NVIDIA's Horizontal Humanoid Platform

NVIDIA is trying to become the Android layer for humanoid robotics, not the Apple of one robot. Project GR00T is built as shared model, simulation, and compute infrastructure that many robot makers can plug into, which helps NVIDIA spread its chips and software across the whole category. That is very different from Google tying Gemini Robotics closely to Apollo, where the AI gains are more tightly linked to one hardware champion.

• NVIDIA launched GR00T as a broad platform for humanoid developers including 1X, Agility, Apptronik, Boston Dynamics, Figure, Sanctuary AI, Unitree, and others. The strategy is to sell the picks and shovels, Jetson and Thor compute, Isaac software, Omniverse simulation, and foundation models, no matter which robot brand wins deployments.
  2 nvidia 3 nvidia 4 sacra
• Google has taken the opposite path in this stack. Gemini Robotics is designed to transfer across robot forms, but Google publicly named Apptronik as its humanoid partner and highlighted Apollo as the robot being built with Gemini 2.0, giving Apptronik a tighter product tie to one frontier model provider.
  1 deepmind 5 deepmind
• For companies like Figure and 1X, NVIDIAs horizontal model lowers the cost of getting world class AI tooling without giving either an exclusive edge. That broad access raises the baseline for the whole field, but it also means differentiation shifts back to who collects the best task data, ships the most reliable hardware, and wins real fleet deployments.
  4 sacra 5 deepmind 6 sacra

The next phase of humanoids will likely split cleanly into horizontal AI infrastructure providers and vertically integrated robot companies. If NVIDIA keeps supplying the common brain and training stack to many OEMs, advantage will accrue to the robot makers that turn shared tools into proprietary deployment data, faster iteration, and customer locked in workflows.