Recently, South Korea officially launched a nationwide AI initiative: providing free and unlimited AI access to all citizens.
With a population of 51 million, 23.45 million are using ChatGPT, a US product that covers nearly half the nation.
The South Korean government's response was to fund it themselves: On August 28th, the Ministry of Science and ICT announced that three major consortia—SK Telecom, KT, and Kakao—were selected for the 'AI for All' project. By the end of this year, they will provide free, unlimited AI services to all citizens, with the condition that 80% must run on domestic Korean models.
This is the first G20 economy to treat AI as public infrastructure.
What Should Countries Outside the US and China Do?
AI model capabilities are concentrated in the United States and China; other countries are consumers.
According to Wiseapp data from April 2026, ChatGPT's monthly active users in Korea were 23.45 million, Gemini had 8.45 million, and Claude had 2.41 million—the top three are all foreign products.
South Korea has models like LG EXAONE, SKT A.X K2, and Upstage Solar, but their user bases are orders of magnitude smaller.
https://www.wiseapp.co.kr/insight/detail/1033/fastest-growing-and-most-used-mobile-apps-in-korea
Domestic models need real users to iterate, but users won't migrate from ChatGPT to a weaker product. Data continues to accumulate on foreign companies' servers.
The market alone cannot break this cycle.
The South Korean government has chosen a globally unprecedented path: instead of competing head-on on model performance, they are using public funds to control the distribution end.
The "AI Basic Act," which took effect in January 2026, provides the legal framework for government intervention in the AI industry. The "AI Action Plan 2026-2028," finalized in February of the same year, further proposes the concept of an "AI Fundamental Society," positioning AI access as a basic citizen right on par with education.
Deputy Prime Minister and Minister of Economy and Finance Choo Kyung-ho called this project the "calculator and computer of the AI era."
The proliferation of calculators relied on unified procurement by schools and government agencies.
Exchanging Distribution Rights for Survival Rights: A New Industrial Policy
The commercial logic of "AI for All" can be summarized in one sentence: The government provides funding, computing power, and access points, channeling 51 million people into the domestic AI ecosystem, using real traffic to nurture and strengthen domestic models.
This year, the government allocated a total of 512 NVIDIA B200 GPUs to the three consortia (one roughly equivalent to 3-4 H200s in inference power). Starting in 2027, operating costs will be covered by the national budget, with the free service guaranteed to last at least until 2028.
But money and computing power are just entry barriers; the real ecosystem shaper is the rule: each consortium must run at least 50% of its services on self-developed sovereign foundation models, with another 30% running on other Korean companies' models, totaling an 80% domestic minimum.
Foreign models are only allowed for scenarios not yet covered by Korean models, and the government does not pay for this portion.
This rule forces each consortium to include multiple domestic model vendors, with the side effect of expanding the order pool for the entire Korean AI industry.
The main difference among the three consortia lies in their distribution channels.
SK Telecom (SKT), with 22.5 million mobile subscribers, focuses on "Mobile AI" that can make calls and handle tasks for users, using its sovereign model A.X K2 with 68.8 billion parameters.
Kakao is embedding AI into the ubiquitous KakaoTalk messenger, enabling users to complete appointments, payments, and applications directly within chat windows.
KT is pursuing a channel alliance strategy, integrating the Daum portal and a batch of vertical apps into a single conversational interface.
All three approaches are betting on the same thing: whoever first turns AI from a chat tool into a public agent capable of filing taxes, booking hospital appointments, and applying for subsidies for ordinary people—doing things ChatGPT currently cannot do.
The first version, launching by year-end, will be a general-purpose chatbot; the real test will come in 2027.
The essence of this design is demand-side industrial policy.
Over the past decade, the mainstream approach for countries supporting domestic technology has been to subsidize the supply side—building production lines for chipmakers and providing R&D funds to AI companies.
South Korea is spending money on the distribution end: using public funds to create a protected user pool, allowing domestic models to accumulate data and iteration experience in real scenarios.
Comparatively, this effort is larger in scale, has a tighter timeline, and places higher demands on consumer experience.
The Ceiling of the Experiment
The biggest variable is Naver's absence.
Naver, equivalent to South Korea's Google, is the nation's largest search and life service platform, owning a full suite of entry points including search, maps, shopping, and payment—conditions far superior to any competitor for creating a nationwide AI assistant.
Yet it didn't even participate.
https://www.naver.com/
In January this year, Naver Cloud was eliminated from the government-led sovereign AI foundation model project because the cosine similarity between its HyperCLOVA X's vision encoder and Alibaba's Qwen reached 99.5%, failing the independence requirement—simply put, it secretly repackaged Qwen.
https://www.cryptopolitan.com/korea-naver-ncsoft-ai-contest/
Samsung is also absent.
In short, the final result is: The platforms with the most user scenarios and the largest hardware manufacturer in South Korea are not at the table. The national-level project is being undertaken by two telecom operators and a messaging app.
Retention is another hard problem.
The basic version of ChatGPT is itself free; Koreans won't stop using it just because there's another free alternative.
Wiseapp data shows ChatGPT had the fastest user growth among all apps in South Korea in the first half of the year, adding 9.86 million users in six months.
The government can get users in the door, but keeping them depends on whether domestic AI agents can do things ChatGPT cannot—functions like filing taxes, booking appointments, and applying for subsidies, which are deeply integrated with local government systems, are precisely what foreign products cannot connect to in the short term.
This is the most noteworthy bet in the South Korean plan: exchanging the exclusivity of government service scenarios for user stickiness to domestic AI.
512 B200s serving 51 million people—compared to the computing power behind ChatGPT's 900 million weekly users worldwide—the gap in scale is evident.
The free promise lasts until 2028; after that, operators must cover their own costs.
If users are accustomed to free service by then, converting them to paying users won't be any easier than it is today.
South Korea's experiment answers a question important to all countries outside the US and China: when model capabilities cannot keep up with the top players, can controlling the distribution end buy a survival window?
After the launch at year-end, the first validation point will be retention rates—when free ChatGPT and free domestic AI are placed side by side, the choices made by 51 million Koreans will determine whether this path is a replicable model or an expensive lesson.
Reference: https://en.yna.co.kr/view/AEN20260828002600320
This article is from WeChat public account "New Zhiyuan" (ID: AI_era), author: ASI Revelation; Editor: Marco.
