NVIDIA and SK hynix form multiyear partnership to co-develop AI memory for Vera Rubin, RTX Spark, and Jetson Thor

NVIDIA and SK hynix have announced a multiyear technology partnership to co-develop next-generation memory for AI infrastructure, personal AI computing, and robotics platforms. The agreement, announced June 7, 2026, builds on years of existing co-engineering work and covers memory for NVIDIA's Vera Rubin AI supercomputers, RTX Spark-powered PCs, and Jetson Thor robotics hardware.

What's new

The partnership spans three distinct market segments NVIDIA is pursuing:

• AI infrastructure: Co-development of memory for NVIDIA Vera Rubin AI supercomputers and Vera CPUs, targeting the AI data center market
• Personal AI: Memory for RTX Spark-powered Windows PCs, enabling on-device agents and AI workloads
• Physical AI / robotics: Memory for NVIDIA Jetson Thor robotic computing platforms

Beyond hardware, the partnership includes applying AI to semiconductor design and manufacturing — using CUDA-X libraries and PhysicsNeMo for physics-informed simulation, Omniverse for fab digital twins, and cuOpt for optimization of manufacturing processes.

No specific investment figures or financial terms were disclosed. The partnership is described as "multiyear" with no defined end date.

Context

High-bandwidth memory (HBM) and advanced DRAM are increasingly bottlenecks for AI compute, with demand driven by larger models, longer context windows, and the transition toward multi-chip GPU architectures. SK hynix is one of the world's three major DRAM manufacturers alongside Samsung and Micron, and has been a key HBM supplier to NVIDIA for its Hopper and Blackwell GPU generations.

The announcement comes as NVIDIA pursues its Vera Rubin generation of AI infrastructure — the successor to Blackwell — which will require next-generation memory technologies not yet in mass production.

Why it matters

AI training and inference are increasingly memory-bound rather than compute-bound. As models grow and context windows stretch into the millions of tokens, the bandwidth and capacity of memory stacks sitting directly on or alongside GPU dies determines how fast those systems run in practice. A formal multiyear co-development agreement — rather than a conventional supplier relationship — signals both companies are aligning roadmaps at the architecture level, not just placing purchase orders.

For the personal AI and robotics segments, the partnership means SK hynix will be engineering memory specifically for the thermal and form-factor constraints of RTX Spark (a superchip targeting thin-and-light PCs) and Jetson Thor (targeting humanoid robots and autonomous machines), rather than repurposing data center memory.

"AI factories are the engines of the next industrial revolution, and advanced memory is essential to their performance," said Jensen Huang, founder and CEO of NVIDIA.