AIR-075: AI Inference System Powered by NVIDIA Jetson Thor
AIR-075The AIR-075 provides powerful edge computing with dual 10GbE and GMSL interfaces to meet data-intensive AI requirements in traffic management and industrial applications. Powered by NVIDIA AI technologies, including NVIDIA Metropolis, NVIDIA Triton, NVIDIA Blueprint for video search and summarization, and NVIDIA Cosmos Reason, and integrated with Advantech Edge AI SDK and DeviceOn, it enables sensor fusion, multi-model inference, visual AI agents, and centralized device management for real-time, predictive edge intelligence.
- Powered by NVIDIA Jetson T5000 and Jetson T4000
- Up to 2070 TFLOPS FP4 inference performance
- 19~36V wide power and -10~40 °C wide temp. supported
- Multiple I/O Expansion: 2x USB-C 3.2 Gen 2, 2x USB-A 3.2 Gen 1
- 1x M.2 B Key (5G/LTE), and 2x M.2 E Key (NVMe / Wi-Fi)
- Supports 10GbE with Base-T connectors
- Support Edge AI SDK / Inference Kit / WEDA
Frequently Asked Questions
The AIR-075 provides the following I/O:
- 2x USB 3.2 Gen1 Type-C (USB 5 Gbps / DP1.4a)
- 2x USB 3.2 Gen2 Type-A (USB 10 Gbps)
- 1x USB 3.2 Gen1 Type-C (RCM only, for system recovery)
- 1x HDMI 2.1 (up to 4K at 60 Hz)
- 1x DisplayPort 1.4a (up to 4K at 60 Hz)
- 10GbE Ethernet ports (4x on the T5000 module, 2x on the T4000 module), RJ-45 with optional PoE (IEEE 802.3af)
- 2x DB9 CANBus
- 1x DB9 8-bit DI/DO
- 2x RS-232/RS-422/RS-485 COM ports
- Audio Line-out and Mic-in
The AIR-075 accepts 19~36V DC input and supports both ATX and AT power modes, with ATX as the default.
Operating temperature depends on the module and PoE status:
- AIR-075A-S6A1U (Jetson T5000): -15~40°C without PoE, -15~35°C with PoE
- AIR-075A-S6B1U (Jetson T4000): -15~50°C without PoE, -15~40°C with PoE
These ranges are rated with 0.7 m/s air flow. Operating humidity is rated at 95% at 40°C, non-condensing, and the unit is rated for 3 Grms vibration (5~500 Hz, random, 1 hr/axis). Certifications include CE/FCC Class B, CB, UL, CCC, and BSMI (no RED certificate).
Storage: 1x M.2 2280 M Key (NVMe) and 1x M.2 2242 M Key (NVMe).
Expansion: 1x M.2 3052 B Key for 5G/LTE and 1x M.2 2230 E Key for Wi-Fi/BT. An optional TPM 2.0 module is also available.
Depending on the ordering SKU, storage capacity is listed as 512GB.
The Jetson T5000 module has 128GB of 256-bit LPDDR5X memory at 273 GB/s. The Jetson T4000 module has 64GB of 256-bit LPDDR5X memory at 273 GB/s.
The AIR-075 runs Ubuntu 24.04 LTS with NVIDIA JetPack 7.1. It supports the Edge AI SDK, an Inference Kit for hardware-validated AI inference deployment, and WEDA. Advantech SUSI is also supported for device management and system monitoring.
The AIR-075 measures 235.8 x 174 x 59 mm (W x D x H) and weighs 2.89 Kg. It supports wall mounting, and a wall mount kit (2 pieces) is included in the packing list.
The AIR-075 has an M.2 B Key slot (M.2 3052) for 5G/LTE and an M.2 E Key slot (M.2 2230) for Wi-Fi/BT. Compatible add-on modules listed include the AIW-169BR-GX1 Wi-Fi 6E M.2 E-Key module and the AIW-356DQ-JK1 5G M.2 3052 B-Key module.
The AIR-075A-S6A1U uses the NVIDIA Jetson T5000 module (14-core Arm Neoverse-V3AE CPU, 2560 NVIDIA Blackwell cores, 96 Tensor Cores, 2070 TFLOPS FP4-Sparse, 128GB memory, 4x 10GbE, 2x CANBus, operating temperature -10~40°C).
The AIR-075A-S6B1U uses the NVIDIA Jetson T4000 module (12-core Arm Neoverse-V3AE CPU, 1536 NVIDIA Blackwell cores, 64 Tensor Cores, 1200 TFLOPS FP4-Sparse, 64GB memory, 2x 10GbE, no CANBus, operating temperature -10~50°C).
Both SKUs share 512GB storage, 1x HDMI, 1x DP, 4x USB, and 2x RS-232/422/485 8-bit DIO, with 19~36V DC power input.
How-to Guides
- Confirm the AIR-075 supports wall mounting as listed in the mechanical specifications.
- Use the Wall Mount Kit included in the packing list (2 pieces provided with the system).
- Refer to the wall-mount dimension diagram in the product documentation for hole placement and orientation.
- Secure the unit to the wall using the mount kit hardware.
- On the front panel, locate the USB-C ports, LAN, USB-A ports, HDMI, DP, and the Power Button/Power input.
- On the back panel, locate the COM ports, CANBus, GPIO, the 4-pin Phoenix power input, USB-C (RCM), Mic-in, and Line-out.
- Connect displays to the HDMI (up to 4K at 60 Hz) or DP (up to 4K at 60 Hz) ports as needed.
- Connect USB peripherals to the front-panel USB-C or USB-A ports.
- Connect serial devices to the RS-232/422/485 COM ports or CANBus DB9 connectors on the back panel.
The AIR-075 provides powerful edge computing with dual 10GbE and GMSL interfaces to meet data-intensive AI requirements in traffic management and industrial applications. Powered by NVIDIA AI technologies, including NVIDIA Metropolis, NVIDIA Triton, NVIDIA Blueprint for video search and summarization, and NVIDIA Cosmos Reason, and integrated with Advantech Edge AI SDK and DeviceOn, it enables sensor fusion, multi-model inference, visual AI agents, and centralized device management for real-time, predictive edge intelligence.
| Product Available On | Apr 22, 2028 |
|---|---|
| Manufacturer | Advantech |
| Short Description | The AIR-075 provides powerful edge computing with dual 10GbE and GMSL interfaces to meet data-intensive AI requirements in traffic management and industrial applications. Powered by NVIDIA AI technologies, including NVIDIA Metropolis, NVIDIA Triton, NVIDIA Blueprint for video search and summarization, and NVIDIA Cosmos Reason, and integrated with Advantech Edge AI SDK and DeviceOn, it enables sensor fusion, multi-model inference, visual AI agents, and centralized device management for real-time, predictive edge intelligence. |