South Korea Picks Three Groups to Give Every Citizen Free AI Access

South Korea on Aug. 28 selected three consortia led by SK Telecom, KT and Kakao to build basic AI services that will be free and have no usage limits for every citizen. The government will supply the groups with a combined 512 Nvidia B200 graphics processors during 2026. The services are intended to identify benefits and help people with public applications, but the cost and computing capacity needed for unlimited use have not been settled.

Three service designs

SK Telecom is building a service that can take a request through planning, execution and confirmation. It will work through an app and the web, as well as phone calls and text messages for people who struggle with digital services.

KT plans to combine public and commercial services inside one conversation. Access points will appear in Daum search and content, on partner shopping and housing platforms, and through KT’s telecom and television channels.

Kakao will make KakaoTalk its main entrance. Its service is designed to handle reservations, applications and payments, with specialized agents for areas such as senior care and taxes.

Domestic models come first

Rules issued in July require at least 50% of each service’s model usage to come from Korean foundation models that meet government independence standards. At least 30% must come from domestic models developed outside the lead operator. Foreign models may cover limited functions, but their use will not receive government funding.

The basic chatbot and public-service agent are supposed to be free without usage limits. Under the July terms, operators may still charge for specialized services or higher-performance features. The public agent is intended to identify benefits for which a user may qualify and help with applications.

A December deadline

The science ministry plans to sign agreements with the three operators in September 2026, begin beta testing after that and launch the full service by year-end. The government will begin supporting part of the nationwide operating expense in 2027, while companies must contribute funds and develop revenue sources.

The government has not disclosed the program’s precise cost and future operating support remains unsettled. It also has not set an adoption target. Seoul officials estimate that more than 20 million South Koreans use free versions of generative AI, but the project has no target for how many will shift to the government-backed services.

The capacity assumptions

A July 17 capacity calculation modeled an operator receiving 256 B200 processors. It began with Nvidia benchmark data showing one B200 generating about 10,000 output tokens per second with the 70-billion-parameter Llama 3.3 model. Assuming delivery of 50 output tokens per second to each user, the scenario put simultaneous capacity at roughly 51,000 users.

A second calculation assumed four questions per user each day, an average response of 500 tokens and peak demand six times the daily average. It produced a theoretical capacity of about 18.4 million daily users. The 51,000 figure measures simultaneous capacity, while the 18.4 million estimate spreads use across a day under those demand assumptions. Those figures are assumptions, not established performance. They depend on lightweight, highly optimized models, short requests and no immediate surge in demand.

The assumptions also predated the selection of three operators. Under the allocation set out in July, the top-ranked consortium would receive 256 B200s while each of the other two would receive 128. The ministry has not disclosed which consortium ranked first. Larger models, longer conversations and public-service agent tasks could cut the number of users each processor can handle while raising token costs.

“What remains unclear is how token costs will be covered and how the service’s financial sustainability will be maintained,” an AI-service company official said in July.

Apple Trains China-Specific AI Model With Alibaba, Sources Say

Apple has trained a large language model specifically for China, according to three people familiar with the unannounced work, Reuters reported in an exclusive. Alibaba helped train the model, while A

The Implicator

Zuckerberg Says US Should Not Ban Chinese AI Models, Warns of Regulatory Capture

Mark Zuckerberg told the Financial Times on Tuesday that the United States should not block Chinese AI models and warned that American frontier labs could gain too much influence over reviews of compe

The Implicator

Trump Officials Revive Push to Bar Chinese AI Models After Kimi K3

Four separate attempts to restrict Chinese AI models reached internal consideration inside the Trump administration last year and were killed before any took effect, Axios reported Monday, and parts o

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai