Nvidia and Hon Hai say their robots improving at building AI components

‘MACHINES THAT MAKE THE MACHINES’: Hon Hai said it plans to use Nvidia’s model-driven humanoid robots in other factories in Texas, Wisconsin and California

Nvidia Corp and Hon Hai Precision Industry Co (鴻海精密) have improved the efficiency of critical assembly steps for Nvidia’s GB300 NVL72 rack-scale artificial intelligence (AI) platform by using robots, the companies said on Friday.

Nvidia detailed in a technical blog titled “The Machines that Make the Machines” how its operations team and Hon Hai trained robots to assemble critical GB300 NVL72 components, improving manufacturing stability and efficiency.

Success rates for assembly steps exceeded 90 percent, it said.

Nvidia Corp’s GB300 NVL72 rack-scale artificial intelligence platform is pictured at Hon Hai Tech Day at the Taipei Nangang Exhibition Center on Nov. 21 last year.

Photo: CNA

Nvidia also said the two companies focused on assembling the GB300’s tester tray, which is used to verify the functionality and performance of GB300 computing modules before shipment.

Key tasks included bus bar assembly and the insertion of multiple electrical connectors, it said.

For bus bar assembly, Nvidia said a major challenge was getting robots to grasp and transport a metal bar that carries high-voltage electricity to the tray, insert it and secure it with screws at 16 locations.

For multi-connector insertion, robots must lift two large and two small cable-mounted electrical connectors from the tray and insert them into sockets with tight clearances, it said.

According to test data, the automation system hit a success rate of more than 95 percent for bus bar assembly, which requires tightening 16 screws, with a target completion time of 124 seconds.

For multi-connector insertion, involving four cable connectors of different sizes, the success rate ranged from 90 percent to 95 percent, with a target completion time of 72 seconds.

During GB300 tester-tray assembly, Nvidia said flexible automation systems are needed to address uncertainty, because rapid design cycles and relatively low production volumes make fixed automation infeasible.

Robots need to be adaptive and robust enough to handle variations in the assembly process, it said.

Hon Hai, known globally as Foxconn Technology Group (富士康), is one of Nvidia’s major suppliers. The company announced last year that it was working with Nvidia to build a smart manufacturing plant at its production site in Houston, Texas, using Nvidia’s model-driven humanoid robots to produce AI servers.

Hon Hai has said it plans to expand its use of robots developed with Nvidia to other plants in Texas, Wisconsin and California.

The company, which accounts for more than 40 percent of the global AI server market, remains upbeat about the sector. At an investor conference in August, Hon Hai said that its AI rack shipments should more than double from a year earlier this year.