SK hynix And NVIDIA Push AI Factory Memory Into A Manufacturing Test

Memory Supply Moves Closer To NVIDIA’s AI Roadmap

SK hynix and NVIDIA have turned their long-running memory collaboration into a multi-year technology partnership aimed at the next phase of AI factory buildout.

The agreement covers next-generation memory for NVIDIA’s infrastructure roadmap and ties supply planning to the longer development cycles, fabrication requirements and capital investment needed for advanced memory.

The partnership is notable because it links memory development to specific NVIDIA platforms rather than treating memory as a separate component market.

SK hynix said it will co-develop memory for NVIDIA Vera Rubin AI supercomputers, NVIDIA Vera CPUs, RTX Spark-powered PCs and Jetson Thor robotics platforms.

That mix stretches from data-center systems to personal AI hardware and robotic computing.

For AI infrastructure buyers, the practical issue is timing.

Advanced memory cannot be added late in a platform cycle without affecting performance, qualification and supply planning.

A multi-year agreement gives SK hynix a clearer view of NVIDIA’s product direction, while NVIDIA gets a memory supplier aligned with the systems it is building for training, agentic AI and physical AI workloads.

The public record still lacks volumes, prices or capacity commitments.

It does show that the two companies are formalizing co-development around the memory layer that determines how much data AI accelerators can feed into compute systems.

Design Tools Become Part Of The Chip Strategy

The agreement also reaches inside SK hynix’s engineering workflow.

SK hynix is using NVIDIA CUDA-X libraries and AI to accelerate semiconductor simulation, including technology computer-aided design and computational lithography workflows.

It is also using NVIDIA PhysicsNeMo for in-house simulation codes and AI physics workflows.

That matters here because semiconductor simulation is one of the bottlenecks between a technology roadmap and actual manufacturing.

Faster modeling can help engineers test process behavior, lithography constraints and physics-driven workloads before decisions are locked into production steps.

The companies also describe a possible three-way collaboration model involving chipmakers, NVIDIA and electronic design automation software vendors.

This is not only a supply announcement.

It places NVIDIA’s software stack inside the way SK hynix designs and validates parts of its manufacturing and engineering process.

For a memory maker serving AI platforms, that creates a closer loop between the customer’s system roadmap and the supplier’s design environment.

Digital Twins Point To Autonomous Fabs

SK hynix is also developing fab digital twins as a foundation for autonomous fab operations.

The plan uses NVIDIA Omniverse libraries, OpenUSD pipelines and scene optimization technologies to build 3D factory environments for visualizing, simulating and optimizing semiconductor manufacturing.

The manufacturing angle extends to operations inside the fab.

SK hynix is looking at NVIDIA cuOpt and the NVIDIA Metropolis platform for optimizing the movement of autonomous mobile robots and other fab assets.

The companies are also exploring links between digital twins, legacy software and agentic AI workflows so AI systems can reason over fab data, automate tasks and support manufacturing decisions.

For the semiconductor sector, the larger point is that AI infrastructure demand is pushing suppliers to change both product development and production systems.

SK hynix is not only preparing memory for NVIDIA’s AI platforms; it is also applying NVIDIA tools to the fab and simulation stack that supports those products.

The next watchpoint is execution across the named NVIDIA platforms and SK hynix’s manufacturing workflow.

The announcement provides the product areas and toolchain direction, but it does not yet provide shipment timing, production volumes or commercial terms.