# NVIDIA Tests Robots to Assemble Tester Trays for GB300 AI Systems

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/27765
Published: 2026-10-08T06:04:04.000Z
Updated: 2026-10-08T06:04:04.000Z
Section: Technology

> The experimental setup handles busbar fastening and four-connector insertion, but the systems have not yet reached factory deployment.

NVIDIA is testing robotic systems for assembling tester trays used to verify GB300 compute modules, targeting precision work around heavy electrical parts and tightly packed connectors before shipment.

The company’s Seattle Robotics Lab and Isaac engineering team developed systems for two operations: attaching and securing an electrical busbar with 16 screws, and fitting four cable-mounted connectors into sockets with limited clearance.

The experimental setup includes two Flexiv Rizon 4S robot arms and a UR10e arm equipped with an OnRobot screwdriver. Foxconn, a GB300 contract manufacturer, helped set the performance targets.

Those targets require a success rate of at least 99.5%. The four-connector operation must also finish within 72 seconds, or no more than twice the time achieved by skilled workers. NVIDIA said the systems were nearing those thresholds but were not ready for factory deployment.

“As of the publication date of this article, we are rapidly approaching their thresholds,” NVIDIA robotics researchers wrote.

The work addresses manufacturing conditions in which cables can bend or deform and components can differ in shape and position. NVIDIA said the project combines conventional robotics engineering with pose estimation, simulation-based reinforcement learning, real-world policy training and custom gripper designs.

The company’s next objective is to place the technology in factories operated by its contract manufacturers. “Our next goal is to deploy our technology into the factories where NVIDIA contract manufacturers produce AI hardware,” the researchers wrote.

The project remains an experimental tester-tray effort rather than a completed production rollout for GB300 systems. NVIDIA has not disclosed when a factory deployment might occur or whether the assembly technology will become commercially available.

GB300 is part of NVIDIA’s Blackwell Ultra platform for artificial intelligence training, inference and reasoning. The GB300 NVL72 combines 72 Blackwell Ultra graphics processing units with 36 Grace central processing units in a liquid-cooled rack-scale system.

The platform includes 20 terabytes of GPU memory, 37 terabytes of fast memory and 130 terabytes per second of NVLink bandwidth. NVIDIA also said a 324,000-square-foot Texas facility produces Grace Blackwell Ultra systems as part of a combined $700 million investment in advanced U.S. manufacturing.