Samsung HBM4 AI memory chip sales top $1 billion by June 23, 2026

The milestone came about four months after mass production began, as Samsung expanded HBM4 supply talks with AMD and discussed next-generation memory with NVIDIA.

Summary

Samsung Electronics said sales of its next-generation high-bandwidth memory, or HBM (stacked memory for AI chips), surpassed $1 billion by June 23, 2026, roughly four months after mass production started. Its HBM4 chips began shipping on February 12, 2026, with the company claiming performance gains of more than 40% over previous standards. The rollout quickly drew commercial backing: by March 18, Samsung had signed a memorandum of understanding with AMD to supply HBM4 for the Instinct MI455X GPU and EPYC processors, targeting bandwidth of up to 3.3TB/s using a 10nm-class process. Samsung also held discussions with NVIDIA at GTC 2026 on HBM4E integration and future HBM5 architecture. By May 2026, the company started shipping samples of its 12-layer HBM4E, which it described as an industry first, with speeds of up to 16Gbps per pin aimed at improving power efficiency for AI workloads. The rapid rise to $1 billion in sales underscores strong demand across the AI hardware supply chain as Samsung tries to win back share from rivals including SK Hynix and Micron in a memory segment that has become strategically central to AI infrastructure.

Terms & Concepts
  • HBM: High-bandwidth memory used to feed AI processors faster
  • GPU: Graphics processing unit, a chip used for AI computing
  • HBM4E: An advanced HBM variant with higher speed and efficiency