South Korea's Free AI Service Faces Capacity Questions
South Korea has opened procurement for an “AI for Everyone” service intended to give residents free, unlimited access to a domestically powered chatbot and, later, an agent for government services. The plan is concrete but not yet proven at national scale: officials expect to choose two or three operators, provide each with 256 Nvidia B200 GPUs, start a beta in September and target a full launch in December. Reporting on the tender says funding, token costs and performance under peak demand remain unresolved, so the initiative should be treated as an ambitious procurement program rather than a finished public service.
South Korea has moved its “AI for Everyone” proposal into procurement, asking private operators to build a free general-purpose chatbot around domestically developed models and connect it to public-service discovery and applications. The Ministry of Science and ICT published the project notice on July 13. The service is targeted for beta testing in September and a broader launch in December, according to The Korea Times.
What the government is procuring
The ministry’s notice describes a service meant to let people use Korean AI easily and without cost. The tender structure reported by The Korea Times calls for two or three operators, with at least half of the models used by each service coming from independently trained domestic foundation models. Operators would also reserve part of usage for other Korean model providers.
The government plans to allocate 256 Nvidia B200 GPUs to each selected operator, while requiring matching investment. The Register reports that bids are due August 11 and that the procurement also covers an agent intended to help residents find and apply for government services. These are procurement commitments and design goals; they do not show that a national-scale service is already operating.
Why capacity and funding remain open questions
The Korea Times estimates that roughly 23 million people in the country use generative AI. Its capacity calculation suggests 256 B200 GPUs could support about 18.4 million daily active users only under favorable assumptions, including a lightweight model, short responses and controlled peak demand. That figure is a journalistic scenario, not a guaranteed service benchmark.
Real capacity will depend on model size, input length, reasoning workload, batching, latency targets, safety systems and how sharply demand spikes. The same reporting says the mechanism for covering token costs and sustaining the operators financially has not been made clear. Larger domestic models or more complex government-service tasks would reduce throughput from the simple estimate.
What practitioners should watch
The program combines public GPU capacity, private matching investment and domestic-model requirements in a single national procurement. That makes the September beta an important test of whether sovereign-AI policy can translate into reliable public infrastructure.
The practical evidence will be measured performance: queue times, rate limits, model mix, service availability, cost per request and completion rates for government-service tasks. Until those results and the long-term funding model are published, the defensible conclusion is that South Korea has launched a serious procurement effort whose accessibility promise remains to be demonstrated.
Key Points
- 1South Korea opened procurement for a free national chatbot and a later public-service AI agent built around domestic models.
- 2The plan targets two or three operators, 256 Nvidia B200 GPUs per operator, a September beta and a December launch.
- 3Token funding, peak-load performance and long-term operating sustainability remain unproven.
Scoring Rationale
A national procurement for free public access to domestic AI models could materially affect sovereign-AI infrastructure and government-service delivery. The score remains below the highest tier because the service is at the tender stage, capacity assumptions are unverified and the operating-cost model is unresolved.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
