Features

• Accelerated by NVIDIA Jetson T5000 and Jetson T4000 modules, built on the Blackwell GPU architecture for next-generation physical AI and robotics
• Up to 128GB LPDDR5X memory and 2070 FP4 TFLOPS of AI compute, engineered for large model inferencing at the edge
• 2x 5GbE LAN and 2x CAN-FD for high-throughput industrial and robotics networking
• 4x USB 3.2 ports, optional speaker and microphone audio, and dual SIM card socket for cellular connectivity
• Built-in IMU and GNSS for real-time motion tracking and positioning, useful in robotics and autonomous navigation
• M.2 M-Key 2280 NVMe storage, M.2 E-Key Wi-Fi/BT, and M.2 B-Key 4G/5G expansion slots
• Wide range DC power input, 19 to 36 VDC, for robotics and industrial power systems
• Ships preloaded with NVIDIA JetPack (Linux for Tegra), including CUDA, cuDNN, TensorRT, and DeepStream SDK drivers and libraries, along with configured client applications, ready to deploy out of the box.

Configuration

Memory (select): 64GB / 128GB

Capability

• Runs large language models (LLMs) and vision-language-action (VLA) models directly on the device, enabling generative AI at the edge
• Built on Blackwell GPU architecture, a generational leap in AI compute over the Ampere-based Orin family
• Ingests high-bandwidth sensor data through up to 4x 25GbE networking, suited for multi-camera and LiDAR-heavy robotics platforms
• Multi-Instance GPU (MIG) support splits a single GPU into isolated instances, running perception, language, and motion planning tasks in parallel
• Delivers up to 2070 FP4 TFLOPS of AI compute, positioning it for humanoid robotics and physical AI development
• Built-in IMU and GNSS support real-time localization for autonomous navigation and robotic motion control
• Designed for machines that understand spoken instructions and reason about their environment, not just recognize objects, a core requirement for humanoid and general-purpose robots

Target Users

Robotics and physical AI developers building systems that need to understand language and reason about their environment, not just detect objects. Common deployments: humanoid robots, vision-language-action systems, and edge devices running local LLM-based assistants where sending data to the cloud isn't an option.