Nvidia and Unitree unveil H2 Plus, an off-the-shelf humanoid for university labs
Nvidia and Chinese robot maker Unitree announced the H2 Plus on June 1, the first reference humanoid built on the Isaac GR00T stack. The six-foot, 150-pound robot bundles Unitree's H2 chassis, Sharpa five-finger hands, and a Jetson Thor compute module, with Ai2, Stanford, ETH Zurich, and UC San Diego signed up as first users. It ships in late 2026.
The H2 Plus is not a new robot so much as a standardized one. Each lab gets the same 31-DOF body, the same 22-DOF Sharpa Wave hands, the same Jetson AGX Thor T5000 with 2,070 FP4 teraflops, and the same Isaac GR00T software stack covering teleop, simulation in Isaac Lab, and ROS middleware. The pitch to researchers is that they stop burning grant money on bespoke integration and start comparing results on identical hardware. Nvidia is also routing software updates through its own chips with secure boot and confidential computing, which is how it answers the obvious question about putting a Chinese-built body in a US university lab.
The strategic read is that Nvidia has decided the humanoid market will not be won at the body layer. Unitree builds the chassis, Singapore's Sharpa builds the hands, and Nvidia owns the compute and the training pipeline that everyone above the silicon depends on. It is the same playbook Nvidia ran on data center AI, applied to physical AI a year earlier than most expected. Unitree gets distribution into elite Western labs it could not have reached on its own; Nvidia gets every published paper from those labs citing GR00T as the substrate.
What this does to the rest of the field is the more interesting question. Boston Dynamics is shipping Atlas to Hyundai factories and Tesla still has Optimus in R&D, but neither is courting the academic pipeline that produces the next generation of manipulation policies. If Stanford and ETH grad students spend the next two years publishing on H2 Plus, the foundation models that emerge will be tuned to Unitree kinematics and Sharpa fingers, not Atlas or Optimus. That is a quiet form of lock-in that compounds long after the late-2026 ship date. It also parallels what Nvidia is doing on the AI infrastructure side — see our analysis of Nvidia's broader supply-chain consolidation — where the same vertical-stack instinct shows up in silicon rather than robot bodies.
Price has not been disclosed, which usually means it is high enough that the press release would rather not lead with it. The base Unitree H1 launched at around $90,000 and the H2 Plus carries far more hardware. For the dozen labs Nvidia named, cost is not the constraint; for everyone else, the reference design will function as an aspirational spec sheet until a stripped-down variant appears. The bet Nvidia is making is that one standardized expensive robot in a hundred top labs produces more usable data than a thousand cheap ones scattered across hobbyists. That bet looks correct on paper. Whether it survives contact with how academic robotics actually publishes is the part worth watching.
CNBC: Nvidia picks Unitree for humanoid robot platform → · NVIDIA Newsroom: Isaac GR00T Reference Humanoid Robot → · PR Newswire: Unitree Announces H2 Plus →