MT

Micron HBM4

Hardware

by Micron Technology

Fourth-generation high-bandwidth memory offering more than 2.8 TB/s per stack, in high-volume production at 36GB (12-high) with 48GB 16-high samples shipping for next-generation AI accelerator platforms.

Find alternatives to Micron HBM4

No public price on the product page; purchase routes through Micron's sales network and distributors.

What it does

• Memory

How it compares

• Samsung HBM4

  Both sell high-bandwidth memory into AI accelerators, the segment both product records cover; Micron's 10-K names Samsung first among its memory rivals, though the filing claims the rivalry for memory broadly rather than HBM alone.

  Samsung Electronics Co., Ltd.; SK hynix Inc.; Kioxia Holdings Corporation; Sandisk Corporation; ChangXin Memory Technologies, Inc. (CXMT); and Yangtze Memory Technologies Co., Ltd. (YMTC)

  Regulatory filing · 25 Aug 2026
• HBM4

  (same competitor list sentence)

  Regulatory filing · 18 Sep 2026

Sources

Pricing

Official documentation · 2 Sep 2026

No pricing shown; sales-network route only.

Description

Official documentation · 2 Sep 2026

Sold within

Not found · 30 Sep 2026

Micron's product page states no price for HBM4; the only routes are the site-wide 'Sales support contact' and 'Authorized distributors' nav links, neither naming this product.

Also from Micron Technology

• Micron 6600 ION NVMe SSD Hardware

  A high-capacity QLC NVMe SSD of up to 245TB, marketed as purpose-built for AI data lakes and large-scale training-data ingestion.

• Micron 9550 NVMe SSD Hardware

  A PCIe Gen5 data-centre NVMe SSD marketed by Micron as its AI-optimised drive, with Big Accelerator Memory and NVIDIA GPUDirect Storage support for training and inference pipelines.

• Micron 9650 NVMe SSD Hardware

  A PCIe Gen6 data-centre NVMe SSD, in mass production, marketed for feeding GPU-heavy AI training and inference deployments.

• Micron HBM3E Hardware

  Third-generation high-bandwidth memory delivering more than 1.2 TB/s of bandwidth per stack for AI accelerators, deep learning and HPC systems.

• Micron SOCAMM Hardware

  A CAMM-based low-power LPDDR5X memory module for AI data centre servers, up to 256GB, built to cut LLM inference latency by offloading KV-cache from HBM.

• Micron GDDR7 Hardware

  Latest-generation GDDR graphics memory (up to 32 Gb/s pin speed, 1.5+ TB/s system bandwidth) supporting GPUs across gaming, AI inference and high-performance-computing workloads; Micron's own comparison table names 'AI Inference Accelerator' as a distinct workload column alongside gaming.

• Micron SOCAMM2 Hardware

  Second-generation SOCAMM — a data-center-class modular LPDDR5X form factor delivering 50% more capacity in the same footprint, designed for NVIDIA Vera Rubin NVL72 and Vera CPU platforms as a flagship AI data-center memory solution.

Open Micron Technology in the directory

Something wrong here? Send a correction — quote this product id: micron-hbm4.