The South Korean government is rallying a broad ecosystem to maximize the competitiveness of domestically developed artificial intelligence (AI) semiconductors. It plans to build AI-powered smart cities to create large-scale demand for neural processing units (NPUs), aiming to reduce reliance on graphics processing units (GPUs) dominated by U.S. chipmaker Nvidia and accelerate the growth of Korean AI chip companies.
The Ministry of Science and ICT (MSIT) has decided to restructure the existing "K-Cloud Alliance" into the "K-NPU Alliance," expanding its scope to center on NPUs, according to IT industry sources on Wednesday. The National IT Industry Promotion Agency (NIPA), an agency under MSIT, is preparing detailed support measures to broaden the AI semiconductor ecosystem, including the alliance overhaul.
The K-Cloud Alliance currently comprises 153 member companies spanning AI chipmakers, cloud service providers, and AI model and service firms. The governance restructuring is seen as focused on more actively generating domestic demand for NPUs, analysts said. NPUs, which consume less power than GPUs, are drawing growing attention as an alternative as the AI industry shifts from the training phase to the inference phase. The application scope of AI semiconductors is also expected to expand well beyond data centers and cloud computing into on-device AI, smart cities, and virtually every industrial sector.
Industry players have increasingly called for a focal point to catalyze NPU industry growth in Korea amid the rapidly changing AI market landscape. Korean-made NPUs have achieved steep technological advances, but critics have pointed out that opportunities to validate their performance at the commercialization stage remain insufficient.
The K-NPU Alliance will serve as a platform for introducing AI services powered by domestic chips across public and private sectors through collaboration among member companies. Accordingly, city-scale projects enabling large-scale NPU trials will be launched in earnest. The projects aim to build NPU-optimized infrastructure suited for large-scale commercial services, including surveillance operations using NPU-equipped high-performance CCTV cameras and drones. For example, AI-powered CCTV systems could protect school zones within cities, while drone-based systems could manage coastal security.
Beyond city-scale trials, the initiative will also pursue the development of NPU-embedded intelligent devices and the discovery of on-device AI services. Korean-made AI chips are expected to be installed in robots and kiosks to develop AI agent services, among other applications. "Large-scale trials provide data and references on an entirely different level from individual device-level testing," an industry official said. "If the stability and efficiency of domestic NPUs are proven in environments with numerous connected devices, it will build a credible track record that strengthens competitiveness in overseas contract bidding."
Dedicated NPU computing infrastructure will also be established. NIPA plans to draw up a roadmap covering technical approaches, cost feasibility, and power supply measures. The initiative is expected to lend momentum to the "National NPU Computing Center" project currently being pursued by the city of Gwangju.
The domestic AI semiconductor value chain is being reorganized around NPUs because they are emerging as a viable alternative to GPUs. Armed with high power efficiency — strong performance relative to energy consumption — NPUs are expected to be applied across diverse industries, making it difficult for any single company to dominate the market, analysts said. While Nvidia's GPUs have effectively monopolized the high-performance chip market for AI training, startups with the capability to design chips tailored to specific devices could capture the on-device and inference segments first.
"The on-device AI market is very likely to see diverse hardware platforms take root depending on national or regional characteristics and demand," said Lee Sang-hyun, director of the AI Semiconductor Innovation Research Center at Sungkyunkwan University. "Korean NPU companies with the ability to develop products quickly have a real chance of competing."