A recent report indicates Samsung Electronics and Micron are the initial suppliers of high-speed PCIe Gen6 SSDs for NVIDIA's new Vera Rubin AI platform, while SK hynix is not yet in the picture.
This development is significant because it reflects a fundamental shift in AI server design. NVIDIA's new CMX (Context Memory eXpress) architecture transforms ultra-fast SSDs from simple storage into an active component of the AI processing pipeline. It works by offloading a critical, memory-intensive workload called the KV-cache directly to these SSDs, bypassing the main CPU. This tight coupling means the SSD's speed directly impacts AI performance, creating immense demand for the fastest drives available—namely, PCIe Gen6 SSDs.
The current supply situation was largely solidified at NVIDIA's GTC conference in March 2026. First, Micron announced it was already in high-volume production of its 9650 Gen6 SSD, demonstrating its immediate supply readiness. Second, Samsung explicitly positioned its PM1763 Gen6 SSD as the "main storage" for the Rubin platform, implying a primary design win. Third, NVIDIA's detailed reveal of the CMX architecture raised the technical bar, creating a market dynamic that heavily favors vendors with ready-to-ship Gen6 products.
Furthermore, this outcome was foreseeable based on earlier announcements. SK hynix had already signaled in January 2026 that its Gen6 SSDs were targeting mass production in late 2027, effectively removing them from contention for the initial launch. In contrast, Samsung and Micron had been signaling their progress for months. Micron, in particular, gained a crucial first-mover advantage by introducing the world's first Gen6 data center SSD back in July 2025.
In conclusion, the Samsung-Micron duopoly for the initial Rubin launch is not a surprise. It is the logical result of strategic timelines and technological preparedness. Both companies were ready for the moment NVIDIA made Gen6 SSDs a performance-critical component of its next-generation AI platform, while SK hynix appears to be targeting the subsequent refresh and expansion cycle.
- Glossary
- PCIe Gen6: The sixth generation of the Peripheral Component Interconnect Express standard, a high-speed interface used to connect components like graphics cards and SSDs to a motherboard, offering double the bandwidth of the previous generation.
- KV-cache: In AI models like transformers, the Key-Value cache stores intermediate calculations (attention keys and values) to speed up the process of generating sequential outputs, like text. Offloading it to fast storage reduces the burden on expensive high-bandwidth memory (HBM).
- CMX (Context Memory eXpress): An NVIDIA storage platform architecture that uses NVMe SSDs to expand the memory available for large AI models, specifically for storing the KV-cache. It is managed by a BlueField DPU to create a seamless, high-performance memory tier.
