Forget the robot. Watch the stack underneath it. NVIDIA's Isaac GR00T push is not a hardware story or a moonshot R&D flex — it's a land grab for the software layer every humanoid robot in the next decade will run on, and most people covering it are still describing the chassis.
On June 1, 2026, NVIDIA announced the Isaac GR00T Reference Humanoid Robot for academic research, built around a Unitree H2 Plus chassis, Sharpa Wave tactile five-finger hands, Jetson Thor onboard compute, and the company's own open GR00T software and models. Read that sentence again. Almost every piece of it that matters — the compute, the simulation framework, the foundation model — is NVIDIA's. The chassis is the part you can swap.
That's the whole game.
The platform play hiding inside a research robot
GR00T started life at GTC 2024 as a general-purpose foundation model for humanoid robots. Two years later it's grown teeth. Isaac Lab handles GPU-accelerated simulation for training robot policies at scale. Jetson Thor, built on Blackwell, delivers 800 teraflops of 8-bit AI performance on the robot itself. The 'Physical AI & Embodied Intelligence Software Market Research Report 2034' describes NVIDIA's Isaac GR00T, Cosmos and Omniverse lines as collectively covering every layer of the physical AI stack — perception, simulation, foundation models, edge inference, cloud orchestration.
Every layer. That's not a product roadmap. That's a moat.
The reference robot is the Trojan horse. Hand a graduate lab a Unitree H2 Plus chassis, equipped with Sharpa Wave tactile five-finger hands, bringing the robot's total degrees of freedom across body and hands to 75, and a pre-baked NVIDIA software stack and a pre-baked NVIDIA software stack, and you are not selling them a robot. You are training the next generation of roboticists to think in CUDA, prototype in Isaac Lab, and ship on Jetson. By the time those students are running labs or startups, the default assumption about how a humanoid gets built will already be settled.
Why humanoids first, and why now
Counterpoint calls NVIDIA's approach a moonshot — solve the hardest robotics problem first, then let the advances cascade down to simpler autonomous systems. It sounds noble. It's also strategically ruthless. If you build the tooling that can train a bipedal robot to fold laundry, you've built tooling that can train basically anything else with wheels, arms, or grippers. The humanoid is the forcing function.
Jensen Huang has put the prize at a multi-trillion dollar economic opportunity. That sounds like CEO theater until you see euRobotics, citing MRFR analysis, projecting the humanoid robots market growing from around $40.85 billion in 2025 to roughly $1.04 trillion by 2035. Pick your skepticism level on those numbers. Even cut in half, the prize justifies the spend.
And NVIDIA is hedging its hardware bets. TNW reported the Unitree-Sharpa research robot is the first of several planned tie-ups, with US, European and South Korean partners to follow. Translation: NVIDIA does not want to be tied to any single chassis maker, especially a Chinese one, when export controls inevitably tighten. The robots are interchangeable. The software is not.
The part that could still break
None of this is inevitable. euRobotics' own analysis lists the unsolved problems honestly: high development and deployment costs, safety and trust gaps in human-robot interaction, missing regulatory frameworks, weak autonomy in dynamic environments, energy and battery limits, fine motor control, context-aware cognition. That is most of the actual job. A foundation model does not fix a battery.
There's also a competitive question NVIDIA can't answer alone. Open-source robotics stacks have a way of fragmenting. Tesla is building vertically. Figure and 1X are building vertically. The Chinese humanoid scene is doing its own thing on its own silicon. GR00T's pitch — use our stack, swap your chassis — only works if enough labs and startups actually take the bargain. Right now, with academic kits going out and reference designs published, the bargain looks attractive. In three years, with a Jetson lock-in and proprietary fine-tuned models, it may look different.
The interesting question isn't whether humanoid robots arrive. It's whether, when they do, every one of them quietly says "Powered by NVIDIA" on the boot screen. That's the bet Jensen is actually placing. The robot on stage is just the demo.
