South Korea will offer free, unlimited generative AI access to every citizen through three telecom-led consortia.

The rules say that at least 80% of the service has to run on Korean models, not on AI systems from the U.S. or China.

The move places South Korea among a growing set of governments pursuing “sovereign AI” — the idea that nations need domestic models and infrastructure to control data, language coverage, and public-sector use of the technology. Countries including France and the United Arab Emirates have backed national models and compute programs for similar reasons.

SK Telecom will route the assistant into Government24 and the PASS app

The Ministry of Science and ICT named SK Telecom, Kakao, and KT as the operators, with beta testing set to begin in the coming weeks.

The “AI for All” program covers all 51 million people. Each selected provider must obtain at least half of its service from South Korean sovereign foundation models.

Another 30% must come from local models made by companies other than the operator, a clause that prevents any company from relying entirely on its own technology. That requirement effectively spreads demand across the domestic model ecosystem, which includes players such as Naver, whose Korean-language models are among the country’s most established, alongside the telecoms’ own systems.

The ministry said foreign models are allowed only where it considers them minimally necessary, and the government will not subsidize that segment.

Seoul launched the competition in July, and the three winners were selected from six applicants. The lineup places the country’s two largest telecom operators alongside Kakao, whose messaging app is a default utility of Korean daily life.

Each consortium will release a standalone app and embed the assistant into products people already use every day.

SK Telecom has been the most explicit about its rollout. The company’s A.X K2 foundation model and other Korean systems will power its service, which will be delivered to users through an app, phone calls, and text messages. That approach is designed for older people who are less familiar with apps — a meaningful segment in one of the world’s most rapidly aging societies.

The company said it will integrate the assistant with Government24 and the PASS identity app so users can request 83 types of official certificates.

It said 20 organizations are directly involved, with another 11 signed up through memoranda of understanding, for a total group of 31.

Seoul hands over 512 Nvidia B200 chips and no cost estimate

In 2026, Seoul will provide up to 512 of Nvidia’s top-of-the-line B200 graphics processors to the three groups and help cover operating costs. There will be no token limits for AI usage.

The funding model changes in 2027. From then on, the government intends to pay the public-service costs from the national budget while requiring the operators to build their own revenue sources to ensure quality over time.

The ministry has not said what the overall cost of the program will be.

In July, Cryptopolitan reported that Seoul is building a sovereign wealth fund of more than 20 trillion won, or about $14 billion, for AI and strategic industries in 2027.

A June report from the Carnegie Endowment for International Peace found that the country has no major publicly known operational AI chip clusters, even as SK Hynix and Samsung supply the high-bandwidth memory inside much of the world’s AI hardware. That gap between South Korea’s strength in AI hardware components and its limited domestic compute capacity is part of the backdrop for the chip handout and the broader state investment.

During Jensen Huang’s June visit to Seoul, Nvidia struck partnerships with SK Hynix, Naver, and SK Telecom. Cryptopolitan reported that SK Telecom plans to build a gigawatt-class AI cloud on Nvidia’s DSX system starting in 2027.

About two-thirds of South Koreans have tried AI, and about 23 million use generative tools regularly. Beta testing could begin as early as the end of September, though SK Telecom now points to October for its own launch, with a full rollout targeted before year-end.