During Nvidia’s developer conference, GTC 2024, held at the Signia Hotel in San Jose, California, on March 19 (local time), Jensen Huang, the CEO of Nvidia, engaged with a global press corps, highlighting the significance of High-Bandwidth Memory (HBM) technology, describing it as a “technology miracle.” HBM, which vertically connects DRAM to enhance data processing speeds dramatically, is a crucial component in Nvidia’s AI semiconductors. Huang then referred to South Korea as the world’s most “advanced” memory producer, predicting that Samsung Electronics and SK Hynix would embark on significant growth cycles. He even encouraged contemplating the extent to which Korean companies’ HBM would infiltrate data centers worldwide, paralleling Nvidia’s growth trajectory.
Currently, SK Hynix effectively monopolizes the supply of HBM for Nvidia’s advanced AI semiconductors. While Micron’s HBM has been integrated into Nvidia’s ‘H200′ since last year, Samsung Electronics lags in deliveries. When questioned about the utilization of Samsung’s HBM, Huang revealed that although not currently in use, Samsung’s HBM is undergoing qualification. This acknowledgment essentially confirms rumors circulating in the market suggesting Samsung’s HBM had not yet passed Nvidia’s quality tests. Huang’s discussion marks the first instance of Nvidia directly addressing Samsung’s HBM usage. However, as Nvidia plans to commence production of next-generation AI semiconductors, such as ‘Blackwell,’ later this year, the mentioned ‘qualification’ likely refers to stability testing for the cutting-edge HBM3E product. Samsung Electronics plans to commence HBM3E production in the first half of this year. Huang also expressed the value of partnerships with SK Hynix and Samsung Electronics, leaving room for the possibility of utilizing Samsung’s products in the future.
Addressing the possibility of manufacturing Nvidia’s advanced AI semiconductors at Samsung’s foundry, Huang affirmed the possibility, highlighting Samsung as an “extraordinary” company. Despite Nvidia’s collaboration with TSMC, Huang stressed that Samsung’s legacy semiconductor production capabilities remain unmatched. However, he suggested that while legacy semiconductor production might be entrusted to Samsung, the production of cutting-edge AI semiconductors would remain solely with TSMC.
