NVIDIA and SK Hynix Sign Multiyear Deal to Co-Develop Memory Across Entire AI Hardware Stack
The June 7 partnership locks in SK hynix as a co-engineer across Vera Rubin supercomputers, Vera CPUs, RTX Spark PCs, and Jetson Thor robotics, every major platform NVIDIA ships next.
NVIDIA and SK hynix announced a multiyear technology partnership on June 7, formalizing what was already an open secret in supply-chain circles: SK hynix is not just a memory vendor to NVIDIA, it's a co-engineer across the entire product roadmap.
According to a joint announcement published on the NVIDIA Newsroom, the two companies will co-develop next-generation memory spanning Vera Rubin AI supercomputers, Vera CPUs, RTX Spark-powered PCs, and Jetson Thor robotics platforms. That's not a narrow deal. It covers data-center training hardware, the first standalone NVIDIA CPU for server racks, consumer AI PCs, and edge robotics in one agreement.
The timing matters. Jensen Huang had confirmed just days earlier, at the GTC Taipei keynote on June 1, that Vera Rubin is in full production, with Samsung, SK hynix, and Micron all certified as HBM4 suppliers. But certification as a supplier and co-development of the memory architecture are different things. This partnership is the latter. As reported by Investing.com citing Bloomberg, supply-chain analysts estimate SK hynix holds roughly 60 to 70 percent of Vera Rubin's HBM4 volume, with Samsung at 25 to 30 percent and Micron supplying the remainder.
The architecture question underneath this deal is memory bandwidth. Vera Rubin is built around HBM4, the sixth generation of high-bandwidth memory. According to the NVIDIA Newsroom announcement, the platform delivers 10x agent throughput at scale compared to Grace Blackwell. That figure is only achievable if memory keeps pace with the GPU's compute density, which is why NVIDIA can't just buy memory off a shelf, it has to co-design it. The partnership codifies that co-design relationship into a contract.
The deal also goes further than memory supply. Per the NVIDIA Newsroom release, SK hynix will use NVIDIA's CUDA-X libraries and PhysicsNeMo to accelerate its own semiconductor simulations and chip design workflows. It'll also build factory digital twins using NVIDIA Omniverse and cuOpt, pushing toward what both companies are calling autonomous fab operations. That's a material expansion of SK hynix's scope: the memory maker becomes a user of NVIDIA's software stack inside its own manufacturing process.
Huang had already signaled the supply pressure publicly. As TechTimes reported, he told reporters during his South Korea visit on June 2 that global semiconductor supply remains tight and urged SK hynix to produce more HBM chips. A partnership agreement that ties SK hynix's roadmap to NVIDIA's for multiple years is one structural answer to that pressure, it gives SK hynix the demand visibility to justify the capital investment required to build out HBM4 and eventually HBM4E capacity.
For Vera Rubin's follow-on platform, Vera Rubin Ultra, TechTimes noted that next-generation HBM4E memory is expected in late 2027. This agreement is almost certainly the vehicle through which that co-development happens.
What this isn't is a guarantee of exclusivity. Samsung and Micron are both certified HBM4 suppliers. The partnership with SK hynix covers co-development and supply alignment, not a lock-out. But depth of integration tends to compound over time, and SK hynix's lead in the HBM4 qualification process, it entered ahead of rivals, combined with a formal co-engineering agreement puts it structurally ahead for at least the next two platform generations.
For anyone building or buying AI infrastructure, the relevant takeaway is straightforward: memory supply is a binding constraint on what NVIDIA can ship, and NVIDIA is now managing that constraint the way chip companies manage foundry relationships, through long-term co-development commitments, not spot purchases.
Sources cited:
- NVIDIA Newsroom (https://nvidianews.nvidia.com/news/sk-hynix-ai-factory)
- TechTimes (https://www.techtimes.com/articles/317994/20260608/nvidia-sk-hynix-sign-multiyear-memory-pact-why-your-next-pcs-ram-just-got-more-expensive-build.htm)
- Yahoo Finance / Investing.com (https://finance.yahoo.com/sectors/technology/articles/nvidia-certifies-samsung-sk-hynix-133001560.html)
- TechTimes (Vera Rubin production) (https://www.techtimes.com/articles/317539/20260602/nvidia-vera-rubin-enters-full-production-samsung-sk-hynix-micron-named-hbm4-suppliers.htm)
This release was originally distributed via ETL Newswire. Visit NVIDIA Newsroom for the full story, related releases, and contact information.
Visit NVIDIA Newsroom →