Nvidia’s vision for humanoid robots now has a blueprint. The company this week released Isaac GR00T, its first open-source reference design for a humanoid robot, a specification sheet and software stack that any manufacturer can build on. The chassis is the H2 Plus from Unitree, the Chinese robotics company; the hands are a different story.
For the dexterous end of the machine, Nvidia chose Sharpa Wave, a company that didn’t exist two years ago. Its hands pack 22 degrees of freedom and tactile sensors with a claimed sensitivity of 0.005 newtons, fine enough to feel a sheet of paper between two fingers. Nvidia executives described the choice as a test of the open design’s promise: the best part wins, regardless of where it was made.
The choice is telling. Nvidia isn’t building the robot; it’s building the ecosystem around it, chips, software and simulators, and letting a global supply chain plug in. Unitree for legs, Sharpa Wave for fingers, Nvidia for brains, and any manufacturer for the rest: that is the reference design as a shopping list rather than a product.
The same day, Unitree cleared its IPO review on Shanghai’s STAR Market, moving toward a listing that could make it the first major humanoid-robot company to go public in China. The convergence of the two events, one American software company anointing a Chinese chassis and one Chinese robot maker passing its listing hurdle, captured the state of the industry in a single news cycle.
Analysts said the reference design effectively anoints Unitree’s platform as the default starting point for a generation of developers, a role Nvidia has played in PCs, graphics and self-driving cars. The company’s playbook is consistent: standardize the interface, let the ecosystem compete on the parts, and collect the rent on the computing that runs everything.
The convergence is the story. Nvidia’s hardware ecosystem is reaching into embodied AI, and Chinese robot makers have become indispensable nodes in a supply chain that once ran exclusively through the United States, Taiwan and Japan. Unitree’s STAR Market listing gives it capital to scale production; Nvidia’s reference design gives it credibility beyond China.
Why the hands matter: manipulation is the unsolved problem in humanoids. Robots can walk, and they are learning to navigate, but picking up an object of unknown shape and weight without crushing it remains the hard part, and tactile precision is the frontier. Sharpa Wave’s 22-degree hand with 0.005-newton sensitivity is aimed at that problem, and its inclusion in the Nvidia blueprint puts a two-year-old startup in the same sentence as companies that have spent a decade on grippers.
Unitree’s road to the listing was fast by Chinese standards. The company, founded in 2016, built its name on quadruped robots before moving to humanoids, and its machines are already used in research labs and security patrols. The IPO proceeds are earmarked for production capacity and overseas expansion, according to people familiar with its plans, and the Nvidia partnership is expected to anchor its international pitch.
Competition looms on every side. Tesla’s Optimus is the most visible humanoid program in the West, Figure and 1X have raised large rounds, and Chinese rivals including UBTech are pushing their own designs. Nvidia’s answer to the crowding is the reference design: rather than pick a winner, it supplies the parts everyone needs.
Market timing is another factor.
The open-source decision is deliberate. Nvidia publishes the design files, the control stack and the simulation environments, so a manufacturer with a chassis and a budget can assemble a robot without paying Nvidia a license fee. The company makes its money on the chips and software inside the machine, not on the blueprint, and the open approach is designed to speed adoption of both. Humanoid production is years from mass adoption, and analysts caution that reference designs outnumber deployed robots by a wide margin. But the flow of capital is real: robotics startups raised record sums last year, and the Nvidia blueprint gives investors a standard against which to judge every pitch.
The deeper point is Nvidia’s. Every robot built on Isaac GR00T runs Nvidia silicon and software, and the company has said it views embodied AI as the next platform after data centers. The reference design is the same strategy that won the PC era, applied to machines with arms and legs.
For Sharpa Wave, the deal is a fast track. A company that was a lab curiosity two years ago is now embedded in the industry’s most visible blueprint, and order books tend to follow blueprints. The robot industry has spent years arguing about whether to build humanoids at all. Nvidia just answered differently: build the parts, standardize the interface, and let the market assemble the machine. Unitree’s IPO and Sharpa Wave’s hands suggest the assembly line is already forming.
