South Korea is turning its AI ambitions into a full-blown public competition — and it’s not playing around.
Late last year, more than a thousand people packed into a convention center in Seoul to watch some of the country’s top engineers show off their latest work in artificial intelligence. The atmosphere sounded less like an academic conference and more like a sporting event. Thousands more tuned in online, and the crowd reportedly cheered as each team took the stage, knowing the stakes weren’t symbolic. For some contestants, it could be their last time in the spotlight.
The event is part of a government-backed tournament that’s already being called the “AI Squid Game,” a nod to the Netflix series where every round ends with someone getting eliminated. And in this case, the comparison isn’t just for dramatic effect. The contest is built around a ruthless cycle: teams develop their own AI foundation models, then every six months those models are evaluated by judges under the Ministry of Science and ICT. Winners advance. Losers get cut.
This week, that cut came faster and harder than expected. Judges eliminated the AI unit of Naver Corp. after criticizing it for relying too much on foreign technology — a major violation in a competition designed to prove Korea can build its own models from scratch. They also dismissed NCSoft’s AI subsidiary, which stood out as the only team led by a woman, after its model scored poorly in performance evaluations. The surprise was that only one team was expected to be eliminated, but two were shown the door.
The teams still standing include subsidiaries of LG and SK Group, along with Upstage, a five-year-old AI startup that’s aiming for an IPO. The science ministry also said it will add one more team in the next stage. Two winners are expected to be named in early 2027, and for the companies still competing, the prize is bigger than bragging rights.
The government is giving participants access to GPUs and datasets — support that can be make-or-break in the foundation model era — and teams that get eliminated lose that access. That’s a brutal incentive structure, but it’s also the whole point. Korea wants speed. It wants focus. And it wants to force its biggest companies and rising startups into a pressure cooker that produces something globally competitive.
Officials have been clear about what they’re chasing: homegrown open-source models that can stand alongside frontier systems like ChatGPT and China’s DeepSeek. It’s a bold goal, especially in a world where the AI conversation is dominated by U.S. and Chinese giants. But Korea sees itself as uniquely positioned to compete, particularly because it already punches above its weight in the pieces that matter: memory chips, industrial hardware, cloud infrastructure, and advanced applications.
Science and ICT Minister Bae Kyung-hoon, appointed in July, has described the country as a “full-stack AI country,” and argues it has a genuine shot at becoming a serious global player. Korea’s leaders don’t want to be “No. 3” in the AI race by default — they want to be the country that can challenge the top two.
This push also reflects how urgently South Korea is taking the AI shift. The country is already one of the world’s biggest markets for ChatGPT when you look at paying users. It also has the highest density of industrial robots anywhere — meaning AI won’t just change apps and chatbots there, it could reshape manufacturing, logistics, and the wider economy.
But even in a competition designed to reward local innovation, defining “homegrown AI” has turned into its own battle. The government wants models trained end-to-end with proprietary learning processes, data, and architecture — not systems stitched together with outside shortcuts. That requirement has sparked controversy, because in modern AI, building entirely from zero is rare. Many teams rely on public research and widely shared architectural ideas, especially when the cost of starting from scratch is nearly impossible to justify.
That tension exploded into public accusations. A rival company claimed Upstage’s Solar Open model looked too similar to a Chinese model called GLM-4.5-Air. Upstage denied it, held a public verification session, and forced the accuser to apologize — but the scrutiny didn’t stop there. Naver Cloud faced criticism for using weights and design elements that resembled architectures from Chinese labs like Alibaba, and SK Telecom also came under questions about whether its model resembled another Chinese system.
In the end, the government drew a hard line: it said Naver Cloud crossed it by partially using an existing Chinese model and was therefore eliminated. It also said it saw no reason to disqualify SK Telecom or Upstage. Naver, for its part, said it respected the decision.
Meanwhile, the competition itself is taking a toll. Engineers have been working around the clock, combing through datasets, tweaking training runs, chasing performance gains — the kind of repetitive, exhausting work that doesn’t look glamorous from the outside but makes or breaks foundation models. One LG research leader said he’s been working so late that he hasn’t seen his toddler awake in months. SK Telecom’s foundation model lead joked that he’s gone gray during the contest and even missed helping with his family’s annual harvest.
But despite the stress, there’s also a real sense of pride — like they’re part of a national project. Upstage’s CEO compared the competition to the intense audition shows that helped launch K-pop, saying this kind of pressure-cooker environment is exactly how Korea has succeeded before.
That historical echo is intentional. South Korea has a long track record of using state-backed strategy to create new industries, from highways to broadband to global manufacturing. President Lee Jae Myung is now framing AI as the next leap, promising an “AI Highway” built around compute infrastructure, data, and eventually 6G rollout.
The “AI Squid Game” may look like a spectacle on the surface, but it’s not entertainment. It’s a national strategy — one designed to pick winners, accelerate progress, and build a homegrown AI ecosystem that doesn’t have to borrow its future from the U.S. or China.
This analysis is based on reporting from financialpost.
Image courtesy of Unsplash.
This article was generated with AI assistance and reviewed for accuracy and quality.
