South Korea has hit the accelerator on Artificial Intelligence. During the APEC summit held in the country, the government — through the Ministry of Science and ICT (MSIT) — and some of its largest industrial conglomerates announced, in partnership with NVIDIA, a coordinated deployment of more than 260,000 GPUs to strengthen the sovereign AI infrastructure, accelerate the digital transformation of their industry, and build an ecosystem of talent and startups centered around “physical” and “agentic” AI.
This isn’t just a single mega-project, but a package of public and private initiatives designed to make Korea one of the world’s leading hubs for accelerated computing. “Accelerated computing infrastructure is becoming as vital as power grids or broadband,” emphasized Jensen Huang, founder and CEO of NVIDIA. The Seoul’s ambition is summarized by Bae Kyung-hoon, Vice Prime Minister and Minister of Science and ICT: “AI has shifted from being an innovation to becoming the foundation of future industries. Expanding the national infrastructure and developing technology with NVIDIA will strengthen our manufacturing capabilities and more.”
Sovereign Computing: 50,000 GPUs for the National Cloud and AI Computing Center
The first pillar is public. MSIT will deploy over 50,000 cutting-edge NVIDIA GPUs over the upcoming years, starting with an initial batch of 13,000 units to be integrated into NAVER Cloud (NVIDIA partner), NHN Cloud, and Kakao Corp. The plan includes the National AI Computing Center, conceived as sovereign infrastructure available to businesses, startups, and research institutes. These entities will be able to train models and build applications on high-performance resources without external dependencies.
The focus isn’t solely on quantity. The South Korean government has also brought NVIDIA into a parallel agenda for advanced mobile networks, with projects like AI-RAN and 6G in collaboration with Samsung, SK Telecom, ETRI, KT, LGU+, and Yonsei University. The goal is twofold: reduce computation costs in the network and extend device autonomy by offloading part of the inference process to the base station.
AI Factories: Heavy Industry Meets Accelerated Computing
The second pillar is industrial. Major Korean conglomerates have publicly announced significant investments in “AI factories” — platforms for computing and software to train, validate, and deploy models — each equipped with tens of thousands of GPUs:
- Samsung Electronics is building an AI factory with more than 50,000 GPUs to accelerate its semiconductor roadmap and digital transformation. The company will leverage NVIDIA technologies such as Nemotron (post-training datasets), CUDA-X, the cuLitho library, and Omniverse to create digital twins that improve the speed and performance of complex lithographic processes. Additionally, it will apply Cosmos, Isaac Sim, and Isaac Lab in its domestic robotics portfolio.
- SK Group is designing another AI factory ready for more than 50,000 GPUs focused on R&D and semiconductor manufacturing, as well as a pioneering industrial AI cloud in Asia that will utilize NVIDIA RTX PRO 6000 Blackwell Server Edition for physical AI and robotics. SK Telecom will operate this “sovereign” infrastructure, enabling local manufacturers to integrate Omniverse and accelerate projects related to digital twins and automation.
- Hyundai Motor Group, in collaboration with the government and NVIDIA, will launch an AI factory featuring 50,000 GPU Blackwell to train and deploy models for smart manufacturing and autonomous driving. The agreement includes the creation of a NVIDIA AI Technology Center, a Hyundai Motor Group Physical AI Application Center, and regional AI data centers, within a package of approximately $3 billion aimed at driving physical AI forward in the country.
- NAVER Cloud will expand its infrastructure with more than 60,000 GPUs— including RTX PRO 6000 Blackwell and other Blackwell accelerators— for sovereign AI and physical AI. This will support Nemotron open models on its own infrastructure and prepare sector-specific models (e.g., naval and security) with a focus on inclusive AI for citizens.
The common thread is physical AI — AI that interacts with and learns from the world — and agentic AI, designed to automate complex tasks in factories, networks, vehicles, or cities. The maturity of digital twins and industrial simulation with Omniverse finally connects with Korea’s manufacturing strength: from cleanrooms to assembly lines, including logistics, quality inspection, predictive maintenance, and network operations.
“Made in Korea” Foundation Models: Sovereign LLMs and Voice/Rationale Agents
Sovereignty isn’t limited to computing infrastructure. MSIT, alongside NVIDIA, is driving the Sovereign AI Foundation Models project to develop sovereign language models based on local data. The initiative will integrate NVIDIA NeMo and open datasets like Nemotron to train and distill reasoning models. Participants include LG AI Research, NAVER Cloud, NC AI, SK Telecom, and Upstage. The goal is to offer core Korean language models that companies, researchers, and startups can adapt to develop voice-enabled AI agents, reasoning capabilities, and other functionalities.
Additionally, LG will collaborate with NVIDIA to foster physical AI and support startups and academia with the EXAONE family of models — including EXAONE Path for healthcare, built with the MONAI framework — to accelerate oncological diagnosis and other high-impact applications.
