NVIDIA, SK Hynix form multiyear partnership on next-gen memory for AI factories

NVIDIA and SK Hynix announced a multiyear technology partnership on June 7 to codevelop next-generation memory optimized for AI data center workloads. The agreement spans NVIDIA's Vera Rubin AI supercomputers, Vera CPUs, RTX Spark-powered PCs, and Jetson Thor robotic computing platforms, signaling deep integration of custom memory into NVIDIA's entire platform stack. The partnership also covers chip design optimization using NVIDIA CUDA-X libraries and advancing semiconductor fab digital twins.

The timing reflects structural shifts in AI infrastructure: memory has become the bottleneck (not compute), and hyperscalers are locking in long-term supply agreements to secure allocation. SK Hynix, one of the three global DRAM and HBM suppliers alongside Samsung and Micron, gains technology parity with NVIDIA on memory-for-AI while signaling confidence in sustained hyperscaler demand. Micron signed 16 similar SCAs worth $22 billion just days earlier; Samsung will likely follow. These partnerships represent a shift from transactional chip sales to co-design and capacity commitments.

For infrastructure teams, this codevelopment model means memory specifications will increasingly become hardware-specific rather than commodity. HBM4 demand remains fully booked through 2027, extending into 2028. Architects shipping GPU clusters should expect tighter coupling between NVIDIA's compute and NVIDIA-optimized memory from SK Hynix and other suppliers. The partnership also implies NVIDIA's broader ecosystem play—integrating CPUs, GPUs, memory, and software—is narrowing the addressable market for generic compute suppliers.