Samsung sample shipments for HBM4E have arrived right on schedule. The chipmaker promised to deliver the upgraded memory to customers in early 2026 after it began shipping HBM4 back in February, and that moment is now here. The new variant keeps the same 12-layer design but packs more capacity, higher bandwidth, and cooler operation.
The current HBM4E configuration from Samsung carries 48GB across its 12 layers, up from 36GB on HBM4. The company is also working on alternative configurations, including a 32GB version with eight layers and a 64GB variant with sixteen layers, to give hardware designers more flexibility in their builds.
Bandwidth has climbed roughly 20% alongside the 33% capacity jump. Per-pin speed now reaches 14Gbps, which translates to 3.6 terabytes per second per stack. That edges past HBM4, which ran at 11.7Gbps per pin for the same 3.6 terabytes per second total.
A stack of Samsung HBM4E memory
The memory relies on sixth-generation “1c” process technology in the 10nm class for its DRAM dies, paired with a 4nm logic base die. Samsung reworked the architecture to cut power consumption by 16%, meaning less electricity drawn and less heat thrown off. Thermal resistance dropped by at least 14% as well, so the modules shed heat more easily under load.
Samsung’s HBM4 production volume is still climbing, and the company expects HBM4E to push AI system performance even further. Headroom remains in the design; the 14Gbps transfer rate could eventually scale to 16Gbps. That reserve matters for data centers running large language models, where memory bandwidth often becomes the bottleneck before compute power maxes out.
The 48GB stack density and 3.6TB/s throughput put Samsung in direct competition with SK hynix and Micron, both of which are racing toward their own HBM4E sample windows. Samsung’s decision to sample now, rather than wait for a full product generation cycle, suggests the company wants to secure design wins ahead of next-generation AI accelerator launches from NVIDIA and AMD. Memory makers typically need twelve to eighteen months between sampling and volume production to validate chips with customers.
“Following the successful mass production of HBM4, Samsung has once again demonstrated its distinct technological edge with HBM4E. Through our advanced manufacturing capabilities and preemptive infrastructure investments, we will continue to drive the growth of the global AI memory market,”
Sang Joon Hwang, Executive Vice President and Head of Memory Development at Samsung Electronics, said in a statement.
The South Korean firm’s semiconductor division has been pushing aggressively across multiple fronts. While memory technology advances, Samsung Display’s OLED Panels Set for MacBook Pro Production is now underway. The company has also confirmed that Samsung OLED displays with holes and a stacked design are headed for the Ferrari Luce.
Industry observers note that Samsung Exynos 2700 Chip for Galaxy S27 Set to Drop WLP Tech reflects the same strategy of tailoring component designs for specific thermal and packaging constraints.
Source: Samsung Newsroom
