'Korean NPUs Guard Runways and Fences': DEEPX Powers Air Force AI Perimeter Surveillance
DEEPX NPUs will power AI perimeter surveillance at an air force base, the first defense case of AI video surveillance on 100% Korean NPUs without foreign GPUs.
- Paired with Pintel's AI video analytics to build AI video surveillance on 100% Korean-made NPUs
- Video decoding and AI inference handled together without foreign GPUs, improving efficiency and cost-effectiveness

Aug. 5, 2026 — DEEPX, a developer of ultra-low-power on-device AI semiconductors, said Aug. 5 that its neural processing units (NPUs) will be used in a project to build an AI perimeter surveillance system at an air force base. It is the first case in the defense sector of an AI video surveillance system, traditionally built around foreign GPUs, being implemented entirely on Korean-made NPUs.
The project is part of the Ministry of National Defense's pilot project for an NPU-based intelligent video surveillance system. It is being carried out by KMNET, which specializes in building network infrastructure for military units, and AI video analytics company Pintel. DEEPX supplies the AI computing platform that runs Pintel's AI video analytics software.
The project will build an intelligent video surveillance system that uses AI to monitor key perimeter areas such as runways and fences at the air base around the clock. When the AI detects people, vehicles or abnormal behavior in real time, security personnel use that information to assess and respond to potential threats.
The project went through open competitive bidding, and the final contract was signed July 30. DEEPX's NPUs reportedly scored highly on performance and price in the bid evaluation and were selected as the final product.
The project uses DEEPX's DX-H1 V-NPU, which handles video decoding and AI inference together, and the DX-H1 NPU card for edge servers, which delivers up to 100 TOPS. Because a single NPU can handle the video decoding and AI inference previously done by GPUs, the system is designed to analyze many CCTV feeds in real time without a separate GPU, supporting stable AI services even in large-scale video surveillance environments such as air bases.
The project is significant because it implements AI video analytics, long dependent on foreign GPUs, entirely on Korean-made NPUs. It is expected to reduce reliance on foreign GPUs while lowering system costs and power consumption and improving operating efficiency.
Following Ulsan's public AI CCTV conversion project, DEEPX is extending its reach to this air force AI perimeter surveillance project, accelerating the commercialization of on-device AI in public safety and defense.
"This project is a meaningful example of an AI video surveillance system, once centered on foreign GPUs, built on Korean-made NPUs. We expect V-NPU technology, which handles everything from video decoding to AI inference on a single platform, to become a new alternative that makes AI video analytics infrastructure more efficient and cost-effective. Going forward, we will expand on-device AI beyond public safety and defense to smart cities, industrial sites, critical national infrastructure and other fields that need AI video analytics, helping grow Korea's AI chip ecosystem." — Lokwon Kim, CEO, DEEPX



