The HBM4 Battle Heats Up: NVIDIA Gives the Green Light to All Memory Giants

NVIDIA has officially completed the qualification process for the three leading global producers of advanced memory for artificial intelligence. During his visit to South Korea, CEO Jensen Huang confirmed that Samsung Electronics, SK Hynix, and Micron Technology have all been approved to supply HBM4 modules for the future Vera Rubin platform.

This is the first public confirmation from the American company regarding the suppliers of the new generation of high-bandwidth memory, a component considered essential to support the growing computational demands of next-generation AI systems. According to Huang, all three manufacturers have already reached production phase and are working to support the introduction of Vera Rubin. This is a notable update, considering that Samsung received NVIDIA's validation for HBM3E memory only later, after failing several tests due to issues related to consumption, temperatures, and missed performance requirements. This allowed SK Hynix first, and Micron, to gain market share.

The new platform, which NVIDIA expects to deliver starting in the third quarter of the year, has entered full production in recent weeks, as confirmed by the executive himself during Computex in Taipei. The Vera Rubin systems are built around the Vera processors and Rubin GPUs, combined with amounts on the order of terabytes of HBM4 memory per single server. This configuration highlights how the availability of high-performance memory has become a decisive factor for the evolution of generative artificial intelligence and HPC workloads.

The competition among Samsung, SK Hynix, and Micron to secure a share of NVIDIA's demand is particularly intense. The three companies dominate the global memory market used in advanced processing systems, and the growing prevalence of AI has transformed HBM solutions into one of the most strategic elements of the entire semiconductor supply chain.

Not surprisingly, during Computex, Huang organized a meeting with some of the leading South Korean partners, including SK Group Chairman Chey Tae-won and SK Hynix CEO Kwak Noh-jung, emphasizing the importance of the memory supply chain for NVIDIA's future plans.

The difficulties in ensuring sufficient production capacity for high-bandwidth memories represent one of the main challenges facing the industry. This is a sentiment shared by René Haas, CEO of Arm, who recently pointed out the memory sector as one of the most complex areas to manage in the current expansion of the global AI infrastructure.

During his stay in South Korea, Huang also announced that NVIDIA is hiring for a new research and development center in the country. The manager plans to meet with several local industrial entities, including Hyundai, LG, SK, Samsung, and Naver, confirming South Korea's growing strategic weight in the international semiconductor and artificial intelligence ecosystem.