SK Hynix and NVIDIA Partner on AI Factory Memory
Key insights
- NVIDIA and SK hynix will co-develop memory for four platforms: Vera Rubin AI supercomputers, Vera CPUs, RTX Spark PCs, and Jetson Thor robotics.
- SK hynix will use NVIDIA's CUDA-X libraries and PhysicsNeMo framework to accelerate TCAD semiconductor design simulations under the partnership.
- SK hynix will build autonomous fab digital twins using NVIDIA Omniverse, OpenUSD, cuOpt, and Metropolis for manufacturing operations.
Why this matters
The partnership shows that leading AI infrastructure vendors are moving beyond procurement to design-level integration, embedding the AI software stack directly into supplier hardware development cycles. For practitioners building on NVIDIA platforms, having SK hynix co-develop memory aligned to Vera Rubin and Jetson Thor roadmaps reduces the risk of memory bottlenecks that have historically constrained AI factory throughput. For founders and technical leaders, SK hynix's adoption of CUDA-X, PhysicsNeMo, and Omniverse for internal semiconductor workflows signals that AI-accelerated chip design and fab automation are becoming standard practice at tier-1 memory suppliers.
Summary
SK hynix and NVIDIA launched a multiyear technology partnership on June 7, 2026, to co-develop next-generation AI memory and bring NVIDIA's AI stack directly into SK hynix's chip manufacturing operations.
Beyond a supply agreement, SK hynix will use NVIDIA's CUDA-X and PhysicsNeMo to accelerate TCAD simulations, and build autonomous fab digital twins using Omniverse, OpenUSD, cuOpt, and Metropolis.
Essentially: (NVIDIA, SK hynix) are integrating at both the roadmap and manufacturing simulation levels.
- Platforms covered: Vera Rubin AI supercomputers, Vera CPUs, RTX Spark-powered PCs, Jetson Thor robotic computing
- SK hynix adopts CUDA-X and PhysicsNeMo for chip simulation; cuOpt and Metropolis for fab automation
- Jensen Huang: advanced memory is 'essential' to AI factory performance
Design-level integration between memory and compute suppliers is now the competitive baseline for AI infrastructure partnerships.
Potential risks and opportunities
Risks
- If Vera Rubin or Jetson Thor platform launches slip, SK hynix's roadmap-aligned memory development could face costly retooling or inventory misalignment
- Other memory suppliers now excluded from NVIDIA's early roadmap access may pivot to rival AI compute vendors, fragmenting the memory supply ecosystem for AI platform builders
- SK hynix's operational dependency on NVIDIA's CUDA-X and PhysicsNeMo for TCAD workflows creates single-vendor concentration risk if partnership terms shift or technology priorities diverge
Opportunities
- SK hynix secures a structural advantage over competing memory suppliers by gaining early access to Vera Rubin and Jetson Thor platform specifications, enabling faster time-to-market for optimized products
- CUDA-X and PhysicsNeMo adoption at a tier-1 memory manufacturer validates NVIDIA's software stack for semiconductor design, likely accelerating uptake at other major chipmakers
- Vendors and integrators in the NVIDIA Omniverse and OpenUSD ecosystem gain a high-profile reference deployment with SK hynix for autonomous fab digital twin buildouts
What we don't know yet
- Financial terms of the multiyear agreement: no dollar figures or investment commitments disclosed in the announcement
- Whether memory co-developed for NVIDIA platforms will be exclusive to NVIDIA or remain available to other AI chip customers of SK hynix
- Production timeline for Vera Rubin- and Jetson Thor-optimized memory: no target dates or volume ramp milestones provided
Originally reported by nvidia.com
Read the original article →Original headline: NVIDIA and SK Hynix Sign Multiyear Partnership to Co-Develop Next-Gen AI Factory Memory Across Vera Rubin, RTX Spark, and Jetson Thor