South Korea Unveils $950B AI Chip Pacts With Nvidia, Broadcom

South Korea outlined $950 billion in AI chip initiatives in San Francisco this week, as Samsung Electronics and SK Group reached major agreements with U.S. technology companies. The package includes an SK Hynix–Nvidia partnership valued at more than $500 billion and a Samsung–Broadcom memorandum of understanding of up to $200 billion, according to the presidential office.

SK Group’s commitments total $750 billion. The centerpiece is a long-term collaboration with Nvidia that secures next‑generation memory supply and joint development of high‑bandwidth memory used to train advanced AI models and run emerging applications.

Samsung Electronics agreed with Broadcom on an MOU covering memory chips, foundry services and advanced packaging. The scope includes work on Broadcom’s next‑generation AI accelerator built with Samsung’s high‑bandwidth memory and sub‑2‑nanometer manufacturing processes.

The announcements came at an AI summit in San Francisco hosted by President Lee Jae Myung. The gathering brought together chip and AI leaders amid tight global supplies of high‑performance processors and memory. In his remarks, Lee introduced a “San Francisco AI Declaration” outlining plans to cooperate with partners on technology development and supply reliability. “I’d like to present a vision for the future of AI and propose global cooperation to make it happen,” he stated. Lee also met with Nvidia CEO Jensen Huang, OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei and Broadcom CEO Hock Tan, alongside South Korean corporate leaders including Samsung Electronics Executive Chairman Jay Y. Lee, SK Group Chairman Chey Tae‑won, Hyundai Motor Group Executive Chair Euisun Chung and Naver founder Lee Hae‑jin.

SK Group detailed an initiative with Nvidia that includes a project by SK Telecom to build a 2‑gigawatt data center using Vera Rubin chips paired with SK Hynix’s planned HBM4 memory. The facility is targeted to come online in 2027 to provide large‑scale computing capacity for model training and inference.

Executives at the summit pointed to demand for AI processors and memory that continues to outstrip available supply, prompting long‑term supply arrangements and joint development programs to secure capacity.

Chey Tae‑won indicated that memory requirements from major AI customers are trending higher than earlier forecasts, referencing recent discussions with Nvidia. “I wouldn’t be surprised if, the next time I see (Huang), he’ll tell me those figures were too low and say, ‘I want more,’” he remarked. He added that Broadcom expects its memory needs to exceed available supply and that Anthropic regularly checks future availability as it expands computing capacity. According to Chey, OpenAI has requested direct chip supply from SK to support its services. He also noted that Anthropic is building its own computing resources.

An Anthropic spokesperson said the company is exploring options and using a range of chips in a diversified approach. Amodei indicated the startup has supply deals with Samsung Electronics and SK Hynix.

Beyond the summit, Lee is scheduled to attend a meeting in Silicon Valley with South Korea’s National Pension Service and U.S. venture capital firms to encourage investment in South Korean startups. Last month, he outlined three large state‑driven projects anchored by Samsung Electronics and SK Hynix to develop semiconductor manufacturing clusters, physical AI ecosystems and AI data centers, with planned investment exceeding $576 billion.