On the morning of the 19th, AMD Chief Executive Officer (CEO) Lisa Su and Upstage CEO Kim Sung-hoon shake hands as they leave after concluding a breakfast meeting. Photo by reporter Han Chae-yeon chaezip@donga.com
Lisa Su, CEO of AMD, met with Upstage CEO Kim Sung-hoon at the Four Seasons Hotel in Jongno-gu, Seoul, on the 19th to discuss cooperation measures to accelerate the development of next-generation artificial intelligence (AI).
CEO Su said, “Upstage is an excellent partner,” adding, “We are currently enjoying our ongoing collaboration, and I look forward to both companies further expanding their cooperation going forward.”
At the meeting, the two companies agreed to adopt AMD’s Instinct MI355 graphics processing unit (GPU) for use in the development of Upstage’s proprietary large language model (LLM) “SOLAR” and its document-processing AI solutions.
The relationship between AMD and Upstage began in August last year when AMD joined as a new investor in Upstage’s Series B bridge financing worth KRW 62.0 billion. Observers note that AMD, which is working to expand its AI infrastructure ecosystem, choosing Upstage as an AI model partner was a predictable step. Upstage is the only AI startup included on Lisa Su’s schedule for this visit to Korea.
Kim said, “We proposed to serve as a testbed to demonstrate that a sovereign AI model runs smoothly on a scale of 10,000 AMD chips, and CEO Su responded very positively.”
On the back of the partnership, the supply of the latest GPUs is also expected to accelerate. Kim added, “They said they would supply a large number of GPUs to Korea at low prices,” and “There was discussion to the effect that Upstage would be given priority access not only to AMD’s latest model, the MI355X, but also to products to be released in the future.”
The volume of GPU supply may expand to more than about 10,000 units. For Upstage, which is seeking to acquire the portal site “Daum” from Kakao, securing large-scale GPU capacity is essential. Kim said, “Once the Daum acquisition is completed, we expect to process tokens at a level of 1 trillion per day,” adding, “To achieve this, around 10,000 GPUs need to be supplied quickly.” He continued, “If we expand into AI agents, GPU demand could increase by up to 100 times.”
ⓒ dongA.com. All rights reserved. Reproduction, redistribution, or use for AI training prohibited.
Popular News