Samsung Electronics outlined the full range of AI computing technologies it will present at Nvidia GTC 2026 in San Jose, California, March 16-19, highlighting memory, logic, foundry, and advanced packaging products aimed at supporting AI infrastructure, on-device intelligence, and manufacturing digitization.
The company said the centerpiece of its exhibit is the sixth-generation HBM4, which is in mass production and designed for the Nvidia Vera Rubin platform. Samsung said HBM4 delivers consistent processing speeds of 11.7 gigabits-per-second (Gbps), exceeding an industry-standard reference cited in the company's materials, and can be enhanced to 13Gbps. Samsung also displayed its successor, HBM4E, for the first time at GTC 2026; the firm said HBM4E delivers 16Gbps per pin and 4.0 terabytes-per-second bandwidth.
Samsung noted it achieved these results by using a sixth-generation 10nm-class DRAM process (1c), which the company said enabled stable yields and industry-leading performance. The company also highlighted hybrid copper bonding technology, which it said allows next-generation HBM to reach 16 or more layers while reducing heat resistance by more than 20% compared with thermal compression bonding.
A dedicated Nvidia Gallery at the Samsung booth will feature items the company said are designed for Nvidia AI infrastructure, including HBM4, SOCAMM2, and the PM1763 solid-state drive. Samsung described SOCAMM2, based on low-power DRAM, as an optimum server memory module offering high bandwidth and flexible system integration and stated it is currently in mass production. Samsung said PM1763 uses the PCIe 6.0 interface to deliver fast data transfers and high capacities and will be demonstrated on servers running the Nvidia SCADA programming model.
Samsung also said its PM1753 SSD will be shown as part of the new Nvidia BlueField-4 STX reference architecture for accelerated storage infrastructure on the Vera Rubin platform, with a focus on energy efficiency and system performance for inference workloads.
The company plans to showcase collaboration with Nvidia on implementing accelerated computing in Samsung's AI Factory and fast-tracking digital twin manufacturing that leverages Nvidia Omniverse libraries. Song Yong-Ho, executive vice president and head of Samsung's AI Center, is scheduled to present a session on March 17 titled "Transforming Semiconductor Manufacturing with Agentic AI from Design and Engineering to Production," which Samsung said will detail the firm's strategic collaboration and real-life use cases.
Samsung will also exhibit memory solutions for local AI, including PM9E3 and PM9E1 NAND for Nvidia DGX Spark, and DRAM products LPDDR5X and LPDDR6, which the company said offer increased speeds and power-management features for mobile and edge-AI workloads.
Article edited by Jack Wu
