NVIDIA and Foxconn partner to automate GB300 assembly with AI robotics

NVIDIA and Foxconn are collaborating to advance robotic automation in the assembly of Grace Blackwell GB300 test trays. The initiative aims to transition complex assembly tasks from skilled human workers to intelligent robotic systems to meet the surging demand for AI servers.

The project focuses on two highly difficult tasks: busbar assembly and multi-connector insertion. Busbar assembly involves robots picking up heavy metal components and securing 16 screws. Current success rates have exceeded 95%, though the process takes approximately 160 seconds, missing the 124-second target. Multi-connector insertion, which requires handling flexible cables with tight tolerances, currently sees success rates between 90% and 95%, with a target of 99.5%.

To overcome challenges like cable deformation and part variability, NVIDIA is utilizing the Isaac Lab for reinforcement learning and DOPER object pose estimation technology. The ultimate goal of this partnership is to establish an “AI manufacturing AI” cycle, where AI-driven robots assemble the hardware used to train the next generation of AI models.