Unitree Robotics' Nvidia-backed H2 Plus has sparked debate in China over who controls the robot body, AI brain and autonomy in a global ecosystem.
Announced by Nvidia CEO Jensen Huang at GTC Taipei on June 1, H2 Plus is due in the second half of this year. It combines Unitree's H2 humanoid robot and Sharpa five-finger dexterous hand with Nvidia Jetson Thor and Isaac GR00T for reasoning, control, data and model deployment, as reported by Global Times, OFweek, and Securities Times.
The debate is not simply alliance versus dependence. As humanoid robots near mass production, few companies can own the full stack alone — from body and brain to tools, data, applications and sales channels.
Unitree owns the body, Nvidia brings the brain
Unitree's real leverage is its robot body, not Nvidia's computing platform.
Humanoid robots are not just chips and algorithms. The hard part is physical engineering: joints, motors, reducers, servo drives, balance, motion stability and fall protection. Unitree has strong in-house capability across these areas, with its self-developed rate for core whole-machine components described as above 90%.
That makes H2 Plus a clear trade-off. Unitree brings motion control, whole-machine engineering and cost control; Nvidia brings Jetson Thor, Isaac GR00T, edge AI inference, simulation tools and a global developer ecosystem. Nvidia needs a body. Unitree needs a stronger brain.
Unitree market director Huang Jiawei said H2 Plus's main upgrade is computing performance, while founder and CEO Wang Xingxing said it gives teams a verified starting point for real-world robotic skills.
For Unitree, H2 Plus lowers customer development costs, improves overseas adoption by developers, universities, research institutions and enterprise pilot users, and helps a capital-market-bound company raise visibility and shorten product definition cycles.
China's robot autonomy depends on boundaries
The risk is not cooperation itself. The risk is unclear control. Robots differ from consumer electronics because they act in the physical world. A phone failure may interrupt information or services. A robot failure can cause loss of motion control, task failure, equipment damage, or safety risks on site.
Four boundaries therefore matter:
The first is data. Robots collect image, spatial, motion and task data in real environments, so ownership, anonymisation, model training rights and customer opt-outs must be clearly defined.
The second is software. Unitree, Nvidia and customers must clarify who controls motion control, task planning, AI inference and cloud management, and which layers can be replaced. Without a clear decision chain, maintenance, upgrades and trust will suffer.
The third is supply. Chips and development platforms face supply cycles, cost swings and roadmap shifts, so Unitree must ensure the same robot body can support multiple computing options.
The fourth is brand. Unitree is a whole-machine company, not a component supplier inside another ecosystem. Its long-term value will rest on system capability, service continuity and technology direction, not hardware pricing alone.
Robot autonomy now hinges on ecosystems, not isolation
The debate points to a broader shift. Humanoid robotics may follow smart vehicles into a layered division of labor, where companies control product definition, system integration, user experience and core data while working with chip, software, sensor and algorithm suppliers. Full-stack players, hardware platforms, embodied AI model providers, application developers and secondary-development specialists may all coexist.
Unitree already has strength in whole-machine engineering and cost control. Omdia's 2025 data put China humanoid robot shipments at 5,168 units for AgiBot, 5,500 for Unitree and 1,000 for Ubtech. Unitree is also said to lead global quadruped robot dog shipments.
The next race is to move robots from "able to move" to "able to work," requiring AI, software tools, developer ecosystems and real-world applications to mature together.
More than one deal, more than one player
China's concerns are understandable. Critics fear Unitree could become a high-end contractor, face supply chokepoints, leak technology or see domestic innovation weakened under US tech restrictions. Yet H2 Plus targets universities and research institutions, not a sweeping strategic alliance, and does not decide China's robotics roadmap or the sector's wider technology route.
Unitree still has room to build its own stack. In its IPO process, it plans to raise CNY4.2 billion (US$620 million), with nearly half earmarked for intelligent robot model development. Since 2024, it has increased investment in embodied large models, shifting from hardware sales towards a software-hardware integrated, AI-driven platform.
China's robotics ecosystem is no longer a one-company race. AgiBot, Ubtech and other local players are rising, while some companies use Nvidia Jetson Thor and others develop around domestic platforms such as Huawei Ascend. One deal cannot hijack the industry's autonomy push.
China's AI, semiconductor and robotics industries now have a stronger domestic base and more fallback options. Even in extreme conditions, robot makers such as Unitree could shift the robot brain from Nvidia to domestic alternatives.
Openness and autonomy are not opposites
For a Chinese robot maker seeking global markets, cutting itself off from global developers, customers and toolchains would be self-limiting. The key is whether Unitree can use Nvidia's ecosystem while retaining control of technology direction, product rhythm, data boundaries and system integration.
Unitree should not be judged simply for working with Nvidia. In humanoid robotics, few companies can own the whole stack alone. The real test is whether Unitree can turn global collaboration into leverage for China's robotics intelligence stack, rather than dependence on someone else's.
Article edited by Jerry Chen
