Free AI for a nation · Part 1 of 7
Budget per citizen is not cost per active user
Spread over every resident, the proposed 2027 budget for free AI is a few thousand won a year. That is the fiscal burden of making it available to everyone, not what the service costs for each person who uses it.
The 2027 budget proposal has a new line for AI for All: 250 billion won ($186 million), before the National Assembly.[11] Over Korea’s 51.82 million residents, that is 4,824 won ($3.58) a year each.[23] As the fiscal burden of making the service available to everyone, it is the right number. The register counts foreign residents too; the ministry says “all citizens”, and I haven’t checked who is eligible.
It is not the cost per user, and the other denominators have their own limits. In the 2025 Internet Usage Survey, 44.5% of Koreans had ever used generative AI; over them the line is about 11,000 won each, but people who once tried a chatbot are not users of this service.[77] The three consortia each aim for 5 million monthly users by the end of 2026, 15 million together: 16,667 won per targeted user.[11] Those targets can overlap, since one person can use two of the services; they are for the end of 2026, not an average over the year the line pays for; and they are targets. How often people will use the service, nobody has measured.
Price is not what it competes on. ChatGPT’s free tier has had no cap on text messages since August 2026, by press reports, and Korean services such as Wrtn and SK Telecom’s A. are free too.[67, 72, 78] What the frontier products sell is the harness around the model: memory, tools, apps, agents. My reading is that a programme that buys GPUs and mandates models is competing on the thing it did not buy.
Most of the line, 170 billion won ($126 million), leases about 2,000 B200s for a year; each consortium gets 10 billion won ($7.43 million), and 50 billion won ($37.1 million) goes to specialised agents for finance, medicine and public services.[11] GPUs cost the same whether a few people use them or millions do, so the cost per resident is fixed by the proposal and the cost per user falls as use rises. That is a reason to measure use, not to call non-users expensive: a reserve is also what peaks and a service guarantee need (chapter 2), and more use is only worth more if it is useful (chapter 6).