake Heimann joined WisdomTree in 2020 and supports the creation, maintenance, and reconstitution of WisdomTree’s indices. Blake began his career in finance in 2017 as an Analyst at TD Ameritrade and, later, a Quantitative Analyst with a focus on the development of machine learning applications in finance.
Blake has bachelor’s degrees in Mathematics and Economics from Iowa State University, as well as a master’s in Computer Science from Georgia Tech, with a specialisation in Machine Learning. He is currently pursuing a master’s in Finance from the London School of Economics