How Korea's AI Action Plan Actually Works

What Korea's AI Action Plan Actually Says

In February 2026, the National AI Strategy Committee, chaired directly by President Lee Jae-myung, finalised the Republic of Korea Artificial Intelligence Action Plan for 2026 to 2028. It is the country's most ambitious sectoral plan since the Korean Wave cultural strategy of the late 2010s, and it is structured for a different audience. Rather than industrial policy framed as soft power, this plan reads as procurement, infrastructure, and rules.

This explainer walks through what the 99 tasks really commit Korea to, where the budget actually lands, and which parts will move first. We also call out where the plan papers over genuine tensions, particularly between the Ministry of Science and ICT and the Korea Communications Commission.

The Five Pillars: What Each One Funds

The plan has five pillars. They are not given equal weight, and the differences matter:

• AI Compute (USD 4.2 billion): A sovereign GPU cluster operated by the Ministry of Science and ICT, plus subsidised access for SMEs through the Korea Information Society Development Institute.
• Data and Open Models (USD 1.8 billion): Korean-language model training corpora, data trust frameworks, and an open-source release schedule for LG, Naver, Kakao, and KT models.
• Talent (USD 2.4 billion): 100,000 AI specialists by 2027, with explicit quota for women in AI engineering and structural support for KAIST and POSTECH AI graduate programmes.
• Industrial Adoption (USD 4.7 billion): Sector-specific procurement-pull. The largest sub-allocation is to manufacturing (Hyundai, POSCO, LG Energy Solution), followed by healthcare, finance, and agriculture.
• AI Rules (USD 1.5 billion): Implementation of the Korean AI Basic Act, sandboxes, certification labs, and international cooperation budgets.

Why Compute Is The Most Concrete Pillar

The sovereign cluster is the closest the plan gets to a single visible commitment. The flagship build, sited in Sejong with a satellite at Pohang, will host roughly 12,000 Nvidia GPUs in its first phase, with capacity to expand to 30,000. The first phase comes online in 2027, and access will be brokered by the Korea Information Society Development Institute through a tariff system that prioritises SMEs and academic users over chaebol-affiliated buyers.

This is unusually direct for Korean industrial policy. The country normally subsidises through tax incentives and R&D grants. The decision to operate compute as a state-managed resource is partly a response to last year's frustration that GPU capacity was concentrated in three or four companies, and partly a hedge against US export controls that have complicated direct purchases.

Where The Plan Is Quieter Than It Looks

The rules pillar is where the politics live. Korea's AI Basic Act, passed in late 2024 and effective from January 2026, set out a tiered risk framework similar to the EU's. The Action Plan funds implementation of that framework, but it is silent on the question of which agency owns enforcement when AI systems touch personal data.

The Ministry of Science and ICT runs the AI rules workstream. The Personal Information Protection Commission (PIPC) owns the data privacy regime. The Korea Communications Commission (KCC) runs broadcasting and platform regulation. All three claim partial jurisdiction over generative AI systems used in consumer products. The Action Plan papers over this with a coordination mechanism that lacks a deciding vote.

The hardest fight will not be between Korea and Brussels. It will be between PIPC and the Ministry of Science and ICT, and that fight is being deferred, not resolved.

Comparing The Plan To Asia's Other National Strategies

The table makes Korea look unusually well-funded. That is broadly accurate, but with a caveat. India and Singapore are leaner because their model is to attract private capital rather than match it directly. Japan's ABCI 3.0 is technically larger in raw FLOPS than Korea's Sejong cluster will be in 2027.

What Actually Moves In The Next 90 Days

1. The Sejong sovereign cluster procurement, with an RFP expected by end-Q2 2026.
2. The release of a publicly funded Korean-language base model, expected at the Korea AI Forum in October.
3. The launch of the AI export support programme, with USD 600 million set aside to anchor Korean AI vendors entering ASEAN markets.
4. A formal call from PIPC and the Ministry of Science and ICT for industry comments on the implementing rules of the AI Basic Act.
5. A KAIST-led AI fellowship for 1,200 mid-career engineers, structured around six-month placements at chaebol AI labs.

The other 78 tasks span the longer 2027-2028 horizon and will move more slowly. Several depend on whether the National Assembly approves the budget tranche due in December.

Where International Investors Should Pay Attention

For outside investors, the most useful signal in the plan is the procurement-pull pillar. Korea is committing real money to buy from domestic AI vendors, and that creates near-term pull for Naver, Kakao Brain, Upstage, and a handful of younger players. International firms can compete in some lots, particularly the cluster build and certification work, but the sector-specific industrial pillar is structured to favour Korean integrators. That contrasts with India's IndiaAI mission and Singapore's model governance work, which both lean more open to foreign vendors.