Tech giant showcases next-generation AI memory for Nvidia’s future platforms
Samsung Electronics unveiled its seventh-generation HBM4E chip at Nvidia’s GPU Technology Conference, widely known as GTC, highlighting its push to strengthen its position in the AI memory market.
Samsung on Monday announced its participation in GTC 2026 in San Jose, California, which runs Monday to Thursday, where it presented the HBM4E technology alongside a range of memory solutions designed to support Nvidia platforms.
At the event, Samsung set up an exhibition area dubbed the “HBM4 Hero Wall” to spotlight its latest high-bandwidth memory technologies. The company also revealed a physical HBM4E chip and a core-die wafer for the first time.
HBM4E is being developed based on Samsung’s 1c DRAM process and a 4-nanometer foundry base-die design. The design reflects Samsung’s integrated device manufacturer structure, combining memory, logic, foundry and advanced packaging technologies.
The chip is expected to support speeds of up to 16 gigabits per second per pin and deliver memory bandwidth of up to 4 terabytes per second.
Samsung also introduced its hybrid copper bonding technology, a next-generation packaging method designed to improve thermal efficiency and stacking capabilities.
Compared with the conventional thermal compression bonding approach, the company said the technology reduces thermal resistance by more than 20 percent and enables memory stacking of 16 layers or more.
The company also showcased memory solutions designed for Nvidia’s new Vera Rubin platform.
Inside a dedicated “Nvidia Gallery” zone, Samsung displayed HBM4 for Rubin GPUs, SOCAMM2 memory modules for the Vera CPU, and its PCIe Gen6-based server SSD PM1763, presenting the components together as part of the Vera Rubin system architecture.
SOCAMM2 is an LPDDR-based server memory module that Samsung recently completed quality validation for and has begun shipping commercially.
Samsung also demonstrated a server equipped with its PCIe Gen6-based PM1763 SSD, which will serve as a main storage device for the Vera Rubin platform. The demonstration showcased Nvidia’s SCADA workload running on the system.
Beyond the Nvidia-focused display, Samsung organized its booth around three thematic zones — AI factories for data center infrastructure, local AI for on-device computing and physical AI — where it also introduced next-generation memory products including GDDR7, LPDDR6 and the PM9E1 SSD.
On the second day of the event, Samsung AI Center head Song Yong-ho is scheduled to speak at a special session hosted by Nvidia, where he will outline the tech giant’s vision for memory technologies supporting next-generation AI systems.
yeeun@heraldcorp.com
