Specifications
Models
Aetina AIB-AT78 (Jetson Thor T5000) and AIB-AT68 (Jetson Thor T4000)
AI Module
NVIDIA Jetson Thor — 100 x 87 mm, 699-pin board-to-board
AI Performance (T5000)
Up to 2,070 FP4 TFLOPS (sparse)
AI Performance (T4000)
Up to 1,200 FP4 TFLOPS (sparse)
GPU
NVIDIA Blackwell architecture with Tensor Cores
Memory (T5000)
128 GB LPDDR5X on-module
Memory (T4000)
64 GB LPDDR5X on-module
CPU (T5000)
14-core Arm Neoverse V3AE
CPU (T4000)
12-core Arm Neoverse
Power Envelope
40 W - 130 W (T5000); 40 W - 70 W (T4000)
Camera Input
Up to 20 cameras via HSB, up to 32 using virtual channels
PCIe
Up to 8 lanes of PCIe Gen5 (root port)
USB
Up to 3x USB 3.2 with integrated PHY
Display
4x shared HDMI 2.1 / DisplayPort 1.4a HBR2 with MST
Vehicle / Sensor I/O
4x CAN, 4x UART, 3x SPI, 12x I2C, 6x PWM, 2x DMIC
Generational Gain
Up to 7.5x the AI compute of Jetson AGX Orin, 3.5x better energy efficiency
Management Software
Aetina EdgeEye for remote monitoring, control and recovery
Provisioning
Aetina EdgeStore software transplant toolkit
Value-Added Services
Custom I/O and form factor, thermal design, BSP tuning, pre-installed third-party software
Partnership
NVIDIA Partner Network Elite Partner with priority Jetson access
Overview
The Aetina AIB-AT78/68 is the company's flagship autonomous edge AI platform built on the NVIDIA Jetson Thor family. The AIB-AT78 carries a Jetson Thor T5000 delivering up to 2,070 FP4 TFLOPS with 128 GB of LPDDR5X and a 14-core Arm Neoverse V3AE CPU; the AIB-AT68 uses the T4000 at up to 1,200 FP4 TFLOPS with 64 GB of memory and 12 Neoverse cores. Both pair a Blackwell GPU with Tensor Cores, configurable from 40 W to 130 W.
The platform is dimensioned for physical AI rather than camera-counting. Up to 20 cameras can be attached through the high-speed sensor bridge, with up to 32 resolvable through virtual channels; eight lanes of PCIe Gen5, three USB 3.2 ports and four CAN interfaces cover high-bandwidth sensors and vehicle or robot buses. Against Jetson AGX Orin, Thor delivers roughly 7.5x the AI compute at about 3.5x the energy efficiency.
Aetina wraps the module with the engineering services that turn silicon into a shippable system: custom I/O and form factors, thermal design, BSP tuning and pre-installed third-party software, plus the EdgeEye remote-management platform for monitoring, control and recovery of distributed devices and the EdgeStore transplant toolkit for moving trained models onto deployed fleets.
Key Benefits
2,070 FP4 TFLOPS with 128 GB unified memory runs large vision-language and generative models directly on the robot. Up to 20-camera input and 8 lanes of PCIe Gen5 keeps multi-sensor autonomy pipelines fed. EdgeEye plus EdgeStore give fleet-level remote management and model migration rather than a one-off box.
Applications
Humanoid and mobile robotics, autonomous mobile robots, physical AI research platforms, multi-camera perception for automation and logistics, medical imaging at the edge, and on-machine generative AI where data cannot leave the site.
Request a Quote — AETINA AIB-AT78/68 — NVIDIA JETSON THOR T5000 / T4000 EDGE AI PLATFORM
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →