# South Korea’s Sovereign AI for All Plan > Source: > Published: 2026-09-05 23:00:00+00:00 [← All news](/news/) Most sovereign-AI programmes concentrate on the supply side: models, chips, data centres and research funding. South Korea is adding the missing last mile. Its government wants the public to use domestic AI in everyday life, and it is prepared to subsidise the compute needed to make that happen. The country’s [AI for All programme](https://www.msit.go.kr/eng/bbs/view.do?bbsSeqNo=42&mId=4&mPid=2&nttSeqNo=1285&sCode=eng) is expected to begin beta testing before launching nationally by the end of 2026. The ambition after 2027 is “one AI agent per citizen”. That makes this more than another model announcement. It is an attempt to turn sovereign AI into public infrastructure. ## What South Korea has announced The Ministry of Science and ICT has selected three consortiums led by SK Telecom, Kakao and KT. They are developing a general-purpose chatbot alongside agents that can help people find and use public services. The plans extend beyond question answering. SK Telecom is developing an action-oriented service; Kakao plans to put reservations, applications and payments into its messenger platform; and KT is concentrating on public and everyday services. Access is expected through apps and the web as well as familiar channels such as messages and phone calls. That matters for citizens who would never seek out a standalone AI product. The government will provide the operators with a combined 512 Nvidia B200 GPUs during 2026. From 2027, public funding will help cover the cost of serving the population. The project therefore subsidises both the infrastructure and the route to market. ## Sovereign AI is the point The programme’s model requirements are unusually explicit. At least 50% of each service must use a Korean model that meets the government’s sovereign-foundation-model standard. A further 30% must use domestic models developed by other Korean companies. Foreign models are allowed only where necessary, and the government will not subsidise that part of the service. South Korea is not trying to isolate itself from the global AI market. It is using public spending to ensure that domestic models have users, distribution and feedback rather than remaining technical demonstrations. This changes the usual sovereign-AI equation: Domestic compute + domestic models + national distribution = a viable sovereign AI market. Building a model is only the first step. Without adoption, local developers receive less real-world feedback and the state remains dependent on overseas products. AI for All is designed to create demand at the same time as capability. ## South Korea is already AI-forward This is not a programme aimed at a population unfamiliar with AI. A 2025 user-panel survey from the Korea Media and Communications Commission and KISDI found that [38.9% of respondents had used generative AI](https://www.kmcc.go.kr/user.do?boardId=1058&boardSeq=68870&cp=2&dc=E04010000&mode=view&page=E04010000), up 14.9 percentage points in a year. Government figures also estimate 23.45 million monthly active ChatGPT users in April 2026, alongside 8.45 million for Gemini and 2.41 million for Claude. Those audiences overlap, so they cannot be added together, but they show the scale of existing demand—and the extent of Korea’s current reliance on foreign platforms. **A figure worth correcting:** some coverage says one-quarter of Korean citizens already pay for AI. The underlying Korea Consumer Agency survey found that 1,101 of 10,000 respondents used generative AI and 268 of those users paid. The reported 24.3% is the share of AI users, not the population; paying users represented about 2.7% of all respondents. The surveys use different samples and questions, so they should not be merged into a single adoption estimate. The defensible conclusion is simpler: use is growing quickly, while paid access remains much narrower than the headlines suggest. ## Is South Korea the first country to do this? The [Wall Street Journal described AI for All](https://www.wsj.com/tech/ai/south-koreas-ai-for-all-push-gives-free-access-to-every-citizen-451f6b2c) as the first major state-led offering to treat generative AI like a public utility. Other reports have shortened that to “the first country to offer AI as a public utility”. The narrower wording is safer. Governments already provide public chatbots, digital assistants and subsidised access in different forms, and there is no authoritative global register against which to prove an absolute first. What is unusual is the combination: nationwide eligibility, no stated usage cap, several private operators, sovereign-model requirements and integration with government services. If delivered as announced, it may be the first national AI service with that breadth. For now, it remains a programme moving towards launch rather than an established universal service. ## The real test starts after launch “Free and unlimited” will attract attention, but it does not tell us what level of model capability or service quality citizens will receive when demand peaks. A pool of 512 B200 GPUs is meaningful launch support; it is not unlimited compute. Performance, queues and the eventual public cost will show whether the offer can operate at national scale. Data governance is the more consequential issue. The ministry’s own project material envisages operators building revenue models that may use prompt data. That could help domestic services improve, but prompts sent to agents handling benefits, tax, health appointments or payments may contain sensitive information. Citizens will need clear answers about consent, retention, model training and who is accountable when an agent acts incorrectly. There is also a public-service design question. A chatbot that explains a form is low risk. An agent that submits an application or makes a payment has crossed into execution. Identity, authorisation, audit trails and routes to human appeal have to mature alongside the interface. ## Why this matters beyond Korea South Korea is testing a version of sovereign AI that other governments can understand immediately. Rather than judging success by benchmark scores or the number of GPUs purchased, it asks whether domestic AI becomes useful enough for citizens to adopt. That has industrial-policy advantages. Public distribution gives local providers customers and feedback. Integration with government services creates uses foreign general-purpose chatbots cannot easily reproduce. Subsidised inference narrows the gap between a domestic model and a heavily capitalised US or Chinese platform. It also raises the stakes. If the programme works, South Korea will have shown that sovereign AI can be built as a national service, not just a national asset. If it fails on trust, quality or cost, it will demonstrate why access alone is not enough. **Sources.** This article uses announcements from South Korea’s [Ministry of Science and ICT](https://www.msit.go.kr/eng/bbs/view.do?bbsSeqNo=42&mId=4&mPid=2&nttSeqNo=1285&sCode=eng), its [2026 policy programme](https://www.msit.go.kr/eng/bbs/view.do?bbsSeqNo=42&mId=4&mPid=2&nttSeqNo=1284&sCode=eng), the [Korea Times](https://www.koreatimes.co.kr/business/tech-science/20260904/korea-to-provide-free-ai-agents-for-all-citizens-through-messenger-texts), the [Korea Consumer Agency survey reporting](https://en.sedaily.com/society/2026/05/29/one-in-four-ai-users-pay-for-subscriptions-led-by-those-in) and the original [TechSpot report](https://www.techspot.com/news/113664-south-korea-giving-entire-population-free-access-ai.html). *New AI policy analysis lands here first. [Sign up for email updates](/contact/).* [← All news](/news/)