Overview

Most people still frame the humanoid robot race through the lens of US players like Figure AI, Tesla and Boston Dynamics – focusing on cinematic demos, frontier AI, and the question of “who has the smartest robot”. However, China’s Unitree forces a different question: what if the first real advantage in humanoids starts from manufacturing cadence, cost structure, fast iteration, and deployment volume?

Unitree is China’s leading robotics company and the clearest expression of a humanoid strategy that prizes scale before intelligence excellence. As Unitree prepares for its IPO, this report argues that China may not need to win the first round of this race by building the most intelligent robot. It can also win by building enough robots at low cost to create the real-world data and field-learning loop that gradually hardens the intelligence layer.

The report combines public IPO and production data with a founder-level, close-to-company perspective. It connects the key signals emerging in the same time window:

• Unitree fast-tracking its IPO and the plan to build 75,000 humanoids per year (June 1, 2026)
• Unitree’s humanoid robot H2+ partnering with NVIDIA’s Isaac GR00T (June 1, 2026)
• The 81-hour full-autonomy live demo from Unitree’s biggest US rival, Figure AI (Mid-May, 2026)
• The proposed GUARD Act from the US that could put the Unitree-NVIDIA collaboration under pressure (June 3, 2026)

Together, these signals reveal a much larger story – humanoid robotics is becoming a new front in the US-China AI race. The US path is centered on frontier autonomy, onboard intelligence and world-model-driven planning. China’s path is built around scale, supply chain integration, cost-down execution and faster deployment.

Through this lens, the report offers a clearer view of where humanoid robots are actually advancing, how the US and China are taking different routes, where Unitree’s opportunities and bottlenecks sit, and what the next phase of the race may look like.

Key Takeaways

• China’s Unitree is a compelling and representative case of taking humanoid robots from the demo stage to real-world deployment. While the US-based Figure AI currently offers stronger proof points in sustained, onboard end-to-end autonomy, Unitree’s advantages lie elsewhere – robust body control with agility, hardware scale (a complete product portfolio), cost structure (benefits from full-stack in-house development), and an expanding developer-customer ecosystem enabled by its NVIDIA partnership.

• Figure AI is trying to prove that a humanoid can work autonomously on a defined task. Unitree is trying to make sure enough humanoids exist for the whole industry to learn, iterate and scale. In one sentence, this reflects the China-US humanoid split: the US is racing to prove autonomy; China is racing to industrialize the platform.

• Unitree-NVIDIA partnership amplifies that if Unitree’s H2+ becomes a widely used reference robot, Unitree becomes part of the global physical AI development substrate. Many labs, companies, universities and developers around the globe can start building on the Unitree humanoid chassis.

• The partnership also becomes more of a US-China AI race story – whether the “body”, the “brain” and the development ecosystem can still be connected across geopolitical lines. Washington’s expanding national-security perimeter is making deep technical collaboration between NVIDIA and Unitree increasingly hard to sustain. Supplying Jetson Thor, integrating Isaac GR00T and building a deeper US-China robot-intelligence stack could face de facto suspension even before any formal ban arrives.

• Unitree’s 75,000-unit question is less about production capacity and more about autonomy conversion. Unitree must turn volume into a data-and-learning flywheel – more robots, more deployment, more real-world feedback, and better autonomy. Without that loop and its in-house robot “brain”, scale risks being discounted as mere hardware hype – impressive humanoid bodies, but not yet true autonomous labor.