TLDR: NVIDIA has introduced its Jetson AGX Thor platform, powered by the Blackwell GPU, which offers a substantial leap in AI performance for edge computing. Delivering up to 7.5 times higher AI compute and 3.5 times better energy efficiency compared to the Jetson AGX Orin, this new system is engineered to accelerate generative AI models and advanced physical AI applications, including robotics. With 2070 FP4 teraflops of AI compute within a 130-watt power envelope, the Jetson AGX Thor is set to revolutionize real-time AI at the edge. The developer kit is now available, with key industry players already integrating the technology.
NVIDIA has officially launched its Jetson AGX Thor platform, marking a significant advancement in edge AI computing designed to power the next generation of robotics and physical AI applications. This new system, featuring the powerful NVIDIA Blackwell GPU, delivers unprecedented performance, boasting up to 7.5 times higher AI compute capability and 3.5 times greater energy efficiency compared to its predecessor, the Jetson AGX Orin.
At the core of the Jetson AGX Thor is its ability to provide up to 2070 FP4 teraflops (TFLOPS) of AI compute within a remarkably efficient 130-watt power envelope. This immense processing power, coupled with 128 GB of memory and a robust 14-core Arm Neoverse-V3AE CPU, enables the platform to effortlessly run the latest and most complex generative AI models, including large language models (LLMs), vision language models (VLMs), and Vision Language Action (VLA) models like NVIDIA Isaac GR00T N.
The platform integrates several cutting-edge technologies to achieve its superior performance. It features the Blackwell Multi-Instance GPU (MIG) technology, which allows a single GPU to be partitioned into isolated instances, ensuring predictable performance for critical workloads while running other tasks in parallel. Furthermore, Jetson Thor introduces native FP4 quantization with a next-generation Transformer Engine that dynamically switches between FP4 and FP8 for optimal performance, accelerating both prefill and decoding in generative AI workloads. A suite of accelerators, including a third-generation Programmable Vision Accelerator (PVA), dual encoders and decoders, and an optical flow accelerator, further enhances its capabilities for high-speed sensor data processing and inference from visual information.
NVIDIA emphasizes the Jetson AGX Thor’s role as the ultimate platform for physical AI and humanoid robotics. It is designed to accelerate low-latency, real-time applications, making robotic systems more flexible, adaptable, and capable of human-level reasoning in real-world settings. The platform runs the comprehensive NVIDIA AI software stack for physical AI applications, including NVIDIA Isaac for robotics, NVIDIA Metropolis for visual agentic AI, and NVIDIA Holoscan for sensor processing, ensuring a seamless cloud-to-edge experience.
Early adoption of the Jetson AGX Thor is already evident across various industries. Major players such as Amazon