Advantech showcased its AI platform, the MIC-743, fully integrated with NVIDIA Jetson Thor, at COMPUTEX 2025 in Taipei. Designed for physical AI and humanoid robotics, the MIC-743 delivers real-time inference and high-throughput data handling from multi-format camera sensors to AI compute, creating a compact, rugged solution ideal for mobility and industrial use cases.
The demonstration highlights Advantech’s end-to-end system architecture, including support for MIPI CSI and industrial camera protocols, NVIDIA Holoscan Sensor Bridge, and high-speed Ethernet aggregation via a dedicated sensor switch. These components funnel high-bandwidth data into the MIC-743, which leverages the Jetson Thor module’s 2,070 FP4 TFLOPS AI performance, 128GB LPDDR5X memory, and integrated 100G networking to process AI workloads such as vision-language models and transformer-based applications in real time.
- MIC-743 powered by NVIDIA Jetson Thor and Holoscan for edge AI
- Features 2,070 FP4 TFLOPS AI performance and 128GB LPDDR5X RAM
- Integrated 100G + 5GbE networking for high-bandwidth camera input
- Built for vision-based decision-making in robotics and mobile platforms
- Live demo at COMPUTEX Booth K0605, featuring NVIDIA Video Search and Summarization
“With the new MIC-743, we’re enabling intelligent systems that combine sensor fusion and real-time AI in a compact, rugged form factor,” said Dr. Bernie Malouin, Flex VP and JetCool Founder. “This is the kind of integration needed to power next-gen robotics and mobile edge AI.”
