NVIDIA and SK Hynix Sign $500B Alliance to Build 2GW AI Factory and Secure HBM Supply.

NVIDIA and SK Hynix Announce Historic $500 Billion Strategic AI Partnership

NVIDIA and South Korean memory giant SK Hynix have officially signed a massive, wide-ranging strategic partnership valued at $500 billion. The multi-year alliance encompasses joint AI Factory data center construction, next-generation memory development, and large-scale advance procurement commitments for High Bandwidth Memory (HBM) supplies.

A cornerstone of this landmark agreement is the construction of a flagship 2-gigawatt (2GW) AI Factory data center for SK Telecom in South Korea. Built on the NVIDIA DSX architecture and powered by NVIDIA’s next-generation Vera Rubin GPUs which integrate SK Hynix’s advanced HBM stacks the hyper-scale facility is projected to come online and begin offering compute services later this year.

Beyond immediate facility infrastructure, the alliance cements long-term supply chain security for NVIDIA's future hardware roadmaps while funding joint R&D initiatives for next-gen HBM4 and customized memory-logic integration.

The sheer power output of a 2-gigawatt data center: Traditional large-scale data centers typically use between 50 and 100 megawatts. A 2-gigawatt data center, equivalent to the power output of two nuclear reactors, represents unprecedented centralization of AI computing. This center operates on the NVIDIA DSX scheme, specifically designed to train and serve advanced models with trillions of parameters and high-level AI workloads in East Asia.

High-bandwidth memory (HBM) remains the most significant bottleneck in today's AI hardware manufacturing supply chain. As NVIDIA shifts its roadmap to the Vera Rubin platform, memory demand is rapidly increasing in both capacity and bandwidth. By pledging pre-orders within this $500 billion deal, NVIDIA gains primary access to SK Hynix's premium HBM production line, protecting delivery deadlines from global semiconductor shortages.

Another crucial industry perspective is the convergence of memory and logic architectures. As HBM evolves towards HBM4, the industry is shifting towards building HBM-based chips on advanced foundry nodes rather than using standard memory manufacturing processes. This collaboration enables NVIDIA and SK Hynix to co-design a custom HBM interface specifically optimized for the Rubin GPU, reducing power consumption and eliminating data transfer latency between the processor and memory.

 

Source: NVIDIA 

💬 AI Content Assistant

Ask me anything about this article. No data is stored for your question.

NVIDIA and SK Hynix Announce Historic $500 Billion Strategic AI Partnership

NVIDIA and South Korean memory giant SK Hynix have officially signed a massive, wide-ranging strategic partnership valued at $500 billion. The multi-year alliance encompasses joint AI Factory data center construction, next-generation memory development, and large-scale advance procurement commitments for High Bandwidth Memory (HBM) supplies.

A cornerstone of this landmark agreement is the construction of a flagship 2-gigawatt (2GW) AI Factory data center for SK Telecom in South Korea. Built on the NVIDIA DSX architecture and powered by NVIDIA’s next-generation Vera Rubin GPUs which integrate SK Hynix’s advanced HBM stacks the hyper-scale facility is projected to come online and begin offering compute services later this year.

Beyond immediate facility infrastructure, the alliance cements long-term supply chain security for NVIDIA's future hardware roadmaps while funding joint R&D initiatives for next-gen HBM4 and customized memory-logic integration.

The sheer power output of a 2-gigawatt data center: Traditional large-scale data centers typically use between 50 and 100 megawatts. A 2-gigawatt data center, equivalent to the power output of two nuclear reactors, represents unprecedented centralization of AI computing. This center operates on the NVIDIA DSX scheme, specifically designed to train and serve advanced models with trillions of parameters and high-level AI workloads in East Asia.

High-bandwidth memory (HBM) remains the most significant bottleneck in today's AI hardware manufacturing supply chain. As NVIDIA shifts its roadmap to the Vera Rubin platform, memory demand is rapidly increasing in both capacity and bandwidth. By pledging pre-orders within this $500 billion deal, NVIDIA gains primary access to SK Hynix's premium HBM production line, protecting delivery deadlines from global semiconductor shortages.

Another crucial industry perspective is the convergence of memory and logic architectures. As HBM evolves towards HBM4, the industry is shifting towards building HBM-based chips on advanced foundry nodes rather than using standard memory manufacturing processes. This collaboration enables NVIDIA and SK Hynix to co-design a custom HBM interface specifically optimized for the Rubin GPU, reducing power consumption and eliminating data transfer latency between the processor and memory.

 

Source: NVIDIA