According to the latest report from Hong Kong securities firms, Samsung Electronics failed to pass the third Nvidia 12 layer HBM3E chip certification in June 2025. This tech giant is currently planning its fourth certification in September.
Samsung's latest certification work failed to meet Nvidia's standards, bringing further uncertainty to its timeline for entering the next wave of AI workload HBM supply. Although Samsung has increased the production of HBM3E ahead of schedule, its supply plan has been postponed due to the failure to obtain certification.
At the same time, Micron Technology seems to be making new progress. Micron Corporation utilizes the thermal bonding (TCB) equipment from Hanmei Semiconductor to improve the yield of 8-layer and 12 layer HBM3E chips. The yield of the 12 layer HBM3E chip has reached 70%, while the yield of the 8-layer HBM3E chip has reached 75%.
UBS Group recently reported that the certification of Samsung's 12th floor HBM3E is still "pending" and predicted that the company's supply to Nvidia may be delayed until the fourth quarter of 2025. Due to the need for faster and more efficient memory solutions from AI chip manufacturers such as NVIDIA, this setback may change the fiercely competitive HBM market landscape.
Micron recently announced that it has delivered its sixth generation HBM4 samples to major customers for use in AI semiconductors. This makes Micron the second DRAM manufacturer to deliver HBM4 samples after SK Hynix, which began delivering HBM4 samples in March 2025.
At the same time, due to the delay in mass production of customized AI chips by major technology companies, UBS has adjusted its market demand expectations for HBM. The company currently predicts that global HBM demand will reach 163 billion GB by 2025, lower than the previously estimated 189 billion GB, and expects HBM demand to reach 254 billion GB by 2026, slightly lower than the previously predicted 261 billion GB.
