Particle.news
Download on the App Store

Samsung Named Preferred HBM4 Supplier for AMD’s Next AI GPU as Foundry Talks Open

The agreement underscores supply security for AMD’s roadmap, opening the door to Samsung foundry services.

Overview

  • Samsung and AMD signed an MoU that makes Samsung the preferred HBM4 supplier for AMD’s Instinct MI455X accelerators and adds collaboration on high‑performance DDR5 for next‑generation EPYC servers.
  • The companies will explore a potential foundry partnership for future AMD products, with discussions ongoing and no manufacturing agreement finalized.
  • Samsung’s HBM4, already in mass production, is built on a 1c DRAM node with a 4 nm base die and targets up to 13 Gbps per pin and 3.3 TB/s bandwidth, while Samsung continues supplying HBM3E for AMD’s MI350‑series GPUs.
  • At Nvidia GTC, Samsung publicly showed HBM4E for the first time with stated targets of 16 Gbps per pin and 4.0 TB/s bandwidth and displayed complementary components for Nvidia’s Vera Rubin platform in a dedicated showcase.
  • Nvidia CEO Jensen Huang confirmed Samsung is manufacturing the Groq 3 LPU used in Nvidia‑aligned inference systems, as Samsung executives push multi‑year memory supply deals to stabilize the business.