NVIDIA and SK hynix sign multi-year alliance to develop next-generation memory for AI factories
NVIDIA and SK hynix have announced a multi-year technology alliance focused on the development of next-generation memory for AI factories, a collaboration with which both companies seek to accompany the sustained growth of the infrastructure necessary for model training, agentic artificial intelligence and physical AI systems. The agreement also includes joint work on semiconductor simulation, chip design and autonomous operations within manufacturing plants.
As explained by the two companies, the alliance is based on previous years of co-engineering between NVIDIA and SK hynix on advanced computing platforms for AI. In this new phase, the focus is on ensuring that the supply and evolution of memory accompany NVIDIA’s infrastructure roadmap, something especially relevant in a segment where development cycles are long, manufacturing is complex and the necessary investments continue to grow.
NVIDIA has presented advanced memory as an essential component for the performance of the so-called AI factorieswhile SK hynix has indicated that this collaboration reflects the depth of a relationship that has been maturing for years. From now on, both companies will work on memory for systems such as Vera Rubin AI supercomputers, Vera CPUs, computers with RTX Spark and robotic platforms based on Jetson Thor.
Memory becomes a strategic piece in NVIDIA’s roadmap
The agreement is not limited to a supply issue. NVIDIA and SK hynix talk about memory co-development aligned with the next steps of the American company’s infrastructure, something that places SK hynix in several of the new markets that NVIDIA wants to promote, from data centers for AI to personal AI and physical AI. That is, memory stops appearing only as a critical component of the present and becomes part of the technological planning of several future generations of products.
The collaboration also addresses how to apply artificial intelligence to the chip design and manufacturing process itself. SK hynix will use CUDA-X and NVIDIA PhysicsNeMo to accelerate semiconductor simulations, TCAD workflows, computational lithography, and internal engineering codes. The idea is to reduce time and improve efficiency in tasks that are increasingly costly due to the complexity of advanced nodes and new packaging and performance requirements.
Furthermore, both companies propose that this work can open the door to three-way collaborations between chip manufacturers, NVIDIA and EDA software providers. This point is important because it shows that the ambition of the agreement goes beyond the memory-GPU binomial and also points to the ecosystem of tools necessary to design more advanced semiconductors.
Digital twins and more alliances to build the ecosystem of AI factories
Another of the axes of the agreement involves the creation of digital twins for semiconductor factories. SK hynix will develop these environments as a foundation for autonomous factory operations, combining NVIDIA Omniverse libraries, OpenUSD flows, scene optimization technologies, and the cuOpt engine to improve the management of autonomous mobile robots and other assets within the factory. Companies are also studying connecting these digital twins with legacy software and agentic AI flows to automate tasks and improve decision making.
The alliance also fits with other recent NVIDIA moves in South Korea. In parallel, the company has announced agreements with LG Group and Doosan Group to work in robotics, mobility, energy and infrastructure of AI factories. Compared to these pacts, more oriented towards industrial deployment, data centers and automation, the collaboration with SK hynix is located further down the technological chain, in an essential component so that all this infrastructure can continue to scale.
Together, NVIDIA continues to expand a network of partners with which it wants to cover various levels of the AI business, from from memory and chip design to energy, robotics and digital twins. In the case of SK hynix, the value of the agreement is in ensuring evolution and supply capacity in a critical piece of infrastructure, just at a time when the global expansion of AI factories begins to depend both on software and GPUs and on the availability of advanced memory capable of sustaining that growth.
