Industry·5 min read·Implicator.ai

South Korea Gives All 52 Million Residents Free AI

Seoul picked SK Telecom, KT and Kakao to build a free, unlimited AI assistant for every resident — the first G20 nation to treat frontier AI as a public utility. At least half the model usage must be Korean.

AI FOR ALL 52M RESIDENTS · FREE BITSMINDS.COM
Share:

South Korea will give all 52 million of its residents free, unlimited access to a general-purpose AI assistant, making it the first G20 country to treat frontier AI as a universal public service rather than a consumer subscription. On 28 August the Ministry of Science and ICT named the three consortia that will build it, led by SK Telecom, KT and Kakao — the country's two largest telecom operators and the company behind its dominant messaging platform.

The bidding opened on 13 July, and the winning proposals differ mainly in where they meet people. SK Telecom plans to reach residents through an app, the web, phone and SMS. KT will fold the assistant into services people already use, including Daum search, shopping and housing platforms alongside its own channels. Kakao will run it inside KakaoTalk, handling reservations, applications and payments in the same thread where users message each other.

What separates this from a free chatbot tier is the plumbing. The services are designed to connect to government systems rather than answer questions in isolation, so a resident should be able to book a doctor's appointment, search apartment listings or get tax guidance without leaving the assistant. Small businesses will be able to calculate taxes and check whether they qualify for government support programmes, and parents will get recommendations for educational content. That turns the assistant into a front door for the state, which is a considerably harder engineering problem than serving tokens.

The government is supplying compute rather than writing a single large cheque. A total of 512 Nvidia B200 GPUs will be distributed across the three groups during 2026, with the top-ranked consortium receiving 256 and the other two taking 128 each; the ministry has not disclosed which proposal ranked first. Operators are required to match the government's GPU support with their own funding, so the public allocation sets a floor, not a ceiling.

The most consequential clause is the domestic-model quota. At least 50 percent of model usage must come from Korean foundation models that meet the government's independence standards, and at least 30 percent must come from domestic models developed outside the lead operator — a rule that forces the three winners to buy from each other's rivals rather than build closed stacks. Foreign models are permitted only for limited functions, and only without government funding. This is industrial policy wearing the clothes of a public service: Seoul is guaranteeing demand for a domestic model industry it has spent the past year trying to stand up, having already committed to an $880 billion decade-long push across AI, chips and robots.

Agreements with the operators are due in September, followed by beta testing, with a full launch targeted for December 2026. The programme runs through 2030, and from 2027 the government begins covering operating costs. That last detail is where the plan is thinnest — the total programme cost, the scale of future operating support and real-world capacity under nationwide demand have all been left unsettled, and industry figures have publicly questioned how token costs will be funded once usage is no longer a pilot. Rough capacity math from the top consortium's 256 B200s suggests roughly 51,000 simultaneous users, or about 18.4 million daily users if you assume four questions per person, 500-token answers and six times headroom for peak load — figures that hold only with heavily optimised models and short requests.

Korea is not building this into a vacuum. Roughly two-thirds of Koreans have already used AI, with about 23 million regular generative-AI users, and as of April 2026 ChatGPT led the market with 23.45 million users against 8.45 million for Gemini and 2.41 million for Claude. Every one of those is an American product. The assistant Seoul is about to hand out for free is, by design, mostly not — and the incumbents now have to compete with something that costs nothing and lives inside KakaoTalk.

Anthropic, for its part, has been building Korean ground presence since it opened a Seoul office with NAVER, Samsung SDS and LG CNS already shipping on Claude. Under the new quota rules, that kind of arrangement — foreign model, domestic wrapper — is exactly what does not qualify for the publicly funded build.

Want AI news before everyone else?

The morning's most important AI stories, straight to your inbox. No fluff.

Related Articles