Micron Plans to Double HBM Capacity by End of 2026, Targeting NVIDIA's Vera Rubin Platform

nashnova research
今天发布阅读约 7 分钟

Micron is preparing to roughly double its HBM wafer output to about 100,000 wafers per month by late 2026, ramping 12-layer HBM4 to supply Nvidia's next-generation Vera Rubin AI platform — a direct capacity challenge to Samsung and SK Hynix.

01

What exactly is Micron planning?

According to Digitimes, Micron aims to lift monthly HBM — high-bandwidth memory, the ultra-fast memory built specifically for AI chips — capacity to roughly 100,000 wafers by end of 2026, nearly double the current level.
The centrepiece is 12-layer stacked HBM4, the latest generation of HBM that vertically stacks 12 memory dies for a major leap in bandwidth and capacity.
This means → Micron is not simply adding volume on existing products; it is betting on the next-generation product to seize an early position.
02

Why target Nvidia's Vera Rubin?

The expansion is explicitly aimed at Nvidia's upcoming Vera Rubin AI platform — its next flagship AI chip platform, which will demand large volumes of HBM4.
In plain terms = Nvidia is the world's biggest AI-chip buyer. Winning a spot in the Vera Rubin supply chain is equivalent to securing the most valuable order book in the HBM market.
This reflects the real nature of the HBM race: it is not just about total capacity — it is about qualifying for the next platform's supply list.
03

How far ahead are Samsung and SK Hynix?

The report notes that Samsung and SK Hynix still hold a lead in overall HBM scale.
If Micron delivers on schedule, it will apply more direct competitive pressure on HBM4 specifically, intensifying the three-way share battle within Nvidia's supply chain.
This means → Micron's playbook is "leapfrog on the generational shift" — the gap on older products is hard to close, so it aims to reshuffle the deck on HBM4.
04

Can Micron actually deliver?

Because the original report sits behind a paywall, detailed capacity allocation and customer-mix data have not been fully disclosed.
Whether Micron can hit the capacity target by late 2026 — and how fast HBM4 mass-production yield ramps — are the key milestones for judging if the scale gap narrows.
Put simply = the plan is ambitious, but between "announced roadmap" and "100k wafers a month" lie equipment installation, yield ramp, and customer qualification — each a hard gate to clear.

市场有风险,内容仅供研究参考,不构成投资建议。

Micron Plans to Double HBM Capacity by End of 2026, Targeting NVIDIA's Vera Rubin Platform · nashnova