ADVANTECH AIR-075 — NVIDIA AI PC
Edge AI inference system based on NVIDIA Jetson Thor T5000 (128 GB, 2070 FP4 TFLOPS) or T4000 (64 GB, 1200 TFLOPS) — 4 × / 2 × 10GbE RJ45 (PoE option), HDMI 2.1 + DP 1.4a, 2 × USB-C + 2 × USB-A 3.2, 2 × CAN (T5000), 8-bit DIO, 2 × RS-232/422/485; M.2 E key Wi-Fi, M.2 B key 5G, 2 × M.2 NVMe, TPM 2.0; 19 – 36 VDC; Ubuntu 24.04 with JetPack 7.1, Edge AI SDK / Inference Kit. Listed as “New” by Advantech.
Status: In production
Product: Jetson Thor edge AI inference system — T5000 / T4000, 4 × 10GbE, HDMI + DP, Edge AI SDK
AI module: Jetson T5000 128 GB (2070 TFLOPS) · T4000 64 GB (1200 TFLOPS)
Interfaces: 4 × / 2 × 10GbE (PoE option) · HDMI 2.1 + DP 1.4a · 2 × USB-C + 2 × USB-A · 2 × CAN · 8-bit DIO · 2 × COM
Expansion: M.2 E key Wi-Fi 6E · M.2 B key 5G · M.2 2280 + 2242 NVMe · TPM 2.0
Power / size: 19 – 36 VDC · 235.8 × 174 × 59 mm · wall mount
AIR-075 is the NVIDIA Jetson Thor model in Advantech’s AIR edge AI inference systems line, offered with a T5000 (14-core Neoverse V3AE, 2560 Blackwell CUDA + 96 Tensor Cores, 128 GB LPDDR5X, 2070 sparse FP4 TFLOPS) or T4000 (12-core, 1536 CUDA + 64 Tensor Cores, 64 GB, 1200 TFLOPS) version. It differs from MIC-742 in its interface set: instead of QSFP28, four (two on T4000) 10GbE RJ45, HDMI 2.1 + DP 1.4a dual display, CAN, DIO and two serial ports on DB9 — a classic industrial box-PC layout. The 235.8 × 174 × 59 mm, 2.89 kg chassis is wall mounted; the T5000 version operates from −15 to 40 °C (130 W TDP), the T4000 version from −15 to 55 °C. It ships with Ubuntu 24.04 + JetPack 7.1; model deployment and hardware validation are done with Advantech Edge AI SDK and Inference Kit.
Four 10GbE ports (Marvell AQR113; PoE with the optional MIOe PSE module) provide enough bandwidth for GigE Vision cameras and a 10G network backbone; with PoE in use the upper temperature limit drops by 5 – 10 °C. The front/rear panels carry two USB-C 3.2 Gen 1, two USB-A 3.2 Gen 2, a USB-C for RCM recovery, HDMI and DP, two CAN DB9 on T5000, an 8-bit DIO DB9, two RS-232/422/485 DB9 and line-out / mic-in. M.2 2230 E key Wi-Fi 6E, M.2 3052 B key 5G (regional Qualcomm modules), M.2 2280 (PCIe x4) and M.2 2242 (PCIe x2) NVMe slots and TPM 2.0 are provided; 512 GB is installed on the SSD SKUs. Power is 19 – 36 VDC through a 4-pin Phoenix connector (typical 13.2 W, 156 W at full load on T5000); the datasheet specifies 200 W continuous / 440 W peak for non-adapter supplies. Certifications CE, FCC Class B, CB, UL, CCC, BSMI; there is no RED certificate.
