Industry Insiders Predict SK Telecom and Upstage to Compete for Final Spot
After the first evaluation of the “National Representative AI” project, industry insiders predict that SK Telecom and Upstage will compete for the final remaining spot among the top two teams. LG AI Research, which demonstrated overwhelming performance in the first evaluation, is highly likely to secure the top position, leaving the two companies to vie for the second spot. Although the government announced it would add one more team for the second evaluation through a public offering, it is challenging for new entrants to compete on equal footing with LG AI Research, SK Telecom, and Upstage, which began development in August last year.
“If Naver Cloud had remained in the race, both SK Telecom and Upstage might have struggled to make it to the final two,” said an AI industry insider. “However, with Naver’s exit, both companies now have a higher chance of surviving until the end.”
SKT Overcomes Hacking Crisis with AI
When SK Telecom was initially selected as a National Representative AI team in August last year, there were skeptical views: “Can a telecom company compete in a field dominated by AI firms?” Out of the five selected teams, SK Telecom was the only non-AI specialist. However, after the first evaluation, the industry reaction shifted: “We now see SK Telecom differently.”
SK Telecom developed an AI foundation model with a scale that dwarfed other competitors. In August last year, Ryu Young-sang, SK Telecom’s CEO, publicly pledged to create a 500-billion-parameter AI model—a goal the company achieved domestically for the first time (519B). LG AI Research’s model had 236B parameters, while Upstage’s was 100B.
Initially, the AI industry raised eyebrows at SK Telecom’s ambitious target. A 500B-scale model was deemed unrealistic given South Korea’s AI infrastructure limitations. Some even speculated the company might prioritize scale over performance. However, SK Telecom tied with LG AI Research for first place in the National Information Society Agency (NIA) benchmark, scoring 9.2 out of 10. A telecom industry insider said, “SK Telecom received full-scale group support to recover its tarnished image from a hacking incident through AI.” For the second evaluation, SK Telecom plans to expand into multimodal capabilities, enabling the model to process diverse data types such as images, voice, and video.
Upstage’s Corporate Value Skyrockets
As the only startup among the National Representative AI teams, Upstage saw its recognition and corporate value surge due to the project. Despite lacking the resources, manpower, and organizational strength of LG, SK, NC, and Naver, the company passed the first evaluation through sheer technical prowess. In individual benchmark tests, it scored a perfect 10, tying with LG AI Research.
Upstage’s greatest strength is its agility, with the entire company moving cohesively under CEO Kim Seong-hoon. When allegations arose last year that Upstage had copied a Chinese model, Kim immediately organized a public verification session the next day, swiftly dispelling the controversy. The startup’s rapid decision-making and transparent approach earned praise from AI experts. Upstage has now entered preparations for the second evaluation. A company representative said, “For the second evaluation, Stanford University and New York University will join our consortium, enabling us to elevate our technical capabilities. We aim to complete a globally top-tier model that surpasses big tech companies.”
Upstage’s strong performance in the National Representative AI competition has also brightened its prospects for an initial public offering (IPO). An investment industry insider said, “We understand Upstage is pushing for an IPO as early as the second half of this year. If it secures a spot in the final two, it could set a record as the most successful domestic AI startup IPO.” Upstage’s corporate value is estimated at 2 trillion Korean won.
Meanwhile, AI startups Motif Technologies and Trillion Labs have expressed interest in participating in the additional public offering for the National Representative AI project.
