NVIDIA is working on robots that could help build its newest AI chips. The company shared details of the research in a blog post published on October 7, 2026.
The project centers on the Grace Blackwell GB300 superchip. These chips power large AI models and high-performance computing systems.
The work is led by NVIDIA's Seattle Robotics Lab and its Isaac engineering team. The goal is to support skilled human workers with robotic systems.
What the GB300 Does
The GB300 is built on NVIDIA's Blackwell Ultra architecture. It is designed for AI training, inference, and reasoning, as well as for large business AI deployments.
Each GB300 compute tray combines four Blackwell Ultra GPUs with two Grace CPUs. The platform is built to handle large AI workloads while using energy efficiently.
The GB300 NVL72 setup can support up to 20 times more AI agents per megawatt than earlier systems, according to the report.
Two Hard Tasks for Robots
The research, described in a blog post by Elizabeth Goodman, looks at two assembly jobs used to build GB300 tester trays. These are busbar assembly and multi-connector insertion.
In busbar assembly, robots must place heavy electrical parts into exact positions and secure them. In multi-connector insertion, robots handle flexible cables and plug them into tight sockets.
Human workers do these tasks smoothly today. Robots struggle because the shape and position of parts can vary slightly.
Manufacturers such as Foxconn require success rates of at least 99.5%. Robots must also finish each task within strict time limits.
To meet these goals, NVIDIA is mixing traditional engineering with newer AI methods. These include imitation learning, reinforcement learning, and vision-language-action models.
The team also built custom grippers and used a vision system called DOPER. NVIDIA said these tools improved how well the robots performed.
Still, NVIDIA said its robots do not yet meet industrial standards for speed and reliability. Coordinating multiple robot arms and training with real-world data remain problem areas.
The company said combining real-world data with simulation training is key to making the robots ready for factories.
Labor is part of the reason for the work. Deloitte projects the U.S. could face a shortfall of 1.9 million manufacturing workers by 2033.
NVIDIA describes the effort as a cycle in which AI-powered robots build the hardware for future AI systems.
The company plans to prepare its robotic assembly tools for use in real factories. One possible site is a $700 million Wistron facility in Fort Worth, Texas, that builds GB300 systems.
NVIDIA has not given a timeline for when its robots will reach factory-level performance. As of October 8, 2026, NVIDIA shares were priced at $237.47.
