South Korea’s sovereign-AI plan is not centred on a single national language model. It links public computing capacity with the industries where the country already holds global strength: semiconductors, telecommunications, automotive production, electronics, shipbuilding, and advanced manufacturing.
At the APEC CEO Summit, NVIDIA announced a broad collaboration with government agencies, cloud providers, universities, research institutes, Samsung, SK Group, Hyundai, LG, NAVER, and others. Hundreds of thousands of GPUs are planned across national clouds and company AI factories.
The objective is to treat accelerated computing like essential infrastructure, then use it to make industrial expertise more productive and locally controlled.
Public Capacity Opens Sovereign AI Beyond Conglomerates
The government plans up to 50,000 additional NVIDIA GPUs for businesses and research groups. An initial 13,000 Blackwell and other GPUs will be offered through NAVER Cloud, NHN Cloud, and Kakao while South Korea develops a National AI Computing Center.
Access gives startups and researchers a domestic place to train models and build applications without each organisation funding a private cluster. The Ministry of Science and ICT is also leading sovereign foundation-model work with NVIDIA NeMo and open Nemotron datasets alongside LG AI Research, NC AI, SK Telecom, and Upstage.
Telecom partners are researching AI-RAN and 6G, moving some computing from devices into base stations. That could reduce battery consumption and local hardware cost while extending national AI services through the network.
Public capacity also matters for competition. If only Samsung, SK, or Hyundai can afford advanced training infrastructure, sovereign AI could reinforce the position of existing conglomerates. Shared access through domestic cloud providers gives smaller companies a chance to build local-language and industry-specific systems without first raising the capital for a data centre.
Corporate AI Factories Connect Compute to Production
Samsung plans an AI factory containing more than 50,000 GPUs. CUDA-X libraries, electronic-design software, computational lithography, and Omniverse digital twins will support circuit development, factory simulation, logistics, predictive maintenance, and chip fabrication.
SK Group is building another 50,000-GPU facility, expected by late 2027, serving subsidiaries and external customers through GPU-as-a-service. SK Telecom is developing an industrial cloud, while SK hynix applies simulation and digital twins to chip design and autonomous fabs.
Hyundai will use Blackwell systems for training, simulation, robotics, factory automation, and autonomous driving. The partnership includes approximately $3 billion in joint investment, an NVIDIA AI Technology Center, Hyundai’s Physical AI Application Center, and new data centres.
These projects share a practical logic: simulate expensive physical processes before changing the real factory. A digital twin can test a production layout, robot path, or logistics plan without stopping an assembly line. AI investment is tied to measurable industrial throughput rather than being treated solely as a race to create a general-purpose chatbot.
Sovereignty Here Depends on a Foreign Platform
NAVER Cloud plans more than 60,000 GPUs and industry models for areas including shipbuilding and public safety. KISTI will use accelerated computing in its HANGANG national supercomputer and explore links between quantum processors and GPU clusters. An Inception-based alliance will connect startups with compute and investors.
The scale creates national capability, but it also reveals a tension. South Korea seeks sovereign AI while much of the infrastructure, software, and development environment comes from NVIDIA. Local control over data, models, and applications coexists with dependence on a foreign accelerator platform.
South Korea’s answer is to embed that platform inside domestic clouds, research, talent programmes, and industrial workflows. Its bargaining strength comes from being indispensable elsewhere in the AI supply chain, particularly memory and advanced manufacturing.
The strategy is less about isolation than coordinated capacity. If public access, corporate investment, and sector expertise reinforce one another, South Korea can produce specialised intelligence in the same way its factories produce chips, cars, and electronics. Sovereignty will be measured not by owning every layer, but by retaining the ability to direct what the infrastructure builds.
