TLDR
- Samsung and SK Hynix receive exclusive supplier status for HBM4 memory in Nvidia’s Vera Rubin AI accelerator
- Micron loses its position in the Vera Rubin supply chain, resulting in a 6.74% decline in MU shares
- Samsung successfully completed Nvidia’s HBM4 quality validation at both 10 Gbps and 11 Gbps performance levels
- SK Hynix continues working toward the 11 Gbps benchmark while maintaining its position as the leading HBM provider
- Manufacturing operations will commence in March, with Vera Rubin hardware scheduled for second-half 2026 release
According to the Korea Economic Daily, Nvidia has designated Samsung and SK Hynix as the sole providers of sixth-generation high-bandwidth memory (HBM4) for the Vera Rubin AI accelerator. Micron, previously a significant HBM partner, will sit out this flagship chip generation.
The announcement triggered a 6.74% drop in Micron shares. Samsung’s Korean-traded stock declined 7.81%, while SK Hynix fell 9.52%. Nvidia shares retreated 3.01%.
Vera Rubin represents Nvidia’s upcoming flagship AI platform, building upon the Blackwell architecture. The complete NVL72 rack setup combines 72 Rubin GPUs with 36 Vera CPUs and achieves 10x superior performance per watt compared to Blackwell.
Micron retains a position in the broader product lineup. The company will provide HBM4 for Rubin CPX, a mid-range inference-oriented accelerator within the Rubin family. The top-tier Vera Rubin product, however, excludes Micron from its supply roster.
Samsung earned its approval by meeting Nvidia’s internal quality standards at both 10 Gbps and 11 Gbps performance thresholds. SK Hynix continues efforts to achieve the 11 Gbps qualification while maintaining its leadership position in overall HBM supply.
SK Hynix Maintains Dominance While Samsung Expands
SK Hynix is anticipated to control approximately 50% of worldwide HBM production in 2026, a modest decrease from 59% in 2025. Samsung’s market share is predicted to reach 28%, advancing from 20% in the previous year.
SK Hynix will deliver more than half of Nvidia’s complete HBM requirements — encompassing HBM3E — throughout 2026, and will command the largest portion of Vera Rubin HBM4 volume.
Both manufacturers plan to initiate HBM4 production this month. Vera Rubin remains scheduled for a second-half 2026 market entry.
Vera Rubin’s Target Applications
The Vera Rubin platform targets large-scale AI training and inference workloads, particularly the mixture-of-experts (MoE) architectures gaining traction in cutting-edge AI development.
Reported prospective buyers include Microsoft, Amazon, Oracle, and Google. These cloud hyperscalers have represented Nvidia’s primary customer base in recent cycles.
The Vera Rubin NVL72 requires twice the power consumption of Blackwell while delivering considerably improved output per watt, a critical factor for data center operators managing these systems at enterprise scale.
HBM4 provides enhanced memory bandwidth compared to earlier generations, addressing one of the primary constraints in training and operating large AI models.
Wall Street analysts maintain bullish views on Nvidia. TipRanks data shows NVDA holds a Strong Buy rating from 39 analysts, with one Hold recommendation. The consensus price target of $272.16 suggests approximately 53% upside potential from present levels.
Nvidia stock has climbed 66.2% over the trailing 12 months, despite Monday’s retreat following the supplier announcement.
Samsung and SK Hynix will launch HBM4 manufacturing in March 2026, with Vera Rubin systems expected to reach customers during the year’s second half.
