NVIDIA Jetson AGX Thor Developer Kit for AI Robotics & Edge Computing
The NVIDIA Jetson AGX Thor Developer Kit is the ultimate edge AI platform for next-generation humanoid robotics and physical AI applications. Powered by the cutting-edge Blackwell GPU architecture and featuring 128GB of LPDDR5X memory, this developer kit delivers an astounding 2070 FP4 TFLOPS of AI compute performance within a compact 130W power envelope. Compared to the Jetson AGX Orin, the Thor offers up to 7.5 times more AI performance and 3.5 times better energy efficiency, making it the definitive platform for running the latest generative AI models, vision-language-action (VLA) models, and large language models at the edge.
Why Developers Choose Jetson AGX Thor
Built from the ground up for physical AI and robotics, the Jetson AGX Thor is designed to accelerate low-latency, real-time applications with advanced Blackwell Multi-Instance GPU (MIG) technology and a robust 14-core Arm Neoverse-V3AE CPU. Whether you're building autonomous machines, intelligent robots, or edge AI systems for manufacturing, logistics, healthcare, or smart cities, the Thor delivers the performance, flexibility, and sensor fusion capabilities you need to innovate faster.
-
Unmatched AI Performance: 2070 FP4 TFLOPS to effortlessly run generative AI models including Isaac GR00T N, LLMs, and VLMs
-
Massive Memory Bandwidth: 128GB LPDDR5X memory for handling large transformer models and real-time multi-sensor data
-
High-Speed Connectivity: QSFP slot with 4 × 25GbE networking, multi-GbE RJ45, USB 3.2 Gen 2 Type-A and Type-C ports, HDMI 2.0b, and DisplayPort 1.4a
-
Advanced Accelerators: Third-generation Programmable Vision Accelerator (PVA), dual NVDEC/NVENC engines for video encoding and decoding, and optical flow accelerator
-
Ready-to-Deploy Software Stack: Ships with JetPack 7, Ubuntu 24.04 LTS, and native support for NVIDIA Isaac robotics, Metropolis visual AI, and Holoscan sensor processing
-
Complete Out-of-the-Box Experience: Includes 1TB NVMe SSD preloaded with software, Wi-Fi 6E module, and 140W power supply
Exceptional Performance for Physical AI
The Jetson AGX Thor is architected specifically for humanoid robots and advanced autonomous systems. The Blackwell GPU introduces FP4 precision, roughly doubling throughput over INT8 for models that support it, enabling real-time multimodal perception, decision-making, and control. The 14-core Arm Neoverse-V3AE CPU ensures ultra-low latency for safety-critical applications, while the suite of accelerators handles demanding vision tasks, video encoding/decoding (up to 4 × 8Kp30 or 10 × 4Kp60), and sensor fusion at scale.
