SK hynix and NVIDIA expand their AI-memory alliance, pulling advanced chips and fabrication into the same deal

SK hynix and NVIDIA have announced a multiyear technology partnership aimed at developing next-generation memory for NVIDIA’s AI computing platforms, broadening an existing supplier relationship into deeper joint work on hardware and production. The agreement, announced in Seoul on June 7 and reported by multiple outlets shortly afterward, positions memory as a strategic component of NVIDIA’s longer-term system roadmap, rather than a standardized commodity delivered on demand.

Under the partnership, the companies plan to co-develop advanced memory products intended to match the performance needs of AI systems built using NVIDIA compute platforms. The reported scope is notable because it extends beyond data center accelerators into other “physical AI” endpoints, including robotics and personal computing. In practice, the message is that memory requirements must be planned early enough to keep pace with the rest of NVIDIA’s platform development.

The deal is also described as including work that connects memory development to NVIDIA’s platform lines, including Vera Rubin AI supercomputers, Vera CPUs, RTX Spark-powered PCs, and Jetson Thor robotics computers. Vera Rubin is NVIDIA’s next-generation AI supercomputing platform line, while Vera CPUs are intended to complement NVIDIA accelerators in AI-focused servers and systems. RTX Spark is NVIDIA’s platform approach for AI-enabled PCs, and Jetson Thor refers to its robotics-focused computing line. Memory designed and validated around those platforms would need tight timing on bandwidth, latency, and reliability, especially as AI workloads scale up.

Industry observers have increasingly described AI accelerators as systems rather than standalone chips, where performance depends on the full hardware stack, including interconnects, networking, power, and memory. In that framing, the partnership gives NVIDIA a way to reduce uncertainty about whether future memory products will arrive quickly enough and with the right characteristics for each generation of its platforms. For SK hynix, it strengthens its role in the memory supply chain for the companies building AI hardware, and it pushes its involvement deeper into the architecture discussions that determine what kinds of memory will be needed.

The timing also comes amid ongoing concerns about memory supply. Separately, SK Group Chairman Chey Tae-won warned in March that global memory shortages could persist for several more years, potentially lasting through 2030. He attributed the shortage to a structural lack of wafer capacity and said securing additional wafer supply can take four to five years. He also suggested the industry supply shortfall could be more than 20% through 2030, underscoring why memory capacity planning matters to AI buyers.

NVIDIA’s reliance on advanced memory for AI training and inference is widely understood across the industry, but the specific partnership terms have not been fully detailed in the publicly available reporting summarized here. It is not clear, for example, what share of future memory output is reserved, whether the companies have agreed on minimum production volumes, or how the economics of the co-development work are structured. The announcements also do not specify which exact memory technologies will be prioritized first within the “next-generation” label.

Even without those specifics, the strategic direction is plain: NVIDIA and SK hynix are trying to align memory development, performance targets, and supply planning with NVIDIA’s expansion from cloud data centers to workstations, PCs, and robotics. If the approach succeeds, it could shorten the time between NVIDIA platform launches and the availability of memory configurations capable of meeting those platform performance needs.

For the next phase, investors and customers will likely watch whether SK hynix’s production planning translates into smoother availability of advanced memory for AI systems, and whether NVIDIA’s upcoming platform milestones continue to move on schedule. The partnership also sets up a clearer competitive lens for the memory market, where the “who designs the memory for the platform” question could matter as much as “who supplies it.”