"With More GPUs, Can't We Keep All National AI Teams?"... Doubts Rise Over Government Selection Process [IT Cafe]

Consortiums Face Mounting Pressure Ahead of First Elimination Round
"This Evaluation Will Decide the Team's Fate"
"With Improved Infrastructure... Need to Broaden the Market for Participating Companies' Technologies"

As the first elimination among the five elite national AI teams participating in the government's "independent artificial intelligence (AI) foundation model" project approaches, voices in the IT industry are asking, "Isn't there a way to keep them all?" The government plans to select the first team for elimination among the five consortia-Naver Cloud, Upstage, SK Telecom, NC AI, and LG AI Research-in January next year. The prevailing opinion in the industry is that, since each team has its own unique strengths, it would be beneficial for the IT sector to maximize these advantages.

There are several reasons behind these opinions. First, some analysts point to the adjustment of the evaluation timeline, which was moved from the end of this year to mid-January next year. As the government postponed the timing of the first elimination in response to requests from the consortia, calls to reconsider the overall operation of the project have gained traction.

The growing pressure on each consortium ahead of the first round of evaluations is also cited as a factor. If a team is selected as the first to be eliminated, it could lead to disappointment within the organization and even raise questions of responsibility, resulting in significant psychological pressure on participating companies. Since the AI business does not immediately generate profits, and with many IT companies preparing for year-end and New Year personnel changes, including organizational restructuring and workforce reductions, some say, "This evaluation will determine the team's future." An industry insider expressed concern, saying, "If we make a mistake and get eliminated, the entire team could be replaced."

The industry also notes that, while the goal is to achieve 95% of the performance of leading global foundation models, it is difficult to evaluate teams by the same standard when the scale of the models, the nature of the data, and their objectives differ. An industry official commented, "Each team has its own strengths, and it is impossible to predict which will ultimately come out on top." Additionally, the conditions for securing graphics processing units (GPUs) have changed completely since the start of the project, leading to arguments that the elite-focused strategy should also be revised. With the United States' Nvidia agreeing to supply 250,000 advanced GPUs following the Asia-Pacific Economic Cooperation (APEC) summit earlier this month, the computational infrastructure environment has improved.

The Ministry of Science and ICT, which oversees the project, maintains that the overall framework-leaving two teams by 2027-remains unchanged. An official from the ministry stated, "The technologies developed by the participating companies could become important references in the future," adding, "We are considering ways to expand the market for these technologies and ensure that the technologies developed by the participating companies are put to appropriate use."