According to reports,Due to technical problems, Samsung HBM memory products have not yet entered the NVIDIA supply chain.Samsung originally planned to supply NVIDIA's 12-layer HBM3E products in the second quarter of 2025, but delivery has now been delayed. Competitors SK Hynix and Micron have already supplied NVIDIA first, which has increased Samsung's competitive pressure in the HBM market.

Samsung is expected to start supplying HBM3E products to Broadcom in the second half of 2025, and is expected to become the largest supplier in Broadcom's HBM3E supply chain, accounting for more than 50%.And because the progress of supplying NVIDIA is blocked, Samsung's supply to Broadcom may be advanced.
The report pointed out that Samsung HBM faced three major technical problems in supplying NVIDIA:
The first is that it fails to meet NVIDIA's stringent heat dissipation standards. NVIDIA's heat dissipation requirements are twice those of Broadcom because its products are general-purpose products that need to demonstrate high performance in various environments. If HBM generates too much heat, it will directly affect chip performance.
Secondly, Samsung's HBM is not as good as SK Hynix and Micron in data transmission speed and accuracy under extreme conditions such as overheating. NVIDIA's GPU requires fast, bottleneck-free data transmission to support, and the DRAM equipped with Samsung HBM has performance issues that affect the accuracy and speed of data transmission.
Finally, the yield rate of Samsung HBM is low, which not only makes it difficult to deliver on time, but also weakens competitiveness in price negotiations. Samsung is redesigning DRAM to improve quality and stabilize yield.
Industry stakeholders pointed out that although Samsung entered the supply chain later than SK Hynix and Micron due to these three major problems, it has recently overcome most of the technical difficulties.However, since NVIDIA has allocated part of its supply to SK Hynix and Micron, supply negotiations between Samsung and NVIDIA may still take time.