Micron to Double HBM Capacity for Nvidia’s Vera Rubin Chips

High-bandwidth memory, or HBM, vertically stacks multiple layers of DRAM and directly affects AI accelerator data-transfer speeds and power consumption. Micron Technology’s entry into Nvidia’s Vera Rubin platform puts it in competition with SK hynix and Samsung Electronics for a high-value segment of the AI data-center supply chain.

Micron said on March 16, 2026, that it began mass-producing and shipping 36GB, 12-layer HBM4 in the first quarter, with most of its 2026 capacity sold out under pricing and volume agreements. On March 18, the company reported fiscal second-quarter revenue of $23.86 billion, up 196% from a year earlier, and net income of $13.785 billion. It also raised its full-year capital spending forecast to more than $25 billion.

High-bandwidth memory, or HBM, is a key component for improving data-transfer efficiency in AI accelerators. As the computing demands of large models rise, HBM4 and subsequent products have become a major competitive focus for memory suppliers. Micron’s expansion of advanced packaging and memory capacity will directly affect AI chip supply and the industry’s competitive landscape.

Micron said its HBM4 capacity expansion remains on schedule, with production ramp-up and yield improvements both progressing better than for previous-generation products. The company expects to begin mass-producing HBM4E in 2027 and, for the first time, will have TSMC manufacture the base dies using its process technology. The companies have not disclosed production capacity, process nodes or investment amounts for the partnership.