Extensive Connectivity for Multi-Sensor Fusion
For high-speed sensor integration and real-time data processing, the developer kit offers extensive I/O options perfectly suited for robotics and autonomous systems:
-
1 × QSFP slot with 4 × 25GbE for high-bandwidth networking
-
1 × Multi-GbE RJ45 connector (5GbE capable)
-
2 × USB 3.2 Gen 2 Type-A ports
-
2 × USB 3.2 Gen 2 Type-C ports (one can be used for power delivery)
-
1 × HDMI 2.0b output
-
1 × DisplayPort 1.4a output
-
16 × CSI-2 camera lanes for multi-camera robotics applications
-
PCIe Gen 5 support for high-speed storage and expansion
-
Wi-Fi 6E module included
-
CAN, RS-232, and additional debug interfaces
Software Stack Designed for Edge AI
The Jetson AGX Thor runs a seamless cloud-to-edge AI software ecosystem designed specifically for physical AI applications. JetPack 7 SDK ships preloaded on a 1TB NVMe SSD with Ubuntu 24.04 LTS, reducing setup time and source-compilation work. The kit supports:
-
NVIDIA Isaac for robotics and Isaac GR00T N foundational models for humanoid robots
-
NVIDIA Metropolis for visual agentic AI and video analytics
-
NVIDIA Holoscan for sensor processing and real-time data pipelines
-
NVIDIA agentic AI workflows like Video Search and Summarization (VSS)
-
CUDA 13.0, TensorRT, and Docker containerized workflows for rapid prototyping
Ideal Applications
- Humanoid robotics and general-purpose robotic platforms
- Autonomous vehicles and mobile machines with multi-sensor fusion
- Smart city infrastructure and visual agentic AI systems
- Manufacturing automation and industrial robotics
- Healthcare and medical robotics
- Retail analytics and edge AI vision systems
- Research and development for frontier robotics and generative AI
What's in the Box
- NVIDIA Jetson AGX Thor Developer Kit with Jetson T5000 module pre-installed on reference carrier board
- 140W power supply with international adapters
- Wi-Fi 6E wireless module (installed)
- 1TB NVMe SSD with JetPack 7 and Ubuntu 24.04 LTS preloaded
- Quick start guide
- Antenna accessories (may vary by retailer)
Full Specifications
Compute
| GPU Architecture |
NVIDIA Blackwell GPU |
| AI Performance |
2070 FP4 TFLOPS |
| CPU |
14-core Arm Neoverse-V3AE |
| Memory |
128GB LPDDR5X |
| Module |
NVIDIA Jetson T5000 system-on-module (SoM) |
| Multi-Instance GPU (MIG) |
Supported (Blackwell MIG technology) |
Video & Vision
| Video Decode |
Dual NVDEC; up to 4 × 8Kp30 or 10 × 4Kp60 (H.265, H.264, AV1, VP9, VP8, MPEG-2/-4) |
| Video Encode |
Dual NVENC; up to 6 × 4Kp60 (H.265, H.264) |
| Vision Accelerator |
3rd-generation Programmable Vision Accelerator (PVA 3.0) |
| Optical Flow Accelerator |
Included |
| Camera Support |
16 × CSI-2 camera lanes |
Display
| Display Outputs |
1 × HDMI 2.0b, 1 × DisplayPort 1.4a |
| Maximum Resolution |
Up to 8K (7680 × 4320) @ 30Hz |
| Multi-Display Support |
Up to 4 independent displays |
Connectivity & I/O
| Networking |
1 × QSFP slot (4 × 25GbE), 1 × RJ45 Multi-GbE (5GbE capable) |
| USB Ports |
2 × USB 3.2 Gen 2 Type-A, 2 × USB 3.2 Gen 2 Type-C (one supports power delivery) |
| Wireless |
Wi-Fi 6E module included, Bluetooth |
| Storage |
1TB NVMe SSD (preloaded with software), PCIe Gen 5 support |
| Additional Interfaces |
CAN, RS-232, JTAG debug connector (2 × 5-pin header), USB Type-C debug port |
Power & Environment
| Power Consumption |
Configurable 40W - 130W (typical 130W for developer kit) |
| Power Supply |
140W AC adapter (included) |
| Cooling |
Heatsink with active fan (included) |
Physical Specifications
| Module Form Factor |
Approx. 87 × 100mm with 699-pin B2B connector |
| Developer Kit Weight |
Approximately 1.94kg |
| Magnetic Cover |
Included for carrier board access |
Software
| Operating System |
Ubuntu 24.04 LTS (preloaded) |
| SDK |
JetPack 7 |
| AI Frameworks |
NVIDIA Isaac (including Isaac GR00T N), Metropolis, Holoscan, CUDA 13.0, TensorRT |
| Container Support |
Docker (primary), additional runtime configuration available |
| Model Support |
Vision Language Action (VLA) models, Large Language Models (LLMs), Vision Language Models (VLMs), generative AI models |
Specifications are sourced from the manufacturer product page and may be updated.
