Author: Su Yang

Editor: Xu Qingyang

On June 5, the South Korean stock market encountered a "Black Friday," with the KOSPI index closing down 5.54%. After opening on June 8, the intraday decline once expanded to over 8%, triggering a circuit breaker at the exchange, with Samsung and SK Hynix both plunging nearly 10%.

At this moment of market panic, Jensen Huang's visit dramatically took on the role of "saving the market."

Previously, on the evening of June 7 local time in South Korea, Jensen Huang held a "dinner meeting" with SK Group Chairman Chey Tae-won and SK Hynix CEO Park Sung-wook, among others.

After dinner, Jensen Huang confirmed to the media present that NVIDIA's newly launched Vera CPU will use SK Hynix DRAM; both parties are preparing for a "super-scale cooperation" in the second half of this year and next year; regarding the current shortage of memory chips, he believes it will last for several years.

Subsequently, NVIDIA and SK Hynix officially announced a multi-year technological cooperation agreement, involving the extension of AI supercomputers to robotics, digital twins, and semiconductor manufacturing.

Jensen Huang even made a direct endorsement at the press conference, stating, "If you are a shareholder of an AI company, you would feel happy; their prices are very low right now."

01 Locking in SK Hynix Memory

Vera is NVIDIA's first independent CPU specifically for data centers, facing competitors such as Intel's Xeon product line, AMD's Epyc chips, and self-developed projects by large cloud service providers like Amazon Graviton.

In this new battlefield, NVIDIA has anchored its memory supply with SK Hynix from the very beginning.

On June 7, NVIDIA and SK Hynix officially announced the establishment of a multi-year technical partnership to co-develop next-generation memory aligned with NVIDIA's AI infrastructure roadmap.

It is understood that the cooperation between the two parties covers a series of product lines for personal and cloud, including NVIDIA's Vera Rubin AI supercomputer, Vera CPU, PCs equipped with RTX Spark, and Jetson Thor robotic computing platforms.

The announcement pointed out that the cooperation aims to ensure the supply of advanced memory in response to the long development cycles, complex manufacturing processes, and high capital investments required for such products, thereby supporting the ongoing construction of AI factories worldwide.

The announcement also listed that SK Hynix will expand into several new markets that NVIDIA is pioneering, including AI infrastructure, personal AI, and physical AI.

02 AI Feeding Back Chip Manufacturing

In addition to supplying memory, SK Hynix has begun to incorporate NVIDIA's AI technology into its chip design and manufacturing processes.

Similar cooperation has already been implemented at TSMC, with "computational lithography" being the most typical example.

According to the announcement, SK Hynix is using NVIDIA's CUDA-X library and AI to accelerate semiconductor simulation, covering segments like technology computer-aided design (TCAD) and computational lithography.

The two parties are also promoting the extension of these tools to semiconductor electronic design automation (EDA) and simulation ecosystems, paving the way for tripartite cooperation among chip manufacturers, NVIDIA, and EDA software suppliers.

This means that the cooperation between the two is no longer limited to SK Hynix's own use, but is exploring a model that can be extended to the entire semiconductor industry.

In the manufacturing segment, SK Hynix is advancing the development of digital twin capabilities in wafer fabs, aiming for fully autonomous factory operations. This work is based on NVIDIA's Omniverse platform. With the Omniverse library and OpenUSD processes, SK Hynix can create 3D factory scenes for visualization, simulation, and optimization of complex semiconductor manufacturing environments.

At the factory operation level, these digital twin capabilities can also integrate with NVIDIA's cuOpt decision optimization engine and Metropolis platform to schedule autonomous mobile robots and other assets within the wafer fab.

The announcement also revealed that both companies are exploring ways to connect digital twins with existing traditional software and AI agent workflows, allowing AI systems to reason based on wafer fab data, execute tasks automatically, and improve manufacturing decisions.

03 Preparing Six Months in Advance

In October 2025, NVIDIA and SK Hynix announced a large-scale infrastructure cooperation.

At that time, SK Group was constructing an AI factory equipped with over 50,000 NVIDIA GPUs, with the first phase scheduled for completion by the end of 2027. Once completed, this is expected to become one of the largest AI factories in South Korea.

This factory will adopt a "GPU as a service" model, open to SK Group's subsidiaries and external organizations, aiming to accelerate the digital transformation and industrial innovation of the South Korean industry.

SK Telecom is also taking on specific deployment tasks within this project.

As NVIDIA's cloud partner, SK Telecom plans to use NVIDIA RTX PRO 6000 Blackwell server GPUs to build an industrial AI cloud in Asia, with an initial deployment scale of over 2,000 GPUs, specifically running Omniverse workloads to provide computational support for SK Hynix's semiconductor manufacturing, wafer fab digital twins, and internal AI agents.

During his visit to South Korea, Jensen Huang also revealed a piece of information: he is in discussions with telecom companies, as future AI will utilize telecom networks. This aligns with the direction of SK Telecom's participation in the cooperation.

04 Three Companies Share HBM4 Orders

Although NVIDIA has signed a multi-year technical cooperation agreement with SK Hynix, it has not put all its eggs in one basket regarding the supply of HBM4.

Upon arriving in Seoul, Jensen Huang made it clear to reporters: "All three suppliers have been qualified. All three suppliers are in production, and they are all competing to support Vera Rubin."

These three suppliers correspond to Samsung Electronics, SK Hynix, and Micron Technology.

In his keynote speech at the Taipei Computer Show, Jensen Huang confirmed that Vera Rubin has entered full production and is scheduled for delivery in the third quarter of this year. This system is built around NVIDIA's Vera CPU and Rubin graphics core cluster, with each server rack system equipped with TB-level HBM4 memory.

From the actual progress of HBM4, SK Hynix remains in the leading position.

Reuters reported last September that SK Hynix had completed internal certification for HBM4 chips at that time, and had established a production system for customers, aiming to be ready for mass production of 12-layer HBM4 products by the second half of 2025. Meritz Securities Senior Analyst Kim Sunwoo predicted at that time that thanks to early supplies to key customers and the resulting first-mover advantage, SK Hynix's HBM market share in 2026 would remain just over 60%.

05 Chip Shortages Will Continue for Years

Opening up the landscape for three suppliers for HBM4 does not mean that the supply pressure has been relieved.

After dinner, Jensen Huang also provided a less optimistic assessment. He told the media present that the shortage of memory chips will not end anytime soon. "The entire industry's supply chain—from wafers to packaging to silicon photonics... everything is in short supply due to such high demand. This situation will last for years."

The background to this statement is the nearly endless consumption of advanced memory resulting from the global construction of AI factories.

The shortage Jensen Huang referred to is not a lack of a specific material, but rather tight supplies at almost every link in the industry chain. NVIDIA's launch of Vera Rubin, promotion of AI factories, and entry into personal AI and physical AI fields are all ramping up demand for memory. This is also why he stated that all three HBM4 suppliers are competing to support Vera Rubin.

No one wants to fall behind in a situation of supply shortage.

During this visit to South Korea, although SK Group was a focus, it was not the only part of Jensen Huang's schedule. Upon arriving, he mentioned that he had arranged meetings with Hyundai Motor, LG, SK, Samsung, and Naver. He also revealed that NVIDIA is actively recruiting personnel for its new R&D center in South Korea. From these movements, it appears NVIDIA is systematically deepening its ties with the entire South Korean technology industry, with SK Group as a key link, but not the only one.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。