Seoul's High-Stakes Battle for AI Sovereignty
South Korea's government has transformed the race for a homegrown AI foundation model into a nationally televised elimination contest. Dubbed the "AI Squid Game" by Bloomberg, the Ministry of Science and ICT's multi-stage competition pits the country's biggest conglomerates against scrappy startups in a fight for government-backed computing power and funding. The prize: building Korea's answer to ChatGPT and establishing Seoul as a credible third force in global AI.
The stakes extend far beyond technology. This competition sits within South Korea's broader strategy to commercialise AI across its industrial base, targeting 500 AI-powered factories by 2030. With recent wins from OpenAI's Stargate deal boosting Samsung and SK Hynix, Korea is positioning itself as Asia's AI manufacturing hub.
The Elimination Process Begins
Five consortia presented baseline models on 30 December 2025. Expert panels scored them on originality, benchmark accuracy, and compute efficiency. Two teams, led by Naver Cloud and NC AI, were eliminated on 15 January 2026. The survivors advanced to Phase Two with government funding and access to high-performance GPU clusters.
Then came the wildcard entry. In February, Seoul-based startup Motif Technologies joined the competition. The ministry allocated 768 Nvidia B200 processors to Motif's team, along with funds for data acquisition and processing. This move signalled the government's willingness to back newcomers over established players.
"South Korea is moving beyond the AI digital hype to launch an economic blueprint that puts intelligence directly onto the factory floor. By targeting the creation of 500 AI-powered factories by 2030, the nation is leveraging its industrial base to become the world's premier test bed for Physical AI." - Sean Kwon and J. James Kim, Stimson Centre
By The Numbers
- 100 trillion KRW: South Korea's planned investment over the next decade in AI semiconductors, data centres, and infrastructure
- 52,000 GPUs: Government target by 2028, scaling to 260,000 by 2030 through public-private investment
- 35.2% CAGR: Projected growth rate for South Korea's industrial AI market through 2035
- 16%: Share of South Koreans more concerned than excited about AI, the lowest of any country surveyed by Pew Research
- 37,000 GPUs: Accumulated by the government for national AI projects as of 2026
David Versus Goliath in Korean Tech
The competition's most fascinating dynamic isn't technical but structural. South Korea's economy has long been dominated by chaebol, the sprawling conglomerate groups like Samsung, LG, and SK that control vast industrial sectors. In AI, that dominance faces its strongest challenge yet.
Upstage, founded by former Naver researchers, has built a reputation for lightweight, efficient models designed for enterprise deployment. Motif Technologies is even newer, focusing on reasoning-first architectures. Neither has the resources of an LG or SK, but both bring agility that large organisations struggle to replicate.
The elimination of Naver Cloud, Korea's closest equivalent to Google, sent a clear message: incumbency means nothing here. Performance and innovation determine survival.
"At 16%, South Korea has the lowest percentage of people who are more concerned than excited about AI of any country surveyed. That public buy-in gives policymakers room to move aggressively." - Pew Research Centre global AI sentiment survey
Korea's AI Infrastructure Push
The competition operates within a broader national strategy called the AI Highway. Under this framework, the government is building a National AI Computing Centre with more than 15,000 GPUs by 2028. Priority sectors include manufacturing, telecommunications, and healthcare.
The infrastructure investment coincides with Korea's dominance in AI memory chips, creating a vertically integrated advantage. Korean manufacturers can access locally-built models running on Korean-made hardware, keeping value chains onshore.
- Manufacturing focus: 500 AI-powered factories by 2030 across automotive, shipbuilding, and electronics
- Telecommunications integration: SK Telecom testing 5G-enabled AI applications for smart cities
- Healthcare applications: LG AI Research developing diagnostic models for Korea's ageing population
- Cultural localisation: Models trained on Korean language data and regulatory contexts
- Open-source commitment: All winning models released for domestic companies and researchers
| Team | Type | Focus Area | Status |
|---|---|---|---|
| LG AI Research | Conglomerate | Multimodal enterprise AI | Phase Two |
| SK Telecom | Conglomerate | Telecoms and infrastructure AI | Phase Two |
| Upstage | Startup | Lightweight enterprise models | Phase Two |
| Motif Technologies | Startup (wildcard) | Reasoning-first architectures | Phase Two |
| Naver Cloud | Tech platform | Search and language models | Eliminated |
| NC AI | Gaming tech | Game AI and NLP | Eliminated |
The Realism Behind the Hype
Sceptics question whether compute alone builds frontier models. OpenAI and Google DeepMind have thousands of researchers and years of institutional knowledge. South Korea's total AI R&D budget, while growing at 19.5% year-on-year, represents a fraction of what a single US lab spends annually.
But the bet isn't about matching GPT-5. It's about building models good enough for Korean enterprise use cases while keeping economic value onshore. If Korean manufacturers adopt Korean-built models for quality control and predictive maintenance, the multiplier effects could be substantial.
The Bank of Korea has raised its 2026 GDP growth forecast to 2%, up from 1.8%, partly on AI and semiconductor momentum. The connection between AI adoption in daily life and economic transformation is becoming clear.
Why did Naver Cloud get eliminated?
Expert panels scored on originality, benchmark accuracy, and compute efficiency. Naver's HyperCLOVA models are established but may have scored lower on originality, given their similarity to existing architectures. The ministry hasn't published detailed scoring criteria.
Could the winning model actually rival ChatGPT?
Not in general-purpose conversation initially, but potentially in Korean-specific applications. The focus is enterprise deployment rather than consumer chatbots, playing to Korea's manufacturing strengths over Silicon Valley's consumer orientation.
What happens to eliminated teams?
Eliminated teams lose government funding and GPU access but can continue development independently. Some may pivot to specialised applications or seek private investment based on their competition performance.
How does this affect Korea's tech giants?
Mixed results. Samsung benefits from increased chip demand, but traditional software leaders like Naver face new competition. The shake-up could accelerate innovation across the sector as companies adapt to government priorities.
Will other countries copy this model?
Several nations are watching closely. The competition format generates public engagement around industrial policy, making it politically viable. Similar programmes may emerge in Japan, Singapore, and other tech-focused economies.
South Korea's AI competition reveals how nations can leverage cultural moments to drive economic strategy. By gamifying industrial policy, Seoul has created public engagement around sovereign AI that traditional procurement processes never achieve. As AI development shifts globally, other countries may find Korea's theatrical approach more effective than Silicon Valley's venture capital model.
Will Korea's AI Squid Game produce world-beating models or just world-class theatre? The answer may determine whether small nations can compete in the AI age. Drop your take in the comments below.
No comments yet. Be the first to share your thoughts!
Leave a Comment