NVIDIA Jetson AGX Thor Developer Kit, T5000 Module, Blackwell GPU, 128GB LPDDR5X, 2070 FP4 TFLOPS, 1TB NVMe, Robotics & Physical AI Edge AI Computer - 945-14070-0080-000

NVIDIA Jetson AGX Thor Developer Kit: the edge AI computer for robotics and physical AI

The NVIDIA Jetson AGX Thor Developer Kit (part number 945-14070-0080-000) is NVIDIA's most powerful Jetson platform, built to run generative AI, vision language models and robot foundation models directly on the machine. Powered by the Jetson T5000 system-on-module, it combines a 2560-core NVIDIA Blackwell GPU with 96 fifth-generation Tensor Cores, a 14-core Arm Neoverse-V3AE CPU and 128GB of LPDDR5X memory to deliver up to 2070 FP4 TFLOPS (sparse) within a 40W to 130W power envelope. NVIDIA rates it at up to 7.5x the AI compute and 3.5x the energy efficiency of Jetson AGX Orin.

Built for generative AI and real-time reasoning A next-generation Transformer Engine switches between FP4 and FP8 precision, so large language models (LLMs), vision language models (VLMs) and vision-language-action (VLA) models such as Isaac GR00T N1.5 and Cosmos Reason run locally with low latency, without relying on the cloud. Multi-Instance GPU (MIG) partitions the GPU into isolated instances, so critical control loops keep reserved capacity while other AI workloads run in parallel. A third-generation Programmable Vision Accelerator (PVA), an optical flow accelerator and dual video encoders and decoders offload vision tasks from the GPU.

High-speed sensor fusion The developer kit connects robots and smart-space systems to many sensors at once. A QSFP28 port provides 4x 25GbE and supports the NVIDIA Holoscan Sensor Bridge, which streams camera data over Ethernet directly into GPU memory for up to 20 cameras. It also includes a 5GbE RJ45 port, two USB-A 3.2 Gen 2 ports, two USB-C ports, HDMI and DisplayPort 1.4a outputs, and CAN, automation and JTAG headers for robot integration.

Ready to develop out of the box The kit includes the Jetson T5000 module with thermal solution on a reference carrier board, a pre-installed 1TB NVMe SSD in the M.2 Key M slot (PCIe Gen5 x4), a Wi-Fi 6E module and a power supply. It runs JetPack 7 with Ubuntu 24.04 LTS, Linux kernel 6.8 and CUDA 13.0, and supports the NVIDIA Isaac robotics platform (Isaac ROS, Isaac Sim, Isaac Lab), Isaac GR00T, Holoscan, Metropolis, PyTorch and Hugging Face models.

Who it is for? Robotics engineers, university and research labs, and teams building humanoid robots, autonomous mobile robots, industrial automation, medical devices, and video search and summarization agents for manufacturing inspection, worker safety, retail and smart cities.

Frequently asked questions

What is the NVIDIA Jetson AGX Thor Developer Kit? It is a complete development platform built around the NVIDIA Jetson T5000 module, designed to build and test physical AI and robotics applications that will run on Jetson Thor hardware.

How much AI performance does Jetson AGX Thor deliver? Up to 2070 FP4 TFLOPS (sparse), which NVIDIA rates at up to 7.5x the AI compute of Jetson AGX Orin.

How much memory does Jetson AGX Thor have? 128GB of 256-bit LPDDR5X memory with 273 GB/s of bandwidth, shared by the CPU and GPU.

What is the difference between Jetson AGX Thor and Jetson AGX Orin? Thor uses a Blackwell GPU with FP4 support, a 14-core Neoverse-V3AE CPU, 128GB of memory and 4x 25GbE networking. AGX Orin uses an Ampere GPU, a 12-core Cortex-A78AE CPU and up to 64GB of memory (275 TOPS).

What software does it run? JetPack 7 (Ubuntu 24.04, CUDA 13.0) with NVIDIA Isaac, Isaac GR00T, Holoscan and Metropolis.

What is the power consumption? Configurable from 40W to 130W.