Samsung Electronics pivots to memory-centric computing for AI era

Samsung Electronics is shifting its strategic focus toward ‘memory-centric computing’ to address the evolving demands of the artificial intelligence era. Speaking at ‘The 65 Summit 2026,’ Sang-yeon Cho, head of Samsung’s Device Solutions Americas, stated that the industry must prioritize memory bandwidth over simple data capacity. He noted that bandwidth determines critical AI performance metrics, including tokens per second, response speeds, and overall operational costs.

To implement this vision, Samsung is developing zHBM, a next-generation architecture that stacks High Bandwidth Memory (HBM) directly on top of AI accelerators to minimize data movement distances. While the company is currently defining specifications for the first generation, commercial products based on zHBM are expected to emerge after 2029. To manage the heat generated by this vertical integration, Samsung plans to utilize advanced technologies such as Hybrid Copper Bonding (HCB).

Furthermore, Samsung emphasized the necessity of a ‘turnkey solution’ that integrates memory, foundry, and advanced packaging services. The company is currently managing the entire process for HBM4, from the 4nm base die design to final assembly and packaging, as traditional process miniaturization alone is insufficient to meet AI power efficiency and performance requirements.