Nvidia secures SK Hynix memory supply in AI infrastructure deal

Nvidia said it has secured Nvidia SK Hynix memory supply under a multiyear agreement that could be worth $500 billion, tightening access to a component central to advanced AI chips. The company said the arrangement includes large data centers expected to start operating in 2027, adding to signs that AI infrastructure spending is broadening beyond the largest U.S. cloud providers.

The agreement was announced late Friday in San Francisco at an AI summit attended by South Korean officials, including President Lee Jae Myung, according to Nvidia. The chipmaker said SK Telecom, an affiliate of SK Hynix, will build a cloud business using Nvidia’s Vera Rubin systems.

Nvidia said it is aiming for enough capacity to require 2 gigawatts of power. The company said that scale points to a buildout involving hundreds of thousands of graphics processing units, the chips used to train and run large AI models.

What does the Nvidia SK Hynix deal include?

The agreement gives Nvidia a more stable supply of high-bandwidth memory from SK Hynix, while also linking SK Telecom’s planned cloud business to Nvidia’s next-generation AI systems. Nvidia enterprise vice president Raj Mirpuri told reporters that the expansion includes a co-development opportunity around next-generation SK Hynix AI memory and would help Nvidia secure stable HBM supply.

High-bandwidth memory, or HBM, is a type of advanced memory stacked close to a processor so data can move quickly between the two. In AI systems, that speed matters because graphics processors must handle vast quantities of data during model training and inference.

Nvidia has been working to secure HBM as demand for AI systems strains global memory supply. Analysts cited by CNBC have described SK Hynix as the leader in HBM production, making the South Korean company a key supplier for AI chipmakers.

The deal also reflects the growing role of governments and large industrial groups in AI infrastructure. Nvidia said the buildout would add capacity outside the small group of hyperscale cloud companies that have dominated demand for advanced AI chips.

Samsung and Broadcom sign separate AI infrastructure pact

Samsung Electronics separately said it signed a memorandum of understanding with Broadcom to expand cooperation across memory and foundry technologies. Samsung said the agreement, estimated at $200 billion, is intended to support the next generation of AI infrastructure.

Nvidia also said Friday it would invest $1 billion in Naver, the Korean cloud company building data centers around Nvidia GPUs. Nvidia said the project could let prospective customers in South Korea and abroad secure AI computing capacity before the planned 200 megawatts of capacity is completed.

SK Hynix, South Korea’s second most valuable company, listed on Nasdaq earlier this month. CNBC reported that the listing was part of the company’s effort to finance infrastructure development.

This story draws on original reporting from CNBC.