Samsung HBM Fails NVIDIA Certification

According to sources reported by Reuters, Samsung's latest High Bandwidth Memory (HBM) chips have encountered challenges in meeting certification standards set by industry leader NVIDIA, primarily due to issues surrounding heat dissipation and power consumption.

These challenges notably affect Samsung's HBM3 chips, widely utilized in cutting-edge Artificial Intelligence (AI) Graphics Processing Units (GPUs), and extend to the upcoming fifth-generation HBM3E chips, slated for release by the South Korean tech giant and its competitors this year.

Samsung has responded by emphasizing the customized nature of HBM products, necessitating tailored optimization processes aligned with customer requirements. The company is actively engaged in optimizing its offerings through close collaboration with clientele. However, Samsung has refrained from commenting on specific clients, while NVIDIA has also opted not to provide public statements on the matter.

The inception of HBM in 2013 introduced a revolutionary approach, employing vertically stacked DRAM chips to conserve space and reduce power consumption, thereby facilitating the processing of vast datasets inherent in complex AI applications. The burgeoning demand for GPUs, fueled by the proliferation of Generative AI, has significantly elevated the need for HBM. NVIDIA's dominance in the global AI GPU market, with a market share exceeding 80%, underscores the strategic importance for suppliers to integrate into NVIDIA's HBM supply chain.

Insights from industry insiders suggest Samsung has been diligently pursuing NVIDIA certification for its HBM3 and HBM3E products since 2023. However, recent reports reveal NVIDIA's certification failure for Samsung's 8-layer and 12-layer stacked HBM3E chips, disclosed in April.

In contrast, SK Hynix has emerged as NVIDIA's primary HBM chip supplier, commencing HBM3 supply since June 2022. The company has further initiated the supply of HBM3E to select clients as of late March, reportedly including NVIDIA among its recipients.

Meanwhile, Micron, another major player in the HBM domain, has expressed intentions to supply HBM3E to NVIDIA, indicating a diverse supplier landscape.

The resolution of these challenges by Samsung remains uncertain. Nevertheless, industry sources suggest that failure to meet NVIDIA's exacting standards has intensified apprehensions among stakeholders and investors, potentially placing Samsung at a disadvantage relative to competitors such as SK Hynix and Micron. Notably, the production capacity for SK Hynix and Micron's HBM has been fully committed, extending well into 2025.