Nvidia Secures SK Hynix DRAM for Vera Data-Center CPU

At Computex 2026 and during a visit to Korea, Nvidia and SK Hynix announced a multiyear memory partnership, and Nvidia's newly unveiled Vera data-center CPU will use DRAM supplied by SK Hynix, according to Seeking Alpha (summarizing Bloomberg), Yonhap, and other coverage. CEO Jensen Huang confirmed the Vera memory sourcing, and SK Hynix said the deal spans memory co-development for Nvidia's Vera Rubin AI systems, Vera CPUs, RTX-class PCs, and the Jetson Thor robotics platform, per Yonhap and Korea Times. Coverage describes Vera as Nvidia's strongest push into standalone server CPUs, succeeding its Grace CPU line and targeting x86 incumbents Intel and AMD; reported Nvidia figures include 88 custom "Olympus" cores and up to 1.2 TB/s of memory bandwidth. Separately, Nvidia has certified Samsung, SK Hynix, and Micron for HBM4 supply in the Vera Rubin platform, so memory sourcing is multi-vendor rather than exclusive to SK Hynix.
What happened
Nvidia and SK Hynix announced a multiyear memory partnership, and Nvidia's newly unveiled Vera data-center CPU will use DRAM supplied by SK Hynix, according to Seeking Alpha (summarizing Bloomberg), Yonhap, Korea Times, and other coverage. CEO Jensen Huang confirmed the Vera memory sourcing, and SK Hynix described a deal to jointly engineer memory for Nvidia's Vera Rubin AI systems, Vera CPUs, RTX-class PCs, and the Jetson Thor robotics platform, per Yonhap and Korea Times. The announcements followed Nvidia's Vera unveiling at Computex 2026 and Huang's visit to Korea.
Vera and the Grace context
Coverage frames Vera as Nvidia's strongest push into standalone server CPUs, positioned against x86 incumbents Intel and AMD. Vera succeeds Nvidia's Grace CPU line, which the company paired closely with its GPUs; Nvidia's own materials describe the move "from Grace to Vera." Reported, Nvidia-supplied figures for Vera include 88 custom "Olympus" cores and up to 1.2 TB/s of memory bandwidth, with the company claiming faster task completion than traditional x86 processors; these are vendor figures and are not independently benchmarked. Public reporting does not provide module-level part numbers for the SK Hynix DRAM, so it is best read as a sourcing and partnership announcement rather than a technical datasheet.
Why it matters
Large-scale AI systems pair high-throughput processors with very large memory footprints, so memory suppliers that secure design wins for server CPUs and accelerators can see multiyear demand. A formalized, multi-product memory partnership reinforces SK Hynix's role in Nvidia's supply chain at a time of tight AI-memory supply. Importantly, the arrangement is not exclusive: Nvidia has separately certified Samsung, SK Hynix, and Micron for HBM4 supply in the Vera Rubin platform, so multiple vendors share the broader memory opportunity even as SK Hynix deepens its CPU-DRAM position.
For practitioners, what to watch
Watch for Nvidia or SK Hynix product briefs listing DRAM type, capacity, and bandwidth for Vera; OEM and cloud-provider announcements citing Vera configurations; and SK Hynix earnings commentary or memory-industry reports as leading indicators of contract scale. For deployment planning, the split between CPU-attached DRAM and the HBM4 on the accelerators, and which vendors win each, will shape availability and pricing for Vera Rubin systems.
Key Points
- 1Nvidia and SK Hynix announced a multiyear memory partnership; Nvidia's newly unveiled Vera data-center CPU will use SK Hynix DRAM, a sourcing CEO Jensen Huang confirmed directly (per Yonhap, Seeking Alpha, and other coverage).
- 2SK Hynix says the deal covers memory co-development across Nvidia's Vera Rubin AI systems, Vera CPUs, RTX-class PCs, and Jetson Thor robotics, deepening an already close supplier relationship.
- 3Memory sourcing is multi-vendor: Nvidia has separately certified Samsung, SK Hynix, and Micron for HBM4 supply in Vera Rubin, so the SK Hynix design win is significant but not exclusive.
Scoring Rationale
A notable AI-infrastructure supply-chain development: Nvidia and SK Hynix formalized a multiyear memory partnership spanning Vera CPUs, Vera Rubin systems, RTX-class PCs, and Jetson Thor, confirmed by CEO Jensen Huang, with direct relevance to practitioners planning AI hardware procurement. It is a vendor partnership and design-win story, and memory sourcing is multi-vendor (Samsung and Micron are also certified for HBM4), rather than a frontier model or technical breakthrough, so impact is solid-to-notable.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems