Stargate Korea progress update coming within the year, OpenAI says
OpenAI says it will announce progress on its Samsung and SK-backed Korean data center project by year-end after design changes delayed plans.
OpenAI Korea General Manager Kim Kyoung-hoon speaks at a press event to celebrate the one-year anniversary of OpenAI Korea's unit launch on Sept. 9 at the Josun Palace in southern Seoul.
OPENAI
Progress on the Stargate Korea project, in which OpenAI plans to build data centers in Korea with Samsung and SK, will be announced within the year, OpenAI Korea General Manager Kim Kyoung-hoon said.
The Korea JoongAng Daily previously reported that Stargate Korea’s progress remains stalled, as little has been accomplished since the companies announced the deal a year ago.
“One thing I’ve learned over the past year is that building data centers involves more complexity than expected,” Kim said Wednesday at a press event marking the one-year anniversary of OpenAI’s Korean unit in southern Seoul.
“Our technology is advancing so quickly that the data center design we envisioned last October is actually somewhat different from what we need now. That has required some adjustments along the way, and it’s taken more time than we anticipated. But we’re continuing discussions, and we hope to share more news soon.”
OpenAI formed separate letters of intent with several affiliates of Samsung and SK, respectively, last October, along with one overarching deal with the two chipmakers to supply memory chips for OpenAI’s Stargate initiative, which involves building data centers across the globe to supply the computing power needed to train and operate OpenAI’s artificial intelligence models.
One part of the deal involved the U.S. AI startup extending its Stargate initiative to Korea with Samsung SDS, Samsung’s IT service arm, and SK Telecom, SK’s mobile carrier.
Samsung Electronics and SK hynix are also supplying high bandwidth memory (HBM) to power OpenAI’s data centers as part of the October deal. The chips are also expected to power OpenAI’s custom AI chip, Jalapeño.
The release of OpenAI’s latest AI model, GPT-6 Astra, has reignited excitement over how far AI can advance. It has also fueled the latest rally in Korean memory stocks, as tasks growing closer to the concept of artificial general intelligence will ultimately require more memory than before.
OpenAI said the number of ChatGPT Enterprise users in Korea has grown 28-fold from a year earlier as of the end of August.
Users of ChatGPT Work and Codex also surged 14-fold over the six months since March. Codex is OpenAI’s coding agent, while ChatGPT Work applies Codex’s agent technology to ChatGPT, handling tasks that range from research and analysis using apps and files to drafting documents.
“The number of ChatGPT users is far higher in Korea [now], but if you look at the share of output tokens, Codex and Work already account for 64 percent, with the chat output tokens making up the rest,” Kim said. “That should be understood to mean a large share of tokens are now being used for actual work.”
The U.S. company is also rolling out its Daybreak cybersecurity initiative in Korea. Work with government agencies and public institutions is proceeding separately, and OpenAI said it is working to give those agencies access to the models on a separate track.
In the private sector, Samsung SDS said Tuesday that it became the first Korean company to join Daybreak as a pilot partner. The cyber model will initially be used by Samsung SDS and then gradually deployed to other Samsung affiliates.
BY LEE JAE-LIM [[email protected]]