Story perspectives

Nvidia, SK Hynix Seal Multi-Year AI Memory Deal.

6/8/2026

Summary

Huang Secures AI Memory

On June 8 in South Korea, Nvidia CEO Jensen Huang announced a multi-year agreement with SK Hynix to co-design and manufacture AI memory and infrastructure chips.
The partnership supplies SK Hynix DRAM for Nvidia’s new Vera CPU and HBM4 memory for the Vera Rubin supercomputer, with deliveries scheduled for the third quarter of 2026.
Nvidia certified SK Hynix, Samsung Electronics and Micron Technology to mass-produce HBM4, the next-generation high-bandwidth memory for AI accelerators.
SK Hynix will also provide memory for Nvidia’s RTX Spark PCs and Jetson Thor robotics platforms under the deal.
Analysts say the long-term agreement gives SK Hynix production certainty as demand for AI training and inference memory grows.
SK Hynix shares slid 10% on June 8 amid a broader Asian tech sell-off, while Nvidia stock rose about 1.5% in early pre-market trading on June 8.
Huang warned that memory shortages will persist for years, prompting Nvidia to lock in supply through the partnership.

Summary

SK Hynix AI Push

SK Hynix will employ Nvidia’s CUDA-X and PhysicsNeMo to speed semiconductor simulation, including TCAD.
SK Hynix will build fab digital twins with Nvidia Omniverse, OpenUSD and cuOpt to enable fully autonomous factory operations and robot coordination.
Nvidia, SK Telecom and SK Hynix will develop an AI cloud platform on DSX architecture for a 2027 commercial launch.
Chairman Chey Tae-won said the venture seeks to become Asia’s leading AI cloud operator.

Summary

Nvidia Shapes AI

Nvidia now acts as a platform provider shaping the AI ecosystem instead of merely supplying chips.
New agreements make South Korea a central AI-infrastructure hub, prompting governments to view semiconductor collaborations through a national-security lens.
Samsung Electronics and Micron Technology feel pressure to secure comparable partnerships and expand capacity.
Nvidia CEO Jensen Huang described Hyundai Motor Group’s data centre in Saemangeum as an “AI Valley,” likening it to Silicon Valley.

Summary

SK Hynix Expands

SK Hynix and Nvidia will embed cuOpt and Metropolis in fab twins to coordinate robots.
At Computex on June 2, SK Group chairman Chey Tae-won warned memory shortages could last through 2030 because of AI demand.
He said SK Hynix aims to double wafer capacity within five years, although building a new fab requires least three years.
The agreement does not disclose chip-supply volumes, leaving Nvidia’s financial contribution and monetary terms opaque.

History

Historical Context

In late 1999 NVIDIA introduced the GeForce 256, the first GPU with onboard T&L; the 2006 CUDA platform, built for GeForce 8 series GPUs and later, let developers program GPUs and became the dominant GPGPU framework, sparking the modern AI hardware boom; today CEO Jensen Huang’s memory pact with SK Hynix extends that programmable-memory strategy to meet soaring AI-training demand.

1 of 5

Huang Secures AI Memory

On June 8 in South Korea, Nvidia CEO Jensen Huang announced a multi-year agreement with SK Hynix to co-design and manufacture AI memory and infrastructure chips.
The partnership supplies SK Hynix DRAM for Nvidia’s new Vera CPU and HBM4 memory for the Vera Rubin supercomputer, with deliveries scheduled for the third quarter of 2026.
Nvidia certified SK Hynix, Samsung Electronics and Micron Technology to mass-produce HBM4, the next-generation high-bandwidth memory for AI accelerators.
SK Hynix will also provide memory for Nvidia’s RTX Spark PCs and Jetson Thor robotics platforms under the deal.
Analysts say the long-term agreement gives SK Hynix production certainty as demand for AI training and inference memory grows.
SK Hynix shares slid 10% on June 8 amid a broader Asian tech sell-off, while Nvidia stock rose about 1.5% in early pre-market trading on June 8.
Huang warned that memory shortages will persist for years, prompting Nvidia to lock in supply through the partnership.

1 / 5