Supercomputing and Quantum: KISTI Looks to the Next Frontier
A third pillar is scientific. The Korea Institute of Science and Technology Information (KISTI) and NVIDIA will establish a Center of Excellence linked to the sixth national supercomputer, HANGANG, based on NVIDIA’s accelerated computing. KISTI will also support the open architecture of NVIDIA’s NVQLink to connect quantum processors with GPU supercomputers via CUDA-Q. The aim: advance in quantum error correction, hybrid quantum-classical applications, and scientific AI foundations using PhysicsNeMo.
The message is clear: AI sovereignty depends not only on GPUs and models but also on supercomputing, quantum technology, and physical models that enable the country to lead in materials, energy, health, and basic science.
Startup Ecosystem and Capital: Turning Computing into Business Fabric
To translate investment into real economic impact, NVIDIA and partners will form a startup alliance supported by the NVIDIA Inception program. Members will gain access to accelerated computing via NVIDIA Cloud Partners — including SK Telecom — along with software and expertise from NVIDIA, with backing from the VC Alliance (including IMM Investment, Korea Investment Partners, and SBVA). Parallel to this, the tech company will collaborate with the government through the N-Up initiative of the Ministry of SMEs and Startups to incubate the next generation of companies.
What Does This Mean in Practice: Energy, Talent, and Value Chains
The announcement signifies a watershed moment for South Korea:
- Energy and Data Centers. Building AI factories with tens of thousands of GPUs requires dozens or hundreds of megawatts of power, new plants, and long-term contracts. Energy efficiency and thermal management — from liquid cooling to heat recovery — will be as crucial as the GPUs themselves.
- Talent and Education. The public-private plan is driving demand for software engineers, data scientists, cloud architects, digital twin specialists, and security experts. Universities and research centers will see an increase in internships, theses, and projects with industry partners.
- Supply Chain. The scale of GPUs, memory, interconnects, storage, and optics will strain suppliers and create opportunities for local manufacturers. Meanwhile, Korea will reinforce its strategic autonomy by reducing dependence on global bottlenecks.
- Intelligence Exports. As Huang pointed out, just as Korea exports ships, cars, chips, and electronics, it can now “produce intelligence” — software, trained models, digital twins, and robots — as a new export category.
Risks and Open Questions
No plan of this magnitude is without its uncertainties. Energy costs, hardware supply, data governance, privacy in sovereign models, regulatory harmonization for AI, and standards for 6G are all areas to watch. International competition — from the US, Europe, Japan, and Taiwan — is also moving quickly, and the market is unforgiving regarding timelines: if the utilization of the new AI factories doesn’t approach 100%, ROI will suffer.
Nevertheless, the layered design of South Korea’s plan — sovereign computing, sector-specific AI factories, own models, supercomputing, and startup ecosystem — mitigates some risks. It distributes investment, expands access points, and fosters virtuous dependencies among universities, clouds, industry, and venture capital.
What’s Next
In the short term, the focus will be on execution: cluster installation, the launch of the AI Computing Center, hiring equipment, and initial AI-RAN tests. Over the next 2-3 years, we should see domestic core models in production, digital twins operating in semiconductors, automotive, and shipbuilding, and pilot projects in robotics and autonomy. If momentum and the startup ecosystem persist, South Korea will be positioned to compete at the highest level in the global AI arena.
Frequently Asked Questions
What is an “AI factory” and how does it differ from a traditional data center?
An AI factory integrates accelerated compute (GPU), high-speed network, storage, software platforms (training, validation, MLOps), and simulation/digital twins to cycle models from training to continuous deployment. Unlike a generic data center, it’s optimized for AI (bandwidth, scheduling, frameworks, thermal profiles) and for physical AI (Omniverse, Isaac).
What are physical AI and agentic AI, and why is Korea investing in them?
Physical AI refers to AI that interacts with the environment (robotics, vehicles, factories, networks) and learns from it; agentic AI aims to automate complex tasks with objectives, planning, and tools. Korea combines manufacturing muscle and compute power to industrialize these capabilities: robots, digital twins, predictive maintenance, and autonomy in mobility.
How does AI-RAN and 6G fit into this plan?
AI-RAN applies AI to the radio layer and network management, optimizing power, spectrum, and coverage. It also shifts part of the inference to the base station, reducing battery consumption and latency. It’s a key component for 5G Advanced/6G and aligns with the vision of sovereignty: smarter, more efficient, and local networks.
What impact will this have on startups and high-skilled jobs in Korea?
The agreement creates a compute channel for startups via NVIDIA Inception and N-Up, with mentorship, software, and local VCs. Anticipated demand includes roles in AI, digital twins, cloud, security, chip design, and quantum. Additionally, collaborative projects between universities, large corporations, and startups are expected to grow.
via: nvidianews.nvidia