Goldman’s Chat With Top Robotics Firm Reveals Skynet Humanoid Timeline

In recent weeks, Ant Group, the Alibaba-affiliated fintech giant, and Mark Zuckerberg’s Meta Platforms have both ventured into the humanoid robotics race, while Elon Musk’s Tesla has been developing a bi-pedal, autonomous humanoid robot for the past several years. 

Goldman’s Jacqueline Du, Yuichiro Isayama, and others provided clients last Sunday with an investment guide to “Skynet”—outlining the companies most exposed to the global humanoid robotics industry.

Goldman’s Investing Guide To Skynet https://t.co/ZRKpiIUFok

— zerohedge (@zerohedge) February 23, 2025

On Thursday, Jacqueline Du, Zhou Li, and other analysts at Goldman penned a new note about key takeaways from meetings with China humanoid robotics company Unitree. The title of their note is “China Humanoid Robots: Unitree (private) trip takeaways: Robust hardware performance albeit still not ready for useful tasks.” 

Du outlined that Unitree’s H1 humanoid robot “is still incapable of handling complex and delicate tasks,” adding, “For commercialization progress, the company commented that humanoid robots can hardly achieve same-level efficiency as human workers in 2-3 years but meaningful application may arise over the horizon of 5-10 years.” 

Here’s more color from the analysts about their meeting with Unitree to gauge what’s happening in the industry: 

As part of our post-CNY pulse check field trip (Feb 24 – Feb 26, 2025), we visited Unitree (private) and watched its live demonstration of H1 humanoid robots and robot dogs.

Unitree showcased the capabilities of its H1 humanoid robot, which include walking and dancing, with robust hardware performance and strong autonomous self-balancing. This could be attributed to the company’s substantial motion control experience from its robot dogs development. During our previous field trip in 2023 and booth visit at 2024 CIIF (China International Industrial Fair), we understand from Unitree’s technical staffs that the underlying technology of motion control for humanoid robots and robot dogs is largely the same. However, H1 only have 19 DoFs (Degree of Freedom), and thus is still incapable of handling complex and delicate tasks. For commercialization progress, the company commented that humanoid robots can hardly achieve same-level efficiency as human workers in 2-3 years but meaningful application may arise over the horizon of 5-10 years. The company’s current humanoid robot shipment is at a few hundred units, meeting demand from R&D in universities’ labs and robot data collection & training.

The company also demonstrated its GO2 robot dog, showcasing its leading technology in robot dogs’ gait control algorithms. Unitree delivers thousands of units of robot dogs (all models including GO2 and B2, etc.) annually, mainly applied to patrol & inspection and substitution of hazardous tasks (fire rescue, power substation/nuclear power plant inspection). Unitree has a 60-70% market share in the quadruped robot market globally with 50% overseas revenue contribution, per the company.

The company has successfully equipped its robot dogs with LLM API, enabling robot dogs with voice control capability. However, robot dogs with LLM is incompetent for fully autonomous operation. Unitree’s humanoid robot only adopted rotary actuators (planetary reduction gear) in terms of hardware design, so further software and hardware iterations are required for large-scale applications. Per the company, its humanoid robot has experienced technology iteration with substantial improvement every year. As a reference, each generation of Unitree’s robot dog can improve its athletic performance (speed, reactivity, step climbing and self-balance ability, etc.) by 2-3X.

All in, the takeaways from our Unitree visit are in line with our view that the technology inflection point for humanoid robots remains unclear due to insufficient capability handling multiple generalized tasks. We model 76k/502k units of global humanoid robot shipments by 2027E/2032E, which is at a slower pace than market expectations as we believe it will likely take longer for AI-empowered robots to be ready. We believe investment opportunity at this stage remains in supply chain component names as we see a comprehensive ecosystem taking shape. We prefer actuator assemblers and harmonic gear suppliers on higher product adoption certainty and stronger technology barriers – stay Buy on Sanhua, Neutral on Leaderdrive, Best Precision and Moons’ Electric on valuation.

What’s more realistic this decade is the proliferation of robotic dogs, which are already being deployed on modern battlefields worldwide. The rise of Skynet is unlikely to be widely seen until the early 2030s, but in the meantime, big tech firms will continue showcasing their robots—much like Tesla has periodically done with Optimus.

This is coming…

. . .