Nvidia and SK Hynix agreed to co-develop next-generation HBM memory and build an AI factory in South Korea, deepening a partnership that already controls 60 percent of the high-bandwidth memory market.
返回
Nvidia and SK Hynix deepen HBM partnership in 3 AI deals
The dinner in Woodside, California on July 24 produced a three-pronged collaboration spanning next-generation HBM memory co-development, an AI factory in South Korea targeting 2027 operations, and a joint research center with KAIST — reinforcing SK Hynix's position as Nvidia's primary HBM supplier at a time when memory supply is the industry's tightest bottleneck.
"This partnership goes beyond supplying memory — it's about co-designing the memory architecture for Nvidia's next AI platforms," SK Hynix CEO Kwak No-jung said at the dinner, according to a company statement.
SK Hynix controls roughly 60 percent of the global HBM market by revenue, and its entire HBM4 production for 2026 is already sold out, according to company disclosures. The chipmaker shipped samples of HBM4E in June — delivering 16 gigabits per second per pin, a 60 percent improvement over the prior HBM4 standard with 20 percent better power efficiency. A single Nvidia GPU using eight HBM4E stacks would have access to 384 GB of high-speed memory, enough to run AI models with hundreds of billions of parameters.
The deepening ties come as SK Hynix's market capitalization has crossed $1 trillion, overtaking Samsung as South Korea's most valuable company in late June. Its shares have surged more than 340 percent year to date, and the company's Nasdaq listing on July 10 raised $26.5 billion — the largest U.S. share sale ever by a foreign company. For Nvidia, the partnership secures access to the most supply-constrained component in its AI data center stack.
The three workstreams outlined at the dinner cover distinct parts of the AI value chain. SK Hynix will co-develop next-generation memory products aligned with Nvidia's AI infrastructure roadmap, extending a relationship that began with HBM3 and now spans HBM4 and HBM4E. SK Telecom, the SK Group affiliate, will partner with Nvidia on physical AI and robotics infrastructure, with plans to build an AI factory in South Korea that is expected to begin operations next year.
Nvidia also announced a partnership with the Korea Advanced Institute of Science and Technology (KAIST) to establish the country's first joint AI research center between a Korean university and a global technology company. The center will be housed at KAIST's Kim Jae-chul AI Graduate School and focus on agentic AI technologies, combining Nvidia's full-stack AI capabilities, its Nemotron open-source models, and computing resources from Nvidia's AI cloud partners.
HBM Supply Tightens as AI Demand Surges
The memory supply-demand imbalance is the structural backdrop for the partnership. DRAM suppliers are currently meeting only 75 percent to 80 percent of demand, and that fulfillment rate could fall to the 60 percent range by 2027, according to Kim Sunwoo, a senior analyst at Meritz Securities. SK Hynix's own CEO warned last week that the global memory industry could face its worst-ever supply shortage in 2027, with demand exceeding production capacity well beyond 2030.
The dinner also carried a symbolic dimension. Madison Huang, Nvidia's senior director of Omniverse and robotics product marketing and Jensen Huang's eldest daughter, attended — a sign of the personal ties between the two companies' leadership. South Korean President Lee Jae-myung is scheduled to meet with Huang at the AI Summit in San Francisco, part of a broader Korean delegation that includes Samsung Electronics Chairman Lee Jae-yong, Hyundai Motor Group Chairman Chung Eui-sun, and Naver founder Lee Hae-jin.
Investment Angle
For investors, the partnership reinforces a structural advantage for SK Hynix over rival Samsung Electronics, which holds roughly 28 percent of the HBM4 market, according to industry estimates. SK Hynix's first-mover advantage in HBM4 mass production and its exclusive co-development relationship with Nvidia create a competitive moat that rivals will find difficult to bridge before 2028. Nvidia, trading at roughly 35 times forward earnings, benefits from supply chain certainty for its highest-margin data center products. The AI Summit this week will serve as the next major event, with Huang expected to outline Nvidia's memory procurement strategy for 2027 and beyond.
This article is for informational purposes only and does not constitute investment advice.
[1] On the eve of the AI Summit, Huang Renxun meets with SK Chairman to discuss next-generation AI memory and AI factory cooperation[2] Jensen Huang Said Nvidia Will Be the First Customer for HBM4. Here's the AI Memory Stock That Has Reportedly Locked Up 70% of Those Orders.[3] SK Hynix Just Jumped 13% in a Day — AI Investors Are Losing Their Minds[4] SK Hynix: Strong HBM Tailwinds, But Mind ADR Premium And The Cycle[5] Micron: It Does Not Need To Beat SK hynix At Nvidia