Technical specifications
| AIR-075 | |
|---|---|
| System (Jetson Thor module) | |
| Jetson module | NVIDIA Jetson T5000 (-S6A versions) or T4000 (-S6B versions) — by part number |
| CPU | T5000: 14-core Arm Neoverse V3AE 64-bit; T4000: 12-core |
| GPU | T5000: 2560-core NVIDIA Blackwell GPU with 96 fifth-gen Tensor Cores; T4000: 1536 cores, 64 Tensor Cores |
| AI performance | T5000: 2070 TFLOPS (FP4, sparse); T4000: 1200 TFLOPS |
| Memory | T5000: 128 GB; T4000: 64 GB — 256-bit LPDDR5X, 273 GB/s |
| Ethernet | |
| Ethernet | T5000: 4 × 10GbE; T4000: 2 × 10GbE — RJ45 (10M/100M/1G/2.5G/10G), Marvell AQR113; optional PoE IEEE 802.3af (MIOE-PSE-DPA1) |
| I/O | |
| Display | 1 × HDMI 2.1 (4K @ 60 Hz); 1 × DP 1.4a (4K @ 60 Hz) |
| USB | 2 × USB 3.2 Gen 1 Type C (5 Gbit/s); 2 × USB 3.2 Gen 2 Type A (10 Gbit/s); 1 × USB Type C (RCM recovery only) |
| CAN | T5000: 2 × CAN, DB9; T4000: none |
| Digital I/O | 1 × DB9, 8-bit |
| Serial port | 2 × DB9 RS-232/422/485 (web table: 1 × RS-232, 1 × RS-232/422/485) |
| Audio | Line-out, mic-in |
| Expansion and storage | |
| M.2 | 1 × M.2 2230 E key (PCIe x1, USB 2.0) Wi-Fi/BT; 1 × M.2 3052 B key (USB 3.2, USB 2.0) 5G |
| Storage | 1 × M.2 2280 M key NVMe (PCIe x4); 1 × M.2 2242 M key NVMe (PCIe x2); 512 GB SSD on the A1/A2/B1 SKUs |
| TPM | TPM 2.0 |
| Power | |
| Input voltage | 19 – 36 VDC, 4-pin Phoenix connector; AT / ATX (ATX default); optional 330 W 24 V adapter (other supplies: 200 W continuous, 440 W @ 6 µs / 330 W @ 1 ms peak) |
| Power consumption | T5000: typical 13.2 W, max. 156 W; T4000: 12 W / 104 W |
| Environment | |
| Operating temperature | T5000: −15 to 40 °C (35 °C with PoE), 0.7 m/s airflow, TDP 130 W; T4000: −15 to 55 °C (45 °C with PoE); the ordering table says −10 to 40 / 50 °C |
| Humidity | 95 % @ 40 °C (non-condensing) |
| Vibration | 3 Grms, 5 – 500 Hz, random, 1 hr/axis |
| Mechanical and software | |
| Dimensions (W × D × H) | 235.8 × 174 × 59 mm |
| Weight | 2.89 kg |
| Installation | Wall mount (kit included) |
| Operating system | Ubuntu 24.04 LTS with NVIDIA JetPack 7.1; WEDA / SUSI / Edge AI SDK / Inference Kit compatible |
| Certification | |
| Certification | EMC: CE, FCC Class B · safety: CB, UL, CCC, BSMI (no RED certificate) |
Key features
- Powered by NVIDIA Jetson T5000 and T4000, delivering up to 2070 TFLOPS FP4 inference performance
- 19 – 36 V wide power and wide temperature range
- Multiple I/O: 2 × USB-C 3.2, 2 × USB-A 3.2, M.2 B key (5G/LTE), M.2 E key (Wi-Fi), 2 × M.2 NVMe
- 10GbE with RJ45 connectors (4 ports on T5000)
- Supports WEDA / SUSI / Edge AI SDK / Inference Kit
Ordering information
| Part number | Description |
|---|---|
| AIR-075A-S6A1U | T5000, 128 GB, 512 GB SSD, 4 × 10GbE, 2 × CAN, made in Taiwan (the SKU in the advantech.com table) |
| AIR-075A-S6B1U | T4000, 64 GB, 512 GB SSD, 2 × 10GbE, no CAN, made in Taiwan |
| AIR-075A-S6A2U | T5000, 128 GB, 512 GB SSD (PCBA China, assembled in Taiwan) |
| AIR-075A-S6A3U | T5000, 128 GB, no SSD (PCBA China, assembled in Taiwan) |
| AIR-075A-S6B3U | T4000, 64 GB, no SSD (PCBA China, assembled in Taiwan) |
| AIR-075A-S6B4U | T4000, 64 GB, no SSD, made in Taiwan |
Part number: S6A = T5000 (4 × 10GbE, 2 × CAN, −10 to 40 °C), S6B = T4000 (2 × 10GbE, no CAN, −10 to 50 °C); the last digit is the manufacturing site / SSD option. Only S6A1U is listed in the advantech.com table; the others come from the datasheet dated 25 August 2026. The package includes the Phoenix connector counterpart, the wall-mount kit and the Ubuntu image; the adapter is optional.
Accessories and optional items
| Part number | Description |
|---|---|
| 96PSA-A330W24P4-3 | AC/DC adapter 100 – 240 V, 330 W, 24 V, 4-pin terminal block; power cord US 1702002600, EU 1702002605, UK 1702031801, PSE 1700000237, CCC 1700013977 |
| AIW-169BR-GX1 | Realtek Wi-Fi 6E M.2 2230 E key; antenna cable 1751000620-01, antenna 1751000651-01 |
| AIW-356DQ-JK1 / -E01 / -N01 | Qualcomm 5G M.2 3052 B key (Japan-Korea / EU / US SKU); antenna cable 1751000623-01, antenna 1750009372-01 |
| MIOE-PSE-DPA1 | MIOe PoE (PSE) module |
| 1700034631-01 / 1700031582-01 | USB-A to USB-C flashing cable 90 cm / DB9 debug cable 20 cm |
Is this Jetson Thor system right for your project?
Share your application (robotics / visual inspection / generative AI), the number and type of cameras (GMSL2 / GigE / USB), the network needs (25GbE / 10GbE), the CAN and I/O requirements and the ambient temperature; let’s decide on the module (T5000 / T4000), the model and the accessory set together. FST delivers the Thor system together with the camera / sensor layer, network design, model deployment and commissioning services as one solution.
All product specifications are subject to change without notice; data compiled from Advantech’s AIR-075 datasheet dated 25 August 2026 and the advantech.com product page.
Source: Advantech product data