NVIDIA "Vera Rubin" AI Systems Set for Launch with Next-Gen Memory Solutions
NVIDIA is preparing to launch its highly anticipated "Vera Rubin" AI systems, with shipments of the VR200 NVL72 rack-scale solutions expected in late summer. These advanced systems are designed to power the next generation of artificial intelligence models, offering unprecedented performance and scalability for data centers and enterprise AI workloads.
HBM4 Memory Supply: Samsung and SK hynix Take the Lead
The memory configuration for the VR200 NVL72 systems has become a focal point in the industry, particularly regarding the supply of HBM4 (High Bandwidth Memory 4). According to detailed supply chain analysis from SemiAnalysis, only Samsung and SK hynix have secured design wins to provide HBM4 for these systems. SK hynix is expected to supply approximately 70% of the HBM4 memory, while Samsung will cover the remaining 30%. Notably, Micron, a major player in the memory market, has reportedly not received any commitment for HBM4 supply in this generation.
Micron’s Strategic Role with LPDDR5X Memory
Despite missing out on HBM4, Micron remains a key contributor to the NVIDIA VR200 NVL72 ecosystem. The company will supply LPDDR5X memory for the "Vera" CPUs, which can be configured with up to 1.5 TB of LPDDR5X per system. This allocation helps offset Micron’s absence from the HBM4 supply chain and highlights the company’s continued relevance in high-performance computing.
The shift in memory suppliers may be linked to NVIDIA’s significant system upgrades for the VR200 NVL72. The system’s bandwidth target increased from 13 TB/s in March 2025 to 20.5 TB/s by September, and at CES 2026, NVIDIA confirmed the system now achieves 22 TB/s of bandwidth. This nearly 70% increase in bandwidth is the result of aggressive memory specification scaling, which placed new demands on memory suppliers.
Vera CPUs and SOCAMM2 LPDDR5X: Micron’s Competitive Edge
The "Vera" CPU, which utilizes SOCAMM2 LPDDR5X memory, positions Micron as a significant—potentially exclusive—supplier for this component. As NVIDIA introduces these CPUs as standalone competitors to established server-grade processors like Intel Xeon and AMD EPYC, the demand for SOCAMM2 LPDDR5X memory is expected to grow, further solidifying Micron’s role in the ecosystem.
While Micron may not be part of the HBM4 supply for NVIDIA’s latest AI systems, its involvement in supplying LPDDR5X memory for the Vera CPUs ensures the company remains a vital player in the evolving landscape of AI hardware. This development underscores the dynamic nature of the memory market and the importance of adaptability as system requirements continue to advance.
