
Kim Seong-hoon, CEO of Upstage, expressed a negative stance regarding the government's plan to proceed with a Frontier AI model project separately from its Independent Foundation Model (Dokpamo) project. As an era in which AI develops AI models has begun and new Frontier AI models emerge daily, he pointed out that simply increasing GPU (Graphics Processing Unit) deployment to scale up size cannot reverse Korea's disadvantage.
On the 1st, after a seminar on "The Theory of Slowing Down AI Development" hosted by the National AI Strategy Committee, Kim (CEO) met with reporters and said regarding the government's Frontier AI model project, "It would be good to create a stage where innovative companies can excel." Upstage is currently in the third phase of the Dokpamo program.
The government is known to plan to select two out of three companies that advanced to the third phase—SK Telecom, LG AI Research Institute, and Upstage—and subsequently begin developing large-scale Frontier AI models similar to Anthropic's "Mito" and OpenAI's "Astra." The government plans to announce specific policies regarding the Frontier AI model project within this month.
Regarding the government's statement that it would deploy GPUs on a large scale to create a large-scale Frontier model with over 1 trillion parameters, Kim (CEO) said one should look at the case of Chinese AI innovation company "DeepSeek." He stated, "Looking at recent situations in the U.S. and China, when DeepSeek creates and releases new technology in China, it becomes the industry standard, after which Anthropic pushes forward," adding, "Innovative companies are at the center, while large corporations do what large corporations do."
He continued, "A model called DeepSeek 4.1 Flash has recently come out, and despite its small size, its performance is very excellent," emphasizing, "Anyone can improve performance by putting in many GPUs, but DeepSeek's approach is entirely different, making it a subject of research. We also aim to develop such technology to reach the frontier."
He further raised questions about sustainability regarding the Ministry of Science and ICT's plan to deploy approximately 10,000 of the latest Vera Rubin GPUs to create a large-scale Frontier model with over 1 trillion parameters.
Kim (CEO) stated, "We may be able to perform well temporarily by deploying GPUs on a large scale, but we need technological innovation like DeepSeek," emphasizing, "Otherwise, whether it is 1,000 GPUs or 10,000, we cannot win."