Samsung unveils HBM4E, highlights AI partnership with Nvidia at GTC

Samsung Electronics' HBM4 / Courtesy of Samsung Electronics

Samsung Electronics unveiled its next-generation high-bandwidth memory HBM4E and a range of artificial intelligence (AI) computing solutions at Nvidia GTC 2026, underscoring its growing collaboration with Nvidia in the rapidly expanding AI infrastructure market.

The Korean chipmaker said Tuesday it is showcasing its latest semiconductor technologies at the conference in San Jose, California, which is being held from Monday through Thursday (local time). The lineup spans memory, logic, foundry and advanced packaging, targeting next-generation data centers and AI systems.

HBM4E, the successor to HBM4, is being introduced for the first time at the event. The chip is designed to provide speeds of up to 16 gigabits per second (Gbps) per pin and bandwidth of 4 terabytes per second, targeting next-generation AI data centers that require significantly higher memory performance.

Samsung is also presenting its sixth-generation HBM4, which has entered mass production and is designed for Nvidia’s upcoming Vera Rubin platform.

Samsung said the chip delivers processing speeds of 11.7 Gbps, exceeding the industry’s typical 8 Gbps standard, and can be boosted to 13 Gbps. The company developed the chip using its sixth-generation 10-nanometer-class DRAM process.

Nvidia CEO Jensen Huang, second from right, poses with Samsung executives at the Nvidia GTC global artificial intelligence conference in San Jose, Calif., Monday. Reuters-Yonhap

The deepening cooperation between Samsung and Nvidia was highlighted by Nvidia CEO Jensen Huang, who said Samsung’s foundry division is currently manufacturing the Groq 3 language processing unit (LPU), a chip designed for AI inference.

"I want to thank Samsung, who manufactures the Groq 3 LPU chip for us, and they are cranking as hard as they can. I really appreciate you guys," Huang said during his keynote speech, adding that the chips would be shipped in the second half of this year.

His remarks signal broader cooperation between the two companies in the AI sector, extending beyond memory to include Samsung’s foundry, or contract chip manufacturing, business.

Last month, Samsung Electronics began its first commercial shipments of HBM4 designed for Nvidia’s Vera Rubin platform, which the company said delivers “ultimate performance” for AI computing.

At a dedicated Nvidia gallery, Samsung is showcasing key components for AI infrastructure, including HBM4 memory, the SOCAMM2 server module and the PM1763 SSD.

The SOCAMM2, based on low-power DRAM, has entered mass production and is designed for high bandwidth and flexible integration in AI servers. The PM1763 SSD targets AI storage systems requiring faster data transfers and larger capacities, with performance demonstrated on servers using Nvidia’s SCADA programming model.

Samsung is also highlighting its collaboration with Nvidia on AI-driven manufacturing, or “AI factories,” using accelerated computing and Omniverse to expand digital twin systems in semiconductor production.

The company is further presenting on-device AI memory products, including PM9E3 and PM9E1 NAND for DGX Spark systems, along with LPDDR5X and LPDDR6 mobile DRAM. LPDDR5X offers speeds up to 25 Gbps per pin with improved efficiency, while LPDDR6 targets up to 35 Gbps with enhanced power management.