South Korea Opens 256-B200-GPU Contest for a Domestic Cybersecurity AI Model
The selected team will get 10 months of Nvidia compute, with continued support dependent on an interim model after five months. The contest extends South Korea’s domestic-AI effort into security infrastructure.
Listen to this story
The audio brief
Story brief
3 key pointsSouth Korea is tying a cybersecurity AI-model competition to a 10-month allocation of 256 Nvidia B200 GPUs (32 nodes), with continued support contingent on an interim-model review after five months. The Ministry of Science and ICT and NIPA will select one team on technical capability, goals, and market impact. Naver Cloud, SK Telecom, and LG CNS are expected to organize bids, but reported consortiums are not final....
- 01
Compute access is performance-gated: the government can stop support after five months if the interim model falls short.
- 02
Applications reportedly close Wednesday; Naver Cloud, SK Telecom, and LG CNS are prospective bidders, with Bidraft considering entry.
- 03
Reported consortiums pair cloud, telecom, model-development, and security firms, but none are confirmed final teams.
South Korea has opened a government-led contest to build a cybersecurity-focused AI foundation model using domestic technology. The project is designed to combine local AI-security expertise into an independent model for deployment across the country’s security ecosystem.
What the government is buying
The Ministry of Science and ICT and the National IT Industry Promotion Agency are leading the Cybersecurity-Specialized AI Foundation Model Development Project. A domestic panel of AI and cybersecurity experts will choose the final team based on technical capability and development experience, development goals, and potential market and broader impact.
The selection criteria
- Technical capability and development experience.
- The team’s development goals for the model.
- Market potential and broader impact.
Compute is paired with an interim model
The winner will receive 256 Nvidia B200 GPUs, described as 32 nodes, for 10 months. After five months, the government will assess an interim model and decide whether to continue support based on the result. The award therefore ties sustained access to compute to progress during development, not only to the initial proposal.
Prospective bids remain unsettled
Applications close Wednesday, according to industry sources cited by The Herald Business. Those sources expect Naver Cloud, SK Telecom and LG CNS to form consortiums or submit bids; AI startup Bidraft is considering participation. These are prospective arrangements, not confirmed final bids.
The reported groupings would combine cloud, telecom, model-development and security expertise. Naver Cloud is understood to be assembling a consortium with the National Security Research Institute, Theori and Hunesion, among others. SK Telecom is reported to be working with SK Shielders, AhnLab and Genians, while LG CNS is preparing an entry with LG Uplus, LG AI Research and other ICT affiliates.
A security track joins two other domestic-AI efforts
The contest adds a security-focused track to South Korea’s wider domestic-AI agenda. Dokpamo, its independent foundation-model project, has narrowed candidates to LG AI Research, SK Telecom and Upstage, with two final teams expected to be selected early next year. Modu’s AI is pursuing a free chatbot and public AI agent built on domestic technology; six consortiums have entered, selected companies will receive 512 B200 GPUs, and the government aims to launch the service by year-end.
Taken together, the programs put domestic AI into three different settings: base-model development, citizen-facing services and cybersecurity. The new contest’s distinct role is to build an independent security model that can be used across South Korea’s security ecosystem.
Sources
- biz.heraldcorp.comSouth Korea launches push to build homegrown security-focused AI model - The Herald Business