Artificial intelligence (AI) semiconductor corporations DEEPX said on the 5th that its Neural Processing Unit (NPU) will be applied to a project to build an Air Force base AI perimeter surveillance system. It is the first case in the defense sector of converting an AI video surveillance system, previously centered on foreign-made graphics processing units (GPUs), to a domestically produced NPU-based system.
The project is part of the Ministry of National Defense's "pilot project to build an NPU-based intelligent video surveillance system," carried out in cooperation with KMNET, a defense unit network infrastructure construction corporations, and Fintel, an AI video analysis corporations. DEEPX will supply the AI computing platform that runs Fintel's AI video analysis software.
The project covers key perimeter zones such as runways and fences at Air Force bases. When AI detects people, vehicles, and abnormal behavior in real time, surveillance personnel will use that to determine whether there is a threat and respond accordingly.
The contract was signed on July 30 through an open competitive bidding process. DEEPX's NPU was reportedly selected as the final product after receiving relatively high scores in performance and price competitiveness evaluations.
The applied products are DEEPX's DX-H1 V-NPU, which integrates video decoding and AI inference, and the DX-H1 edge server NPU card with performance up to 100 TOPS. Because a single NPU can handle both video decoding and AI inference—tasks previously handled by a GPU—it is designed to analyze multiple CCTV feeds in real time without a separate GPU, and it is said to enable stable operations even in large-scale video surveillance environments.
Following the Ulsan Metropolitan City public AI CCTV transition project, DEEPX is expanding the scope of application with this project, accelerating the commercialization of On-device AI for public safety and the defense sector.
Kim Nok-won, CEO of DEEPX, said, "This project is a case of implementing an AI video surveillance system, previously centered on foreign-made GPUs, on a domestically produced NPU basis, and we expect that V-NPU technology, which processes everything from video decoding to AI inference on a single platform, will be an alternative that boosts the efficiency and cost-effectiveness of AI video analysis infrastructure," adding, "Going forward, we will expand the application of On-device AI not only to public safety and defense but also to Smart City projects, industrial sites, and national core infrastructure, contributing to the spread of the domestic AI Semiconductor ecosystem."