- Published on
Samsung to Devote Half of 4nm Capacity to HBM4 Base Chip Production
- Authors
- Name
- DeepHorus
Samsung Electronics is increasing its investment in HBM4, the next-generation high-bandwidth memory. According to South Korean media reports, its foundry division has allocated about half of its 4-nanometer capacity to the production of HBM4 base dies to meet the demand for high-performance memory in the AI market.
HBM4 is the latest generation of high-bandwidth memory technology and a key component for future high-performance AI accelerators. Compared with previous products, the importance of the HBM4 base die has significantly increased. This chip is located beneath the multi-layer stacked DRAM dies, responsible for data transfer between memory and compute cores, and can integrate more logic circuit functions.
Currently, Samsung uses its own 4nm SF4 process to produce HBM4 base dies, with production lines at S5 and S6 in the Pyeongtaek Phase 2 and Phase 3 campuses. According to industry insiders, Samsung's 4nm production lines are running near full capacity, with approximately 50% to 60% of capacity used for HBM4 base die manufacturing, and this proportion is expected to remain high in the coming months.
An important reason driving this change is the rapidly increasing demand for customized HBM solutions. More and more customers want to integrate proprietary logic modules, such as memory controllers and PHY interface circuits, into the base die. This can transfer some circuits from the main compute chip to the HBM base die, reducing the area occupied by the compute chip and improving design efficiency.
This trend is particularly evident in the ASIC market. Cloud service companies such as Google, Microsoft, and Amazon continue to develop their own AI accelerators, and demand for customized HBM4 products is growing. Samsung, which simultaneously owns memory, foundry, and advanced packaging businesses, has a unique advantage in providing customized HBM solutions.
Market share changes also reflect Samsung's rapid catch-up. Data shows that in the second quarter of 2026, Samsung's HBM market share reached 38%, compared with about 15% a year earlier; during the same period, SK Hynix's share fell from 64% to 50%. More and more customers are adopting a multi-supplier strategy to reduce supply chain risks.
Samsung has already provided HBM4 samples to customers and plans to fully expand shipments in the second half of 2026. The company expects HBM4 revenue to grow more than threefold quarter-on-quarter in the third quarter, and to account for more than 60% of total HBM revenue in the second half. Analysts believe that with NVIDIA's Rubin platform and more custom AI chips entering mass production, HBM4 will become one of the fastest-growing areas in the semiconductor industry, and Samsung is expected to further narrow the gap with SK Hynix.
