TLDR
- Nvidia has chosen Samsung and SK Hynix as the only suppliers of HBM4 memory for its next-gen AI chip, Vera Rubin
- Micron has been cut out of the Vera Rubin supply chain, sending MU stock down 6.74%
- Samsung passed Nvidia’s HBM4 quality tests at 10 Gbps and 11 Gbps; SK Hynix is still working on the 11 Gbps threshold
- SK Hynix is expected to supply over half of Nvidia’s total HBM in 2026; Samsung’s share rises to 28%
- Production is expected to begin in March, with Vera Rubin launching in the second half of 2026
Nvidia has picked Samsung and SK Hynix as the exclusive suppliers of sixth-generation high-bandwidth memory (HBM4) for its upcoming Vera Rubin AI accelerator, according to the Korea Economic Daily. Micron, which had been a key HBM supplier, is not part of this cycle for the flagship chip.
The report sent Micron stock down 6.74% on the day. Samsung’s Korean-listed stock fell 7.81%, and SK Hynix dropped 9.52%. Nvidia itself slid 3.01%.
Vera Rubin is Nvidia’s next flagship AI platform, succeeding the Blackwell architecture. The full NVL72 rack configuration pairs 72 Rubin GPUs with 36 Vera CPUs and delivers 10x better performance per watt than Blackwell.
Micron is not being shut out entirely. It will supply HBM4 for Rubin CPX, a mid-tier inference-focused accelerator in the Rubin lineup. But it won’t be part of the top-tier Vera Rubin product.
Samsung was approved after passing Nvidia’s internal quality benchmarks at both 10 Gbps and 11 Gbps speeds. SK Hynix is still working to clear the 11 Gbps bar but remains the dominant player in HBM supply overall.
SK Hynix Leads, Samsung Gains Ground
SK Hynix is projected to hold around 50% of global HBM output in 2026, down slightly from 59% in 2025. Samsung is expected to grow its share to 28%, up from 20% last year.
SK Hynix is forecast to supply more than half of Nvidia’s total HBM — including HBM3E — across 2026, and is expected to lead on Vera Rubin HBM4 volume.
Both companies are expected to begin HBM4 production this month. Vera Rubin is on track for a second-half 2026 launch.
What Vera Rubin Is Built For
The Vera Rubin platform is designed for large-scale AI training and inference, including the kind of mixture-of-experts (MoE) models that are becoming more common in frontier AI development.
Potential customers named in reports include Microsoft, Amazon, Oracle, and Google. These are the same cloud hyperscalers that have been Nvidia’s biggest buyers in recent years.
The Vera Rubin NVL72 draws double the power of Blackwell but delivers substantially better output per watt, which matters for data center operators running these systems at scale.
HBM4 offers higher memory bandwidth than previous generations, which is one of the key bottlenecks in training and running large AI models.
Analyst sentiment on Nvidia remains positive. According to TipRanks, NVDA carries a Strong Buy rating from 39 analysts, with one Hold. The average price target of $272.16 implies around 53% upside from current levels.
Over the past 12 months, Nvidia stock has gained 66.2%, even as Monday’s session saw it pull back on the supplier news.
Samsung and SK Hynix are set to begin HBM4 production in March 2026, with Vera Rubin hardware expected to ship in the second half of the year.





