EVOC (Yanxiang Intelligent Technology) recently launched the AIB-3600 industrial AI computer. As an industrial-grade platform deployed at the production line edge to complete AI inference and real-time control where data is generated, the AIB-3600 targets the long-overlooked compute vacuum between cloud and device: in 2026, semiconductor inspection has entered the sub-micron level and embodied intelligent robots are moving toward volume delivery, making the real-time compute gap at the edge increasingly prominent — cloud links carry latency and disconnection risks, while insufficient edge compute causes queuing jitter and missed captures in vision inspection.
The AIB-3600 is powered by the NVIDIA Jetson AGX Orin 32/64G series processors, delivering up to 275 TOPS with 64GB of LPDDR5 memory, and can locally process multiple streams of high-resolution vision, radar, and force data simultaneously. On interfaces, it offers 8 channels of automotive GMSL 2.0 (Fakra-Z) coaxial video suited to the repeatedly flexing sections of robot arms; up to six 2.5GbE ports with optional PoE+ that power and connect industrial cameras over a single cable; expandable CAN FD for direct connection to servo motors and grippers; and 8 USB ports (up to 10Gbps) plus M.2 expansion supporting 4G/5G/WiFi/Bluetooth. The machine adopts an aluminum alloy plus sheet-metal fully sealed fanless design, supports -20°C to 60°C wide-temperature operation, DC 9-50V wide-voltage input with ignition control, and dynamically replaceable I/O including LAN, COM, GPIO, CAN, and GMSL modules. Official data indicates the AIB-3600 compresses inspection latency from seconds to milliseconds, achieving target recognition accuracy above 99.9%, planning-and-control latency under 10ms, roughly 30% lower integration and debugging costs, and 70% shorter algorithm iteration cycles.
From an industry landscape perspective, the industrial AI computer is becoming a third product curve beyond traditional industrial PCs and edge servers: industrial PCs excel in reliability and I/O expansion but lack AI compute structures, while general-purpose edge servers struggle with harsh production-line conditions. As embodied intelligence and sub-micron inspection make real-time inference a production-line necessity, platforms combining high compute density, native industrial interfaces, and wide-temperature sealed designs will take the lead in robot workstations, inspection stations, and mobile chassis — and the focus of edge AI hardware competition is shifting from raw compute parameters to a system-level contest of compute, interfaces, environmental tolerance, and software ecosystem.
Junyi Microelectronics’ K-series edge intelligent motherboards and IPC-i5-1235U industrial mini PC likewise serve on-site compute needs: the K series leverages RK platforms for multi-channel vision and lightweight AI inference, while the IPC-i5-1235U pairs a 12th-generation Intel Core processor with industrial software compatibility and the x86 ecosystem. Together with the AI-BOX-3588 edge computing host and touch panel all-in-one machines, Junyi offers domestic edge computing foundation options along the same lines as the AIB-3600 for semiconductor inspection, production-line quality control, and equipment host computers, working with industry leaders to advance the large-scale deployment of edge AI on manufacturing floors.
Source: Chinapp: EVOC Industrial AI Computer — A Millisecond-Level Compute Foundation Filling the Edge Compute Vacuum (2026-10-23)