Recent announcements from Computex Taipei 2024 reveal that NVIDIA's CEO, Jen-Hsun Huang, disclosed during a global media briefing that Samsung's latest High Bandwidth Memory (HBM) chips have yet to receive official certification from NVIDIA. This certification milestone marks the final step before Samsung can commence the supply of HBM3 and HBM3e chips, which holds significant implications for NVIDIA's advancement in artificial intelligence (AI) technology.
Reports preceding this disclosure had indicated that Samsung's HBM modules faced challenges in passing NVIDIA's certification due to concerns regarding overheating and power consumption. However, Samsung refuted these claims, asserting the smooth progress in the development of its most advanced HBM3e products.
In response, Jen-Hsun Huang emphasized that Samsung's products are pending certification and deployment, necessitating further engineering design work, which is currently underway. Notably, he did not address any specific issues regarding overheating or power consumption with Samsung's HBM.
Despite facing some production capacity challenges, Samsung retains its position as the leading global memory chip manufacturer. The company has initiated mass production of its 8-layer stacked HBM3e products and is poised to commence mass production of 12-layer stacked products, anticipating a significant increase in HBM supply in 2024 compared to the previous year.
Meanwhile, SK Hynix, based in South Korea, remains at the forefront of HBM3 and HBM3e chip production and serves as a key supplier for NVIDIA. SK Hynix's production capacity is fully committed until 2025, prompting a substantial investment of $14.6 billion in new production facilities to meet burgeoning demand. Samsung also intends to intensify its research and development efforts to reclaim its leading position in the memory market.