Korea’s National AI Computing Center and Sovereign AI Plan, Explained

This guide is for expats, investors, and Korea-watchers who want to understand what the South Korean government is building when it talks about a “national AI computing center” and “sovereign AI.” The short answer: Korea is investing more than 10 trillion won (roughly $7.3 billion) in 2026 alone to build domestic AI computing infrastructure and fund homegrown large language models (LLMs — the technology behind tools like ChatGPT), with the explicit goal of joining the world’s top three AI powers.

How It Works

The National AI Computing Center

Korea does not have one single building called the National AI Computing Center. Instead, the Ministry of Science and ICT (MSIT — the government department that oversees technology policy) is distributing a large pool of high-performance GPUs (graphics processing units, the specialist chips that train AI models) across three domestic cloud providers: Naver Cloud, NHN Cloud, and Kakao. These three companies act as the physical hosts of the national computing pool.

In July 2025, MSIT announced it would invest approximately 1.46 trillion won (around $1.1 billion) to procure 13,000 high-performance NVIDIA GPUs as the first tranche of the center. Startups, research institutes, and companies can apply to rent time on this infrastructure at subsidized rates, rather than buying chips themselves or paying full price to foreign cloud providers.

The government’s longer-term GPU targets are 52,000 units by 2028 and 260,000 units by 2030, to be reached through a mix of public procurement and private co-investment.

The Sovereign AI Foundation Model Program

Separately from the computing infrastructure, MSIT launched the Sovereign AI Foundation Model Project in mid-2025. The goal is to produce a Korean-language AI foundation model (a large, general-purpose AI trained from scratch, comparable in scale to GPT-4 or Gemini) that is built and owned domestically.

Rather than picking one winner upfront, the government ran an open competition. Fifteen consortia applied. In August 2025, five were selected: teams led by Naver Cloud, LG AI Research, SK Telecom, NC AI (the AI arm of game company NCSoft), and Upstage (a Korean AI startup). The government allocated approximately $381 million to fund these teams collectively.

The selection process is structured as a tournament. Teams are evaluated every six months. At each stage, lower-scoring teams are eliminated and their allocated resources — compute time and funding — are redistributed to the remaining teams. The first evaluation, held in late December 2025, eliminated Naver Cloud and NC AI. MSIT said Naver Cloud’s entry relied too heavily on fine-tuning existing overseas models rather than training independently. By 2027, two finalist teams are expected to be designated as Korea’s national AI champions.

As of the first-round results (January 2026), three teams remain: LG AI Research (which ranked first across all evaluation categories), SK Telecom, and Upstage.

Legal Framework

Underpinning both programs is the AI Framework Act (formally: the Framework Act on the Development of Artificial Intelligence and Establishment of Trust), signed into law on January 21, 2025 and effective from January 22, 2026. It makes Korea the second country in the world, after the EU, to pass comprehensive national AI legislation. The Act requires transparency from AI developers, mandates impact assessments for “high-impact” AI in sectors like healthcare and public services, and creates legal hooks for the government’s AI investment programs.

Why Korea Does It This Way

Korea’s economy runs on semiconductors, displays, and industrial exports. The concern inside government is straightforward: if AI becomes the key input to manufacturing, logistics, finance, and healthcare — and if the underlying models and chips are built abroad — Korea becomes dependent on foreign infrastructure in the same way a country without energy resources depends on imported oil.

Korea also has a specific language disadvantage. Because Korean is not as widely represented in global AI training data as English or Chinese, general-purpose models from US or Chinese developers tend to perform worse in Korean. A domestically trained model using Korean text and cultural context is expected to outperform adapted foreign models on Korean-language tasks.

The tournament structure for the foundation model competition reflects a deliberate policy choice: rather than funding a single state champion (which risks backing the wrong technology), the government funds multiple teams in parallel and lets performance data decide who continues. This approach borrows from how some countries have handled semiconductor R&D consortia.

The Numbers That Matter

What It Means for You

For workers and researchers

If you work in AI, data science, or related technical fields, the national GPU pool is available to apply for. Startups and university research groups can request subsidized compute time through MSIT’s allocation process. This lowers the cost barrier for training large models in Korea significantly compared to renting from US or Chinese hyperscalers (large-scale cloud providers).

The three surviving foundation model teams — LG AI Research, SK Telecom, and Upstage — are all expected to grow headcount through 2027. Job postings in LLM pre-training, evaluation, and Korean NLP (natural language processing) have increased at all three organizations.

For businesses and investors

The AI Framework Act introduces new compliance obligations starting January 2026. Companies deploying AI in high-impact sectors (healthcare, education, finance, critical infrastructure) must conduct impact assessments and maintain transparency documentation. Foreign operators using AI services in Korea must designate a domestic representative. If your business uses AI tools in any of these sectors, check MSIT’s guidance on whether your systems qualify as “high-impact” under the Act.

The government’s stated preference for domestic cloud providers (Naver Cloud, NHN Cloud, Kakao) as the hosts of national AI infrastructure signals continued policy support for Korean tech incumbents. This is a structural factor, not an investment recommendation.

For foreign residents

The immediate practical impact is limited. The sovereign AI push is aimed primarily at enterprise and research use cases. However, Korean-language AI tools — in customer service, public services, and healthcare — are likely to improve as the national model competition raises the quality of Korean LLMs available to product developers. Government digital services may also begin integrating AI tools built on the national model infrastructure.

FAQ

Does Korea own the NVIDIA chips it is buying?

The government is procuring chips through contracts with domestic cloud providers (Naver Cloud, NHN Cloud, Kakao), not buying chips directly in all cases. Some GPUs are purchased outright as public assets; others are leased. The national computing pool is operated by these private companies under government contract, with access governed by MSIT’s allocation rules. Ownership structures vary by tranche and have been subject to ongoing negotiation.

Why were Naver Cloud and NC AI eliminated from the sovereign model contest?

MSIT’s evaluation board found that Naver Cloud’s submitted model relied significantly on fine-tuning an existing overseas model rather than being trained independently from scratch — a core requirement of the program. NC AI did not meet technical performance benchmarks. Both eliminations were announced by MSIT in January 2026 after the first six-month evaluation round.

Is this related to the US–China chip restrictions?

The geopolitical context is relevant but the government has not publicly framed the program as a direct response to US export controls on chips to China. Korea is a US ally and currently receives NVIDIA chips without the same restrictions applied to China. The sovereign AI drive is officially motivated by economic competitiveness and linguistic self-sufficiency. That said, several policy analysts — including researchers at the Carnegie Endowment for International Peace — have noted that reducing dependence on single foreign suppliers is an implicit goal of the infrastructure buildup. Check MSIT’s official policy documents at msit.go.kr for the government’s stated rationale.

Browse all Le Won guides