Samsung Reportedly Dedicates Over Half of 4nm Capacity to HBM4 Base Die Production as AI Memory Demand Soars
Samsung Electronics is reportedly shifting a major portion of its advanced chip manufacturing resources toward high bandwidth memory, as demand for AI-focused memory continues to exceed available supply.
According to a report from Korean industry media, Samsung has reserved more than 50% of its 4-nanometer production capacity for HBM4 base dies. These base dies are a crucial part of next-generation high bandwidth memory, serving as the foundation beneath the stacked DRAM layers that make HBM chips suitable for AI accelerators, data center GPUs, and other high-performance computing hardware.
The report claims Samsung’s 4nm lines are currently running at full capacity, with around 50% to 60% of output being used for HBM4 base die manufacturing. That would represent a significant commitment from Samsung, especially as the company works to strengthen its position in the fast-growing AI semiconductor market.
Unlike the DRAM layers in HBM, which are manufactured using specialized memory processes, the base die is built using logic chip manufacturing technology. This makes Samsung’s foundry capabilities especially important, since the company can produce both memory and logic components in-house.
Samsung is reportedly manufacturing these chips across technologies ranging from 4nm to 7nm at its Pyeongtaek Campus 2 and Campus 3 facilities. Production is said to be taking place on the S4 and S6 foundry lines. The company’s total 4nm capacity at these facilities is estimated at around 30,000 wafers per month, with roughly 15,000 wafers allocated to HBM4 base dies.
The growing focus on base die production reflects how HBM technology has evolved. As memory bandwidth requirements rise, especially for artificial intelligence workloads, the base die has become more complex and more important. It handles critical functions that support faster data transfer, improved power efficiency, and stronger overall performance in stacked memory packages.
This shift became more noticeable with HBM3 and HBM3E, where higher data rates and faster pin speeds required more advanced logic manufacturing. HBM4 is expected to push those requirements even further, making advanced process nodes such as 4nm increasingly valuable for memory production.
Samsung is also believed to be evaluating even more advanced manufacturing options for future HBM base dies. A previous industry report suggested the company had considered using its 2nm technology for certain HBM products, potentially including customized chips for major AI hardware customers such as NVIDIA.
Competition in the HBM market remains intense. Samsung is relying on its internal manufacturing ecosystem to produce HBM base dies, while SK hynix has reportedly turned to TSMC for outsourced production. Micron, meanwhile, is believed to be using its own customized approach.
The race to secure HBM supply has become one of the most important battles in the semiconductor industry. AI servers and accelerators require massive memory bandwidth, and HBM has become essential for training and running large artificial intelligence models. As a result, chipmakers are investing heavily to expand output, improve yields, and prepare for the next generation of AI hardware.
If the latest report is accurate, Samsung’s decision to allocate more than half of its 4nm capacity to HBM4 base dies shows just how central high bandwidth memory has become to the company’s strategy. With HBM4 production expected to expand further in 2026, Samsung appears to be positioning itself for a larger role in the AI memory supply chain.
