Last updated: March 18th, 2026 at 09:59 UTC+01:00


AMD will use Samsung's HBM4 chips in its next-generation AI GPUs

Samsung started the mass production of HBM4 chips last month, and has already supplied them to Nvidia.

Asif Iqbal Shaik

Reading time: 2 minutes

amd ceo lisa su sign memorandum of agreement samsung hbm4 chips ai gpus

Samsung

Business

AMD CEO Lisa Su and Samsung Device Solutions head Jeon Young-hyun - Source: Samsung

AMD and Samsung have signed a memorandum of understanding (MoU) that includes the use of Samsung’s sixth-generation high-bandwidth memory (HBM4) chips in AMD’s next-generation AI GPUs. The agreement was signed during AMD CEO Lisa Su’s visit to Samsung Electronics’ semiconductor plant in Pyeongtaek, South Korea.

During her first business trip to South Korea since becoming AMD’s CEO, Lisa Su attended the signing ceremony at Samsung’s facility. As part of the agreement, Samsung has been selected as the preferred supplier of HBM4 chips for AMD’s Instinct MI455X GPU, which was announced earlier this year and is expected to begin shipping in the second half of 2026. It rivals Nvidia's latest AI platform, Vera Rubin, which also uses Samsung's HBM4 chips.

Samsung’s latest-generation HBM4 memory offers data transfer speeds of up to 13Gbps per pin and bandwidth of up to 3.3TB/s. The chips are built using a 10nm-class 1c DRAM process along with a 4nm base die. Samsung first announced the technology last month and has already started shipments.

Lisa Su said, “Close collaboration across the industry is essential for realizing next-generation AI infrastructure. We are very pleased to combine Samsung’s leadership in advanced memory technology with AMD’s Instinct GPU, EPYC CPU, and rack-scale platform.

The AMD Instinct MI455X is a general-purpose GPU designed for AI workloads such as training and inference. It delivers up to 40 PFLOPS of FP4 and 20 PFLOPS of FP8 compute performance, making it nearly twice as fast as its predecessor, the MI350.

AMD and Samsung also announced plans to expand their partnership. This includes the use of high-performance DDR5 memory in AMD’s Helios AI data center rack platform and sixth-generation EPYC server CPUs. Samsung Foundry is also in discussions to manufacture AMD’s future products.

Samsung also confirmed that it has been supplying HBM3E chips for AMD’s existing MI350X and MI355 GPUs. The two companies have collaborated for nearly 20 years. Samsung has supplied VRAM for AMD GPUs in the past, while AMD has helped Samsung develop RDNA-based GPUs for Exynos mobile chips.