Key Points:
- Nvidia CEO Jensen Huang officially announced the H2 Plus at Computex 2026, the first open humanoid robot reference design built on the Isaac GR00T platform.
- The hardware integrates Unitree Robotics’ six-foot-tall H2 body with Singapore-based Sharpa’s five-finger Wave hands, which feature 25 degrees of freedom.
- Powering the robot’s brain is Nvidia’s Jetson Thor computer, delivering 2,070 FP4 Teraflops of performance via its Blackwell GPU architecture.
- Designed primarily for academic researchers, the open-source platform aims to eliminate “franken-robots” and accelerate the development of physical AI.
The global robotics industry has officially entered a new era of standardized development. Speaking at the Computex 2026 technology conference in Taipei, Taiwan, on Monday, June 1, 2026, Nvidia Chief Executive Officer Jensen Huang unveiled the company’s first open humanoid robot reference platform. Named the H2 Plus, this state-of-the-art system aims to unify the highly fragmented hardware and software landscape for academic and frontier research. By creating an out-of-the-box blueprint, Nvidia and its strategic partners plan to dramatically reduce setup times for developers, shifting the industry’s focus from basic hardware troubleshooting to high-level artificial intelligence training and real-world deployment.
To bring this ambitious reference platform to life, the U.S. chip giant coordinated a major international partnership with two leading robotics hardware manufacturers. Nvidia selected Chinese robotics pioneer Unitree Robotics to supply the physical humanoid body, using its nearly six-foot-tall, 150-pound H2 framework. For dexterous physical manipulation, the robot uses specialized five-finger mechanical hands from the Singapore-based startup Sharpa. This collaboration marks a major step forward, as equipping humanoid systems with human-like, five-fingered hands has historically been one of the toughest mechanical hurdles in advanced robotics.
The true differentiator of the H2 Plus lies in its onboard computing power, which functions as the physical robot’s central nervous system. The platform runs on Nvidia’s Jetson Thor processor, a highly advanced computing module designed specifically for embodied artificial intelligence. Leveraging the company’s cutting-edge Blackwell GPU architecture, the onboard computer delivers a staggering 2,070 FP4 Teraflops of processing power, backed by a 14-core Arm CPU and 128 gigabytes of system memory. This immense processing capacity allows the robot to run complex, real-time AI reasoning and physical coordination models directly on the device without relying on external cloud networks.
While the physical hardware provides the body, Nvidia’s software stack provides the cognitive intelligence. The H2 Plus runs on the newly announced Nvidia Isaac GR00T foundation platform, a comprehensive software workflow that spans data collection, simulation, reinforcement learning, and real-world deployment. The system uses the GR00T 1.7 software stack and Unitree’s whole-body control mechanisms to execute complex physical movements smoothly. By unifying the entire software pipeline, Nvidia is giving developers the tools they need to train robots in virtual digital twins before transferring those learned skills directly to physical hardware in the real world.
Nvidia’s decision to launch an open-source reference design directly addresses a major bottleneck in academic research. Rev Lebaredian, Nvidia’s Vice President of Physical AI, explained to reporters in Taiwan that building humanoid robots at scale remains insanely hard for individual research labs. Historically, university laboratories have spent most of their time and funding on assembling custom hardware parts, resulting in unreliable franken-robots that struggle with basic movement. By providing a fully integrated, standardized platform, the H2 Plus allows academic teams to bypass the basic assembly phase and immediately focus on developing useful robotic skills for factories, warehouses, and logistics systems.
Although Unitree serves as the launch partner for this reference design, Nvidia is actively expanding its global robotics alliance. Company executives confirmed that Nvidia plans to work with other humanoid robot manufacturers in the United States, Europe, and South Korea to build out a diverse portfolio of research systems. This global outreach includes deep partnerships with major European industrial chipmakers like Infineon, NXP, and STMicroelectronics. These chipmakers supply critical physical components—including high-precision sensors, safe motion control, power management systems, and internal communication lines—to flesh out the physical bodies running on Nvidia’s Jetson Thor platform.
The financial motivation driving this massive engineering effort is the promise of a virtually untapped global market. Jensen Huang has repeatedly predicted that physical AI—the merger of artificial intelligence models with physical robotic systems—will soon become a market worth tens of trillions of dollars. Huang told investors last month that he expects to see rapid, exponential growth in the commercial robotics segment over the next five years. By establishing its chipsets and software platforms as the foundational brain of these systems early on, Nvidia plans to secure the same dominant position in robotics that it currently enjoys in the generative AI software market.
This technological push arrives as real-world industrial demand for automation reaches a critical tipping point. Facing projected global labor shortages of over 85 million people by 2030, multinational corporations are actively transitioning from experimental pilots to permanent robotic workforces. For instance, Japan Airlines deployed Unitree-based humanoid platforms at Tokyo’s Haneda Airport in May 2026 to solve severe ground-handling labor shortages. Similarly, BMW expanded its humanoid program to Europe following an 11-month pilot in South Carolina, where humanoid robots successfully loaded over 90,000 sheet-metal components to support vehicle manufacturing.
Ultimately, the launch of the H2 Plus reference design represents a defining moment for the commercialization of physical artificial intelligence. By bridging the gap between advanced computing power, robust software, and highly dexterous hardware, Nvidia has cleared a major hurdle for the next generation of roboticists. Unitree expects to make the H2 Plus commercially available in late 2026, while developers can access the Isaac GR00T reference workflows immediately on open-source platforms like GitHub and Hugging Face. As researchers around the world begin training this unified platform, the dream of deploying general-purpose humanoid robots into homes, factories, and warehouses is rapidly approaching reality.
