NVIDIA locks down SK Hynix HBM supply with $500B AI infrastructure deal
NVIDIA announced Friday a multiyear technology partnership with SK Hynix to secure next-generation high-bandwidth memory (HBM) supply alongside a major infrastructure build. The agreement, announced at an AI summit in San Francisco with South Korean President Lee Jae Myung, involves the construction of large-scale data centers powered by NVIDIA's Vera Rubin systems with expected capacity of 2 gigawatts—requiring hundreds of thousands of GPUs when operational in 2027. The partnership spans NVIDIA's entire product roadmap from Vera Rubin AI supercomputers to RTX Spark PCs and Jetson Thor robotics platforms, with SK Hynix participating in co-development of next-generation AI memory across all three markets.
The deal underscores a critical supply-chain dynamic: memory has become the bottleneck in AI infrastructure scaling. SK Hynix is the global leader in HBM production for AI accelerators, and NVIDIA is securing supply years ahead as the chipmaker deploys increasingly memory-intensive systems. NVIDIA also committed $1 billion to South Korean cloud company Naver to expand AI data center capacity. SK Hynix affiliate SK Telecom will build a 2-gigawatt facility using Vera Rubin chips and SK Hynix's HBM4 memory, beginning operations in 2027. The companies will also apply AI to chip design and manufacturing, using NVIDIA CUDA-X and PhysicsNeMo to accelerate semiconductor simulations and develop factory digital twins for autonomous fab operations.
Architects tracking AI infrastructure economics should mark this: the deal signals that memory cost and supply—not GPU availability—is now constraining deployment velocity. For teams planning data center buildouts, it reinforces that HBM pricing and allocation will remain a material variable for 2027 and beyond. The partnership also reflects a broader shift toward government-scale and conglomerate-backed AI infrastructure buildouts beyond hyperscalers.
Sources
- Primary source
- nvidianews.nvidia.com
“SK hynix will codevelop memory for NVIDIA's platforms—from NVIDIA Vera Rubin to Jetson Thor—while advancing fab digital twins using NVIDIA Omniverse and cuOpt”
- koreaherald.com
“SK Hynix, Samsung and Nvidia are partnering to advance chip design and memory design as global demand rises for computing power and memory needed for AI”
- investing.com
“SK Telecom plans to build a 2-gigawatt AI data center powered by Nvidia's Vera Rubin chips and SK Hynix's HBM4 high-bandwidth memory, with the first facility due to come online in 2027”