NVIDIA and SK hynix sign multiyear tech partnership to co-develop memory for AI factories

NVIDIA and SK hynix said on June 7, 2026 that they will pursue a multiyear technology partnership aimed at advancing next-generation memory for the global buildout of “AI factories,” and to accelerate semiconductor design and manufacturing. The companies positioned the work as an extension of years of co-engineering behind some of the most widely deployed AI computing platforms.

In the announcement, NVIDIA CEO Jensen Huang said “advanced memory is essential” to AI factories’ performance and that SK hynix has been a central partner in delivering those memory technologies. The companies said the multiyear agreement is intended to support memory supply as development cycles lengthen for advanced memory, aligning output with NVIDIA’s infrastructure roadmap.

SK hynix also said the partnership would expand beyond server memory supply into additional markets NVIDIA is creating across AI infrastructure, personal AI, and physical AI. Specifically, the announcement said SK hynix will codevelop memory for NVIDIA’s Vera Rubin AI supercomputers, Vera CPUs, RTX Spark-powered PCs, and Jetson Thor robotic computing platforms.

To support the memory roadmap, SK hynix said it is using NVIDIA CUDA-X libraries, alongside AI, to speed semiconductor simulation, including technology computer-aided design (TCAD) and computational lithography workflows. In plain terms, TCAD is a set of computer models used to predict how semiconductor processes and device designs will behave before expensive physical fabrication begins.

The companies also pointed to PhysicsNeMo, NVIDIA’s AI framework for physics-based learning, as part of SK hynix’s efforts to accelerate core simulation workloads and AI physics workflows. NVIDIA described the approach as extending its simulation tools into semiconductor electronic design automation and simulation ecosystems, with the goal of enabling “three-way” collaborations across chipmakers, NVIDIA, and electronic design automation software vendors.

Separately, SK hynix said it is developing “fab digital twins,” meaning 3D virtual replicas of semiconductor manufacturing environments that can be visualized, simulated, and optimized. In its description, SK hynix will build those scenes using NVIDIA Omniverse libraries and OpenUSD pipelines, and it will use NVIDIA cuOpt (a decision optimization engine) and the NVIDIA Metropolis platform to optimize operations such as the movement of autonomous mobile robots and other fab assets.

The announcement landed in a broader industry moment where memory and data-center infrastructure are increasingly treated as co-designed systems rather than interchangeable components. NVIDIA has previously described AI factories as a new kind of manufacturing plant enabled by accelerated computing and software, and SK hynix has linked its own messaging to the idea that AI factories are data-center infrastructures that generate value from data.

While the companies outlined the technical areas of collaboration, the announcement did not disclose the specific memory product targets (for example, particular generations or density targets), the length of the multiyear term, or any financial terms. It also did not specify whether there are binding commitments on quantities, pricing, or delivery schedules for memory used in NVIDIA systems.

Looking ahead, investors and customers may watch for engineering milestones that connect the software and digital twin work to measurable manufacturing throughput and for announcements tied to NVIDIA platforms named in the release. If the digital twin and accelerated simulation efforts deliver faster design iteration and higher factory optimization, the partnership could become a template for how memory makers and GPU platform vendors iterate together as AI infrastructure scales.