Nvidia and SK Hynix Form Multi-Year AI Memory Partnership

Nvidia and SK Hynix have entered into a multi-year technology partnership to co-develop next-generation memory solutions designed to support the global expansion of AI infrastructure [1]. This agreement aims to address the extended development cycles required for advanced memory as the two companies collaborate on hardware for AI factories, ranging from frontier model training to physical AI applications [2].

Key takeaways

• The partnership focuses on co-developing memory for Nvidia’s Vera Rubin supercomputers, Vera CPUs, RTX Spark PCs, and Jetson Thor robotic platforms [1].
• SK Hynix is integrating Nvidia’s CUDA-X libraries and PhysicsNeMo framework to accelerate semiconductor simulation and design workflows [2].
• The collaboration includes the development of "digital twins" for SK Hynix factories using Nvidia Omniverse and cuOpt to optimize manufacturing and autonomous operations [1].
• The deal is intended to ensure memory supply keeps pace with Nvidia’s infrastructure roadmap amid high global demand [2].

Expanding the AI Hardware Ecosystem

The partnership builds upon years of existing co-engineering collaboration between the two firms [1]. By formalizing this multi-year agreement, the companies intend to synchronize their development cycles, specifically targeting the hardware requirements of Nvidia’s upcoming platforms [2]. SK Hynix has already been certified to supply fourth-generation high-bandwidth memory (HBM4) for the Vera Rubin platform, with production for these lines now scheduled for the third quarter of 2026 [2].

Beyond memory hardware, the collaboration extends into the semiconductor manufacturing process itself. SK Hynix is utilizing Nvidia’s AI-driven software, including CUDA-X libraries, to enhance computational lithography and technology computer-aided design [1]. Furthermore, the company is implementing digital twin technology through Nvidia Omniverse to create 3D replicas of its manufacturing facilities [2]. These virtual environments are designed to simulate and optimize factory operations, including the movement of autonomous mobile robots and other assets, using the cuOpt decision optimization engine [1].

Why it matters

This strategic alliance is designed to sustain the buildout of global AI infrastructure by aligning memory production with Nvidia’s long-term technology roadmap [1]. As demand for AI hardware continues to outstrip supply, the partnership serves as a mechanism to manage the complexities of advanced semiconductor manufacturing [2]. While SK Hynix has faced challenges with production lead times—some of which have been pushed back to 2028—the company remains a central supplier for Nvidia’s next-generation computing platforms [2]. By integrating AI into the design and fabrication stages, the companies aim to improve manufacturing efficiency and accelerate the development of future AI computing systems [1].