South Korea's government will build government-wide artificial intelligence services on homegrown neural processing units (NPUs), a move aimed at reducing reliance on Nvidia graphics processing units (GPUs) and nurturing domestic AI chip companies.
The National Information Society Agency (NIA) signed agreements on August 26 with FuriosaAI and Rebellions to purchase NPU servers worth a combined 10 billion won ($7.2 million), according to information technology industry sources on the 9th. A total of 25 AI servers will be installed to run new public-sector AI services.
The flagship service the government plans to build is an AI chatbot for public officials. The chat service is expected to give priority to AI models selected for the country's independent AI foundation model project, sources said. FuriosaAI and Rebellions are participating in consortiums led by LG AI Research, developer of Exaone, and SK Telecom (017670), developer of A.X, respectively.
The NPU-based government-wide AI services are seen as complementing existing GPU-centered systems. Relying solely on power-hungry GPU resources could drive up costs excessively as the public sector shifts toward AI transformation. NPUs offer higher performance per watt than GPUs and are specialized for AI inference, making them better suited to public-sector work centered on repetitive tasks.
The deal also carries weight for expanding the domestic NPU ecosystem. Once domestic companies secure cases of actual service operation and performance verification in the public sector, those can serve as references proving their technology and stability as they target overseas markets for public AI infrastructure. Domestic NPUs are expected to continue entering the public market, after AI chips from Rebellions, FuriosaAI and DeepX were designated last month by the Ministry of Science and ICT as Excellent R&D Innovative Products, an official government certification. "Domestic NPU companies are staking everything on building long-term partnerships with overseas clients, based on their track records with domestic public infrastructure and telecom and cloud partnerships," an official in the AI chip industry said.
