NVIDIA and SK Hynix Sign Multi-Year Agreement to Co-Develop Next-Generation AI Memory
NVIDIA and SK Hynix have signed a multi-year cooperation agreement to jointly develop next-generation memory chips tailored for NVIDIA's AI infrastructure roadmap, including those for the Vera Rubin platform.
The agreement covers chip design and manufacturing, with SK Hynix providing customized memory for NVIDIA's Vera Rubin AI supercomputer, Vera CPU, RTX Spark PC, and Jetson Thor platform, while leveraging NVIDIA's tools such as CUDA-X, PhysicsNeMo, and Omniverse to accelerate design and simulation.
This deepened cooperation focuses funding and supply chains on AI memory and infrastructure, benefiting NVIDIA's ecosystem partners, while SK Hynix aims for growth through diversification into areas like physical AI. Competitors face pressure on supply share due to locked agreements.
Source: Public Information
ABAB AI Insight
NVIDIA has previously certified Samsung, SK Hynix, and Micron as suppliers for Vera Rubin HBM4, with SK Hynix expected to capture about 60-70% of the supply share, continuing its close ties with NVIDIA in the HBM sector, having previously dominated the Blackwell platform supply with products like HBM3E.
In terms of capital pathways, NVIDIA will output technical specifications and ecosystem tools to SK Hynix, mobilizing the latter's production capacity and R&D resources through joint design, motivated by the need to ensure stable memory supply for next-generation platforms like Vera Rubin and to accelerate expansion into scenarios such as physical AI and robotics, while consolidating supply chain dominance.
This is similar to NVIDIA's long-term ties with TSMC and Samsung in process technology, and the past shift of HBM supply chains from Samsung to SK Hynix, aligning with the current transition of AI infrastructure from chips to a full-stack memory-computing synergy.
Essentially, this represents a restructuring of the industry chain: the exponential growth of AI computing power drives memory customization, mechanism-wise shifting capital and technology from generic components to vertically integrated supply chains through multi-year agreements and joint development, further strengthening NVIDIA's pricing power and barriers with core partners, avoiding supply bottlenecks that could constrain overall ecosystem expansion.
ABAB News · Cognitive Law
The thirst for computing power persists, and memory binding serves as a supply chain leverage; winners sell ecosystems rather than individual products.
General to customized, short-term costs rise while long-term barriers are built, with structural restructuring stemming from demand asymmetry.
Most suppliers compete for share, while a few partners lock agreements, with top capital concentrating on scarce interfaces.