Korea narrows its AI model race to three, then plans to fund only two

South Korea's Ministry of Science and ICT said at a briefing at the Government Complex Seoul on the 18th that Upstage, SK Telecom and the LG AI Research Institute had passed the second-stage evaluation of its homegrown AI foundation model program, while Motif Technologies finished last of the four elite teams and was eliminated ^[1][2]. The field is now three ^[8], and according to the Seoul Economic Daily's account it was real-world usability rather than benchmark performance that decided the outcome, with the competitive arena widening into public administration, manufacturing, defense and education ^[3].

The number worth staring at is the user evaluation. It was worth 25 points, the overall average came in at 17.6, and the gap between first and fourth place was 5 points, the widest of the three evaluation categories ^[4]. That average is roughly 70 percent of the available marks ^[6], and the spread equals a fifth of the category's total weight ^[7]. Read plainly: nobody aced the part where people actually use the model, and the separation between teams showed up there rather than on a leaderboard.

Then the policy turn, which is the more consequential half. Vice Minister of Science and ICT Ryu Je-myung said the government is discussing a plan to completely overhaul its support approach, moving away from a method that selects only a few winners, and that after a third-stage evaluation two teams will be chosen as early as the end of this year ^[5]. Those two statements point in opposite directions at once: broaden support in principle, narrow from three funded teams to two in practice ^[12]. For operators, the near-term fact is that one of the three survivors loses its place within months ^[12], and that a state designation is not the same thing as a deployment inside the ministries, plants and defense programs the source frames as the expanding arena ^[3][5].

The capital backdrop in the same daily briefing is not flattering to sovereign-scale ambitions. Funds flowing into the global IPO market in the first half of this year reached $205.1 billion, three times the year-earlier level, while the number of listings rose just 8 percent to 689 and the U.S. share of inflows expanded to 75 percent from 44 percent ^[9]. SpaceX alone drew $86.25 billion after its June Nasdaq listing, 42 percent of the global IPO market ^[10]. European Central Bank economists warned the same day that a correction in U.S. tech stocks is inevitable ^[11]. Money is concentrating in a handful of very large assets; a national program that awards two slots is doing something structurally similar with a much smaller pot.

What to watch: the design of the third-stage evaluation, and specifically whether usability keeps or increases its 25-point weight now that it has proved to be the discriminating category ^[4]; which two of Upstage, SK Telecom and LG AI Research are selected, and whether the third retains any support once the promised overhaul is spelled out ^[5][12]; and whether any of the three can show paid, running deployments in public administration, manufacturing or defense rather than pilots ^[3].