Korea Subsidized Access. Malta Required the Course.
Ricardo Argüello, September 13, 2026
CEO & Founder
General summary
South Korea announced on August 28, 2026 that three consortia will deliver free, unmetered access to domestically built AI services for its entire population, launching in December. Malta got there first on May 16, conditioning its free year of ChatGPT Plus on completing an AI literacy course. Korea's own ministry says about a third of citizens cannot properly use AI.
- Korea's program selected SK Telecom, Kakao and KT consortia, with official launch in December 2026
- Malta conditioned its free year of ChatGPT Plus on completing a University of Malta course
- Around 21 million Koreans were already using free tiers of foreign AI before the program
- Humlum and Vestergaard's Danish study of 25,000 workers found precise null effects on earnings and hours
- Only 9.2 percent of Korean firms used AI in 2024, per the national business activity survey
A city gives every resident a bicycle to cut traffic. It works if people can ride and there are bike lanes. Otherwise you end up with a lot of new bicycles on balconies and the same traffic as before. Access was never the hard part of the problem.
AI-generated summary
South Korea’s own ministry says about a third of its citizens cannot properly use AI.
Its response, announced on August 28, is to give all of them free access.
Two governments, two theories
Malta went first. On May 16 it announced a deal with OpenAI giving every citizen a free year of ChatGPT Plus, on one condition: finish an AI literacy course built by the University of Malta.
Korea announced its program on August 28. Three consortia selected, SK Telecom, Kakao and KT, the last distributing through Naver. Beta from this month, official launch in December. No condition attached.
Two different bets about where the bottleneck sits. The evidence on which one is right already exists.
Two corrections to what circulated
English coverage repeated two claims that do not hold, and one of them is the whole argument.
First, Korea is not the first country to give citizens free AI access. Malta beat it by three months. The genuine first, and it is not a small one, is funding a domestically built, unmetered AI service for roughly 52 million people out of its own national budget, as public infrastructure rather than a vendor partnership. The localization mandate requires at least half the models to be Korean-made.
Second, and this is the one that matters. The claim going around was that 25 percent of Koreans already pay for AI against 2.5 percent of Americans. The real 24.3 percent comes from the Korea Consumer Agency and describes the share of generative AI users who pay, not the share of Koreans. In a sample of 10,000 adults, 1,101 used generative AI and 268 of those paid. That is 2.68 percent of Korean adults.
The ministry’s own numbers land in the same place by another route, roughly 1.8 million paying subscribers out of 52 million, about 3.5 percent.
Against the 2.2 percent of US households PNC measures from card transactions, the payment gap between the two countries barely exists.
The gap that is real
It is in usage, not payment.
Microsoft measures 37.1 percent of Korea’s working-age population using generative AI, against 31.3 percent in the US, with Korea posting the fastest growth in the world last quarter.
Here is the number that makes sense of the whole program. Before announcing it, Korea already had around 23 million generative AI users, and roughly 21 million of them were on free tiers of foreign services.
Twenty-one million people had already solved the price problem themselves.
So a skills problem got a spending solution. That is the part worth sitting with, and it is not unique to Seoul.
Denmark already ran the experiment
Humlum and Vestergaard studied roughly 25,000 Danish workers, linking two survey waves to administrative employment records. Most employers in exposed occupations adopted chatbots. Workers reported genuine productivity gains.
The difference-in-differences found null effects on earnings and hours. Not noisy nulls, precise ones, ruling out anything larger than 2 percent two years after ChatGPT launched, at both the worker and the workplace level.
Access, adoption, perceived productivity, and no movement in the outcome you can measure on a payroll.
Korea’s own enterprise data repeats the pattern. A KRIVET analysis of the national business activity survey shows the share of Korean firms using AI went from 1.4 percent in 2017 to 9.2 percent in 2024. Fewer than one in ten. Over the same period the gap between firms above 300 employees and those under 100 widened from 2.8 points to 11.8.
Access rose for everyone. So did the distance between the firms that know what to do with it and the ones that do not.
We wrote about the same distinction when the McKinsey numbers landed, in adoption is not transformation.
What a technology director takes from this
None of this is an argument against what Korea is doing. It is an argument about sequence.
If your company is about to buy AI licenses for everyone, you just watched the national-scale version of that plan get funded, and you watched the government that moved first decide to require the course before handing over the key.
Three questions I would answer before signing. How many of your people already use AI on their own, today, with you paying nothing? Which specific process turns an hour saved into a line on a financial statement? Who decides what gets automated and what does not?
If those three answers do not exist yet, the licenses will produce exactly what they produced in Denmark.
In AI Maestro, that sequence is the entire product. Two months of discovery, a map of where decisions actually live in your processes, and a go or no-go gate before anything gets installed.
Korea starts producing results in December. Worth checking back in a year, because it will be the largest study anyone has ever run on this question.
Let’s fix the sequence before you buy licensesFrequently Asked Questions
It is a Ministry of Science and ICT program giving the entire population free, unmetered access to the basic functions of domestically built AI services. The SK Telecom, Kakao and KT consortia were selected on August 28, 2026, with beta running from September and official launch in December.
No. Malta announced a deal with OpenAI on May 16, 2026 giving every citizen a free year of ChatGPT Plus after completing an AI literacy course. Korea's genuine first is funding a domestically built national AI service from its own public budget, treated as infrastructure.
Between 2.7 and 3.5 percent of the population, per the Korea Consumer Agency and the ministry's own figures. The widely circulated 24.3 percent is the share of generative AI users who pay, not the share of Koreans, and it is not comparable to the US household figure.
The available evidence says not on its own. Humlum and Vestergaard's study of roughly 25,000 Danish workers found broad adoption and self-reported productivity gains alongside precise null effects on earnings and hours, ruling out effects larger than 2 percent two years in.
Related Articles
AI Agents in Enterprise Ops: A Practical Playbook
A practical playbook for deploying AI agents in procurement, customer service, and compliance, with frameworks to bridge the demo-to-production gap.
AI Removes the Filter: Neurodivergence and Hidden Talent
15-20% of your talent pool thinks differently, and AI removes the barriers that excluded them. Companies that skip redesigning the pipeline lose out.
The 100x Employee Already Exists (And Changes How You Hire)
One AI-literate professional now produces what used to take a team. Jensen Huang confirmed it at GTC 2026. Here's what it means for your hiring strategy.
Claude Cowork: AI on Every Employee's Desktop
Anthropic launched Claude Cowork with plugins for operations, finance, and HR. What it means for B2B companies and how to get ready.
AI Beyond Engineering: A Guide for Every B2B Team
AI isn't just for developers. Specific playbooks for Marketing, Finance, Sales, HR, and Operations with 90-day adoption plans for each team.
78% of Your Employees Already Use AI Without Permission. Don't Stop Them.
Boris at Anthropic watched it happen: one data scientist opened Claude Code, and within a week the entire floor had it. 98% of companies have shadow AI.