Nvidia and SK hynix just unveiled a deep, multiyear partnership that ties AI chips, memory, and factory automation even closer together. The two companies will co-develop custom memory for Nvidia’s next wave of AI, PC, and robotics platforms while also using Nvidia AI to design and build chips inside SK hynix fabs.
Under the deal, SK hynix will create specialized high-performance memory for four Nvidia product lines: Vera Rubin AI supercomputers, Vera CPUs, RTX Spark-powered PCs, and Jetson Thor robotic computing systems. The goal is to match memory bandwidth, capacity, and power use directly to each platform’s workload instead of relying only on generic parts.
Vera Rubin systems and CPUs will need massive parallel memory pipelines to feed giant AI models and data analytics jobs. RTX Spark PCs, which target AI-enhanced personal computing, will lean on fast, power-efficient memory for local inference and creative tools. Jetson Thor will bring those same ideas into robots and edge devices, where real-time response matters more than raw size.
Nvidia CEO Jensen Huang summed up the idea by saying, “AI factories are the engines of the next industrial revolution, and advanced memory is essential to their performance.” His comment puts memory on the same level as GPUs and networking in the AI stack.
Using Nvidia AI to Design and Build Chips
The partnership also runs in the other direction. SK hynix will use Nvidia’s CUDA-X libraries and PhysicsNeMo framework to speed up chip design and manufacturing. That means it will run AI-powered simulations of transistors, materials, and packaging to test new ideas faster and reduce costly trial-and-error in the fab.
SK hynix will also use Nvidia Omniverse to develop “digital twins” of its factories. These precise virtual replicas of actual fabs will enable engineers to model production lines, tweak workflow and evaluate how changes can affect yield or downtime before they touch actual equipment. The ultimate goal is fully autonomous fab operations, with AI taking on more of the day-to-day choices.
The transaction underscores how strongly coupled AI is to memory tech today. As models expand in size and are deployed in more locations, from cloud supercomputing to home computers and robots, bottlenecks change from pure compute to the rate at which data can enter and leave memory.
READ MORE: Zcash Price Recovers After 50% Crash – Is a Bull Trap Forming?
