South Korea to deploy 10,000 Nvidia GPUs to startups, AI projects

South Korea will allocate 10,000 Nvidia GPUs to SMEs, startups, and AI projects starting in February.

The government recently spent 1.4 trillion won (US$947.2 million) to purchase these GPUs as the first phase of a plan to acquire 50,000 units by 2030.

The GPUs will be used in a large-scale cluster to support high-speed computing for AI model training and inference in industry, academia, and state projects.

Applications for access are open online until January 28, with up to 256 H200 GPUs or 128 B200 GPUs available per project.

Another 6,000 B200 GPUs, to be delivered later, will help develop homegrown AI foundation models.

Nvidia CEO Jensen Huang said the company would deploy 260,000 GPUs to South Korea to build large-scale AI factories with the government and major firms, including Samsung Electronics.

Of these, 50,000 GPUs will build a national sovereign AI platform, while Samsung Electronics, SK Group, Hyundai Motor Group each receive 50,000 units, and Naver Corp. will get 60,000 GPUs.

🔗 Source: Yonhap

🧠 Food for thought

Implications, context, and why it matters.

The pool’s access rules and GPU caps will decide if SMEs can compete

China’s voucher programs (government-issued credits that offset cloud or compute bills) cover up to 80% of AI costs for SMEs in Shanghai (CN¥600 million) and Chengdu (CN¥100 million for research institutions) 1. The 10,000-GPU pool has not disclosed prices or subsidy terms, so smaller firms cannot judge affordability.
Caps of 256 Nvidia H200 GPUs or 128 Nvidia B200 GPUs per project look high. If slots renew yearly and not quarterly, startups could crowd out others, including the 163,000 SMEs in Korea’s Digital New Deal (a government program to spur digital transformation) 2. China also paired compute aid with CN¥100 million in training data vouchers 1, which Korean SMEs may need to use their GPUs well.

Data center builders can win work as Korea targets 260,000 GPUs

Reaching 260,000 GPUs needs sites. PDG (Princeton Digital Group) put its 48 MW Incheon AI campus at $700 million 3. Korea will add gigawatt scale builds. Korea Electric Power Corporation (KEPCO) needs 1 to 4 years for facilities over 5 MW 4. New data centers must generate 2 to 20% of power on site by 2040 4.
Samsung SDS (Samsung’s IT services arm) leads a consortium as the sole bidder for the National AI Computing Center 5. Targets 15,000 GPUs by 2028 and 50,000 by 2030. Contractors should watch Ministry of Science and ICT (MSIT) procurement posts, which offer tax credits up to 25% plus fast track power reviews 6.