A view of the server rack for Nvidia’s next-generation artificial intelligence (AI) computing platform “Vera Rubin.” LG Electronics is expected to begin full-scale mass production of Vera Rubin server racks next year. Nvidia
LG Electronics (066570.KS) is developing an AI chip-based server rack targeting Nvidia and plans to begin mass production next year. The move expands the recent AI infrastructure alliance between LG and Nvidia into server rack manufacturing.
According to the semiconductor industry on the 12th, LG Electronics’ Production Research Institute (PRI) completed development of an AI server rack prototype matching the compute tray specifications of Nvidia’s “Vera Rubin” in the first half of the year. LG plans to secure orders from global big tech companies after reliability evaluations in the second half and begin mass production next year. The research and development was conducted at a 1.8-megawatt data center cluster established at LG PRI’s Pyeongtaek site.
A compute tray is a module that integrates computing units such as graphics processing units (GPUs) into a single board, with dozens of trays stacked vertically to form a single rack. In the Vera Rubin rack, a total of 108 AI chips operate simultaneously, generating enormous heat, so how quickly and stably the heat is cooled determines the performance of the AI server rack.
LG Electronics has transferred the cooling technology accumulated over 58 years, since it launched Korea’s first home air conditioner in 1968, to its AI racks. Already holding air-cooling and water-cooling technologies for AI servers, LG Electronics applied a “direct-to-chip (DTC)” cooling method to this AI rack, circulating coolant around the chips to control heat generation.
AI server rack manufacturing is the final piece that will complete LG Group’s data center turnkey strategy. LG is targeting the AI data center construction market with its “One LG” strategy, which combines LG Electronics’ heating, ventilation and air conditioning, LG Energy Solution’s batteries, and LG CNS’s data center construction capabilities. LG Chairman Kwang-mo Koo is known to have discussed expanding data center cooperation, including AI rack supply, with Nvidia CEO Jensen Huang, who visited the company’s Yeouido headquarters last month.
“With server rack development, LG will cover all areas of AI data centers except the chip and power sectors,” an industry official said.
A compute tray of the Grace Blackwell (GB)200 rack, one of Nvidia’s computing platforms, shown pulled out of the rack. Nvidia