27.3 Billion Won Project to Strengthen Domestic AI Semiconductor Ecosystem
The South Korean government is moving forward with plans to establish micro data centers (MDCs) that leverage domestic AI semiconductors, aiming to expand the local AI semiconductor industry. These small-scale data centers will be located in non-metropolitan areas, providing affordable AI computing services for SMEs and medical institutions while creating demand for domestic semiconductor manufacturers.
The Ministry of Science and ICT announced on July 6 that it is pursuing a 27.3 billion won project to build these MDCs. The project is designed as a proof-of-concept stage where domestic AI semiconductor companies can deploy their products within existing AI data center markets, which have been dominated by NVIDIA’s GPUs.
What Are Micro Data Centers (MDCs)?
MDCs are compact, self-contained data centers that integrate servers, storage, networking, cooling, power, and security systems into containerized units. Their key benefit is ultra-low latency AI inference, achieved by situating MDCs close to end-users such as factories, hospitals, or government offices.
Current Market Challenges
- The domestic data center market is largely controlled by global cloud service providers like AWS and Microsoft Azure.
- These providers impose strict requirements on hardware, including GPU specs, server configurations, and infrastructure designs for power and cooling.
- As a result, local telecommunications and IT companies find it difficult to independently design and operate data center infrastructure.
- More than 90% of commercial data centers in Korea are concentrated in metropolitan areas, with over 40 large centers expected by 2028, according to Savills Korea.
- Global CSPs prefer metropolitan locations for latency and network efficiency, limiting the growth of decentralized data centers in regional areas.
Impact on Domestic AI Semiconductor Companies
Companies like Furiosa AI and Rebellions face challenges in commercializing their NPUs and ASICs due to the dominance of NVIDIA’s CUDA-based AI frameworks and market uncertainties. For example, Furiosa AI has developed an NPU for server inference that is five times more power-efficient than NVIDIA GPUs, yet securing large-scale demand remains difficult.
The Government’s Approach
The MDC project aims to address these issues by developing MDCs powered by domestic AI semiconductors. The Ministry of Science and ICT requires participating companies to develop platforms that can link and expand multiple MDCs, effectively creating a distributed cloud network over five years from 2025 to 2029.
Potential locations for these MDCs include Daegu and Gwangju, focusing on decentralizing AI computing resources outside of metropolitan centers.
Why This Matters for Government Officials
This initiative supports the growth of local industries by creating infrastructure that encourages adoption of domestic AI semiconductor technology. It also promotes regional development by situating data centers in non-metropolitan areas, potentially improving access to AI computing for SMEs and public institutions.
For more information on AI training and development relevant to this sector, explore resources like Complete AI Training.
Your membership also unlocks:
