�۶���Ƶ

Imagine machines that move and look like humans. Humanoid robots have long captured the imagination, and their appeal is clear: human-shaped robots could operate in a world designed by humans for humans, climbing stairs, turning door handles, and using our tools and infrastructure. Millions of humanoids could take on a variety of everyday tasks and seamlessly slot into our human world, alleviating labor scarcity, performing dirty and dangerous tasks, and leaving us humans more time for creative pursuits.

Today, many companies are investing in humanoid development. While we expect rapid progress, experts on our Thematic Equity Advisory Board¹ raised questions about form factor and cost. The world today may be suitable for humanoids, but will future factories, hospitals and warehouses be shaped differently to make use of the latest technologies and most cost-effective automation solutions? After all, you don’t need a humanoid robot to drive an autonomous vehicle.

Humanoid robots are costly, complex and relatively fragile compared to simpler more dedicated automation systems.

Professor Gery Colombo, member of Thematic Equity Advisory Board, March 2025

Humanoid robots are fiendishly complex. Agility Robotics suggests around 5,000 parts are needed² and 36 ‘degrees of freedom’³ for each joint, excluding the hands. While every degree of freedom increases the ability to perform human-like movements, such as walking and carrying objects, each one requires actuators, sensors, gears and motors, adding to the cost, weight, power consumption and complexity of the system.

Replicating the versatility of a human hand is a major challenge for roboticists. Even tying a shoelace is highly complex. The state-of-the-art ‘Dexterous Hand’ system from Shadow Robotics has 20 motors and 24 degrees of freedom, resulting in a price tag of USD 74,000.⁴ Ironically, hands and legs might not be the best solution. Wheels are far more efficient in certain environments, and a toolbox of plug-and-play ‘end effectors’ such as screwdrivers and suction cups might be more suitable than hands.

Table 1: Humanoid robots

Given the number of joints, it is no surprise that that actuators alone make up roughly half of a humanoid’s cost.⁵ Batteries, force sensors, optical cameras, LiDAR (light detection and ranging), and AI chips for processing, are also expensive. Given these cost pressures and huge complexity, there is a lot that could go wrong in a humanoid robot. How much certainty do we have that these machines will deliver 24/7 uptime and withstand bumps and falls or operate efficiently and reliably in dirty and dusty environments?

Bill of materials of a humanoid robot

Source: BoA Research, “Who Makes the Humanoid Robot”, 7.03.2025, reproduced by �۶���Ƶ AM.

The flexibility and physical capabilities, especially for dexterous hand manipulation, required for a genuinely useful household humanoid robot, appear unlikely in the foreseeable future.

Dr. Roland Siegwart,��Professor of Autonomous Systems at ETH Zurich, April 2025

Elon Musk, with characteristic ambition, targets production of 50,000 humanoids in 2026 and 500,000 in 2027.⁶ In contrast, Agility Robotics has set a more modest goal of producing “10,000 units over the next few years”.² China, leveraging its vast manufacturing and supply chain capabilities, also plans for a significant production ramp up in coming years.

Financial analysts forecast that 1 million humanoids will be in in production by 2030.⁷ For investors, this poses the risk of a ‘hype cycle’: lofty expectations set early on, leading to disillusionment after several years, before genuine use cases for the technology finally emerge. This pattern is familiar; autonomous driving systems, after decades of development, have been much slower to be deployed than initially expected. However, since humanoid robots are now starting to be employed in production, we may not have to wait long to find out if they can deliver a positive return on investment.

We expect humanoids will be successfully deployed in certain factory and logistics-center tasks, but we are more sceptical about their ability to handle a diverse and dynamic range of varied tasks. Large businesses tend to roll out new technologies in a deliberate manner, once they are tried and tested. Reliability is critical. Until production volumes are at scale, the high price tag of humanoids will also remain a barrier to adoption.

Non-humanoid robots already dominate the industrial landscape. Wheeled AGVs, CNC machines and robot arms are used extensively in factories and warehouses. They are typically designed to perform specific tasks, such as picking and placing items or welding, but in that task they are precise, fast, cheap, efficient and easy to maintain. As AI makes these dedicated systems more ’intelligent’ and easier to use, perhaps they are the real shape of the future of automation.

It seems we are still a long way from having fully versatile humanoid robots performing our household chores and doing our laborious work. However, progress has been fast, and early feedback from logistics companies running pilot tests with humanoids has been positive.⁸ The question is for what use cases and at what price point are they economically viable. Our Advisory Board members maintain some scepticism of the humanoid form and the speed with which physical AI will advance. However, they remain highly positive on the outlook for dedicated automation solutions designed to perform specific tasks and empowered by AI technology.

Nonetheless, we should not underestimate human willpower. When highly influential people, with huge balance sheets at their disposal, such as Jensen Huang (Nvidia) and Elon Musk (Tesla), make humanoid robots a central pillar in the future vision for their firms, we would be foolish not to keep an open mind. In fact, many corporate R&D budgets today dwarf those of government and privately funded programs, and with AI and other technologies progressing fast, it could be a case of “where there is a will, there is a way.”

Whether humanoid or dedicated machines are the future, the current buzz around physical AI (‘embodiment’) illustrates the innovative and dynamic nature of the AI and robotics theme. For the patient investor, we believe this theme will deliver above-market returns over the long term, and a pure-play approach offers both high exposure to the theme and diversification away from standard equity indexes.

^1. The Thematic Equity Advisory Board allows extensive dialogue between our investment team and experts from industry and academia
^2. Agility Robotics. .
^3. Animation of Human Walking in Virtual Environments, Shih-kai Chung and James K. Hahn. Institute for Computer Graphics, School of Engineering and Applied Science, The George Washington University. May 1999
^4. IEEE Spectrum. The New Shadow Hand Can Take a Beating. May 2024
^5. Bank of America Research. Who makes the Humanoid Robot. 7 March 2025
^6. Tesla Earnings Call, 2025-01-30
^7. Bank of America Research, Daiwa Capital Markets, �۶���Ƶ CIO Office Humanoid Report, Bernstein Research
^8. In house meeting at �۶���Ƶ Asset Management with GXO Logistics Investor Relations (18.02.2025)
^9. Agility Robotics. .��
^10. The RobotReport. How Agility Robotics crosses the Sim2Real gap with NVIDIA Isaac Lab. November 2024
^11. The RobotReport. Meet Agility Robotics' Digit bipedal at RoboBusiness. September 2022.��
^12. UBTECH. .��
^13. Just o Born. .
^14. Heise online. . January 2025
^15. Unitree Robotics. .��
^16. Robotics 24/7. Figure AI unveils second-generation humanoid robot Figure 02. August 2024.
^17. The RobotReport. Figure 01 humanoid takes first public steps. October 2023
^18. Viso.ai. Tesla Bot Optimus – General Purpose Humanoid Robot. May 2024
^19. Yahoo!Tech. Fox News. Humanoid robot breakdances its way into history. April 2025
^20. Linkdood Technologies. New Electric Atlas AI Humanoid Robot. April 2024