AMD expands Samsung HBM4 deal for next-generation accelerators

AMD has secured Samsung HBM4 supply for its next-generation AMD Instinct MI455X graphics processing units, while the agreement also points to broader memory collaboration around future server chips. The arrangement suggests Samsung gained leverage as demand for advanced memory remains tight.

AMD has secured HBM4 capacity from Samsung for its next-generation AMD Instinct MI455X Artificial Intelligence accelerators after Lisa Su traveled to South Korea to finalize supply. Samsung said the companies signed an MOU covering primary HBM4 supply for the AMD Instinct MI455X graphics processing unit and advanced dynamic random-access memory solutions tied to AMD’s upcoming server platform plans.

Samsung said, “Under the MOU, Samsung and AMD will align on primary HBM4 supply for the next-generation AMD AI accelerator, the AMD Instinct MI455X GPU, as well as advanced DRAM solutions for 6th Gen AMD EPYC CPUs, codenamed ‘Venice.” These technologies will support next-generation AI systems that combine AMD Instinct GPUs, AMD EPYC CPUs, and rack-scale architectures such as the AMD Helios platform.

The agreement appears broader than a straightforward memory procurement. Samsung’s announcement links HBM4 supply to AMD’s next Artificial Intelligence graphics processing unit while also adding memory work for server chips, indicating a deeper commercial relationship than a standard component order. At the same time, AMD is still behind Nvidia in Samsung’s HBM4 allocation line, a difficult position as supply remains constrained.

Samsung has recently built momentum in high bandwidth memory by pushing HBM3 and HBM3E into the Artificial Intelligence market. Its HBM4 is being marketed with pin speeds up to 13 Gbps, helping make it one of the most sought-after memory products for upcoming accelerator platforms. Nvidia secured Samsung HBM4 capacity early for Vera Rubin, leaving AMD negotiating from a weaker position.

Reports also suggest Samsung may have pressed for more than memory business alone. Chosun Biz reported that Samsung has maintained conditions for producing some of AMD’s advanced Artificial Intelligence chips at Samsung Foundry, though the exact scope remains unclear. The report leans toward foundry work for accelerators rather than the EPYC server central processing unit line, despite earlier rumors around EPYC Venice and SF2. If expanded, that would place more AMD silicon closer to Samsung’s manufacturing roadmap and strengthen Samsung Foundry’s customer list alongside NVIDIA, Tesla and Apple.

55

Impact Score

OpenAI acquires Astral to strengthen coding workflows

OpenAI is acquiring Astral, the developer of open source Python tools including uv, Ruff and ty, to integrate them with Codex. The move signals a push to make Artificial Intelligence coding systems more reliable across the full software development workflow.

Artificial Intelligence tool targets forged radiology reports

University at Buffalo researchers developed a detection system aimed at identifying radiology reports generated by Artificial Intelligence rather than clinicians. The work targets a growing risk of fraud in health care, insurance, and other record-driven industries.

NSF funds teacher training to expand Artificial Intelligence education nationwide

The U.S. National Science Foundation is awarding 11 million to the Computer Science Teachers Association to train K-12 educators in computer science and Artificial Intelligence instruction. The multistate initiative is designed to scale classroom-ready teaching capacity and broaden high-quality learning opportunities for students across the country.

Contact Us

Got questions? Use the form to contact us.

Contact Form

Clicking next sends a verification code to your email. After verifying, you can enter your message.