Nvidia Partners With SK Hynix on Next-Gen Memory as AI Factory Buildout Accelerates

As Jensen Huang continues his high-profile tour of Asia, Nvidia (NVDA) and SK Hynix revealed a multiyear technology partnership to advance next-generation memory for the global AI factory buildout. The focus has shifted from Computex in Taiwan to South Korea. The deal builds on years of working together as co-engineers to speed up the creation and production of semiconductors. SK Hynix recently hit a record high of $1.12 trillion, making it the third company in Asia to have a market value of more than $1 trillion. It joined TSMC and Samsung in this exclusive club.

Nvidia's founder and CEO, Jensen Huang, said, "SK Hynix has been an amazing partner to NVIDIA, playing a central role in delivering advanced memory technologies for NVIDIA AI computing platforms." We will work together to create the next generation of memory for AI companies and help the world's AI infrastructure grow faster, covering everything from cutting edge model training to physical and agentic AI.

Chey Tae-won, Chairman of SK Group, said: "Together, we are codeveloping the next generation of memory for AI factories and applying AI to how we design and manufacture semiconductors — work that will shape the future of AI infrastructure."

What the Partnership Covers: From Vera Rubin to Robotics

SK Hynix will keep offering supplies to meet the longer development cycles of advanced memory. Together, they will work on memory for Nvidia Vera Rubin AI supercomputers, Nvidia Vera CPUs, Nvidia RTX Spark-powered PCs, and Nvidia Jetson Thor robotic computing platforms. This partnership also includes new markets that Nvidia is building.

SK Hynix will use AI and Nvidia CUDA-X libraries to speed up semiconductor simulation. These will include computer-aided design and computational lithography processes. Along with CUDA-X and the Nvidia PhysicsNeMo framework, the Korean chipmaker will also use them to speed up their own simulation codes and AI physics processes.

Adding these tools to the ecosystem for semiconductor electronic design automation makes it possible for chipmakers, Nvidia, and EDA software companies to work together.

The Korea Expansion: Naver, Doosan, and SK Telecom Join

Huang's trip to Korea led to more business deals. SK Telecom, which is related to SK Hynix, will build a new gigawatt-scale AI cloud in South Korea. It will use Nvidia's AI chips and infrastructure to run it. Early next year, the first data center should be up and running.

Naver, the huge internet company, plans to grow its Gak Sejong data center and then build more gigawatt-scale AI factories, but this depends on when they can get more space and power.

Doosan Group is planning a number of new projects. They are already working on clever robots and parts for Nvidia GPUs. The energy options that Nvidia will use in its data center platforms come from Doosan. Doosan will use physical AI technology from Nvidia to power its robots.

The partnerships help Nvidia do more business in South Korea, which is one of the world's biggest technology industry hubs and home to huge chip, electronics, car, and ship-building industries. Two of the three biggest memory chip makers in the world are SK Hynix and its peer Samsung Electronics. AI data centers need these chips to work.

The market cap that SK Hynix reached was $1.12 trillion. This proves that memory is no longer a commodity play but a strategic AI infrastructure layer. The long-term deal between Nvidia and Intel guarantees that there will be a clear picture of demand for advanced memory through the Vera Rubin supercomputer cycle and beyond.

The size of the data centers that SK Telecom, Naver, and maybe even Doosan want to build is gigawatt-scale. This level of building out AI infrastructure in Korea provides a regional demand anchor for Nvidia's chips and SK Hynix's memory that is not affected by how much hyperscalers in the U.S. spend.

Simulation tools like PhysicsNeMo and CUDA-X could help SK Hynix speed up the development of improved memory. It gives you a lead over Samsung in the HBM (High Bandwidth Memory) race because faster simulation means faster node transitions.

The longer-term sign is the three-way EDA cooperation. If Nvidia's simulation tools become the standard for designing memories, the company will be even more locked into the semiconductor manufacturing stack, not just GPUs.

The clock for the trade is the Vera Rubin machine. If Nvidia announces its next-generation AI platform, keep an eye on whether SK Hynix memory is mentioned. If not, see if Samsung's countermoves with its own HBM4 plan pick up speed. Asia is home to three chipmakers in the $1 trillion club. They are just beginning to fight over memories.