South Korea Gives 52 Million Citizens Free AI Access

A Whole Country Gets an AI Agent

South Korea's Ministry of Science and ICT announced on August 28, 2026 that it has selected three industry consortiums to build free, general-purpose AI services for the entire population. SK Telecom, KT and Kakao each lead one group, chosen from six bidders. Beta testing begins in September, with the full service due by the end of December.

The programme is called AI for All, and the design goal is worth pausing on: not a chatbot demo, but AI agents that can complete real tasks on a citizen's behalf across public services, healthcare, finance, education and housing.

• Three consortiums selected — led by SK Telecom, KT and Kakao — from six applicants, with agreements to be signed in September 2026
• 512 Nvidia B200 GPUs will be split among the three groups for 2026, with the government subsidising nationwide service costs from 2027
• Beta in September, full launch by December 2026, free to residents at the point of use
• Agent-first requirement: the brief demands services that carry out reservations, applications and payments, not just answer questions

What Does "Free AI for Everyone" Actually Cover?

South Korea's population is roughly 52 million, and the stated ambition is an AI-basic society reachable regardless of income, age or digital literacy. That framing matters more than it might sound. The gap that shows up in every AI adoption study is not between people who have heard of these tools and people who have not — it is between people who pay a monthly subscription for the good ones and people who do not.

Reported details from outlets covering the launch describe access without the usage caps that normally define a free tier. That specific claim comes from press reporting rather than a ministry document, so treat the exact limits as unconfirmed until the September beta terms are published. The direction of travel, though, is not ambiguous.

The practical examples cited in coverage are refreshingly mundane: booking a medical appointment, searching rental listings, working through tax guidance, checking whether a small business qualifies for a support programme. These are the tasks where an agent that can actually complete a form is worth considerably more than one that can describe how to fill it in.

How the Three Consortiums Differ

Each group is building on its own distribution advantage, which is the sensible way to run a three-horse public programme.

SK Telecom is pitching an action-oriented assistant reachable through apps, the web, phone calls and text messages — including AI-placed phone calls to businesses and government offices that have no digital front door at all. Its consortium includes Shinhan Card, Hana Card, Tmap Mobility and the search startup Liner.

Kakao routes through KakaoTalk, the messaging app that already sits on essentially every phone in the country, pairing a general chatbot with agents that handle bookings and payments. Its partners include Lunit, Seoul National University Hospital, Shinhan Bank and several LG AI units alongside the chip designer FuriosaAI.

KT connects its service to partner platforms including Daum, Danawa, Musinsa and Zigbang, plus its own telecom and IPTV channels, with Rebellions and Upstage AI supplying silicon and models.

Notably, Naver did not participate, citing the timeline and the requirement to support multiple models — a reminder that a national programme's constraints are real engineering constraints for the companies inside it.

Why This Is the Most Interesting AI Story of the Week

Most of what we cover in AI is a capability story: a model gets better, cheaper, or smaller. This is a distribution story, and distribution is where capability turns into consequence.

There is also a supply-side reading. The 512 B200 allocation is a modest number by hyperscaler standards, and it forces the consortiums toward efficiency — smaller models, aggressive caching, careful routing — rather than throwing frontier-scale compute at every query. That is a useful constraint, and the engineering answers to it tend to be reusable. South Korea has been building toward this for a while; we covered NAVER and Nvidia scaling sovereign AI to gigawatt capacity earlier this year, and Anthropic opening a Seoul office shortly after.

The thing to watch is not the launch date. It is the first honest usage report in mid-2027 — how many of those 52 million people use an agent more than twice, and for what. That number will tell the rest of the world considerably more about AI adoption than another benchmark chart.

Sources: The Korea Herald — August 28, 2026; The Korea Times — August 28, 2026; TechSpot — August 31, 2026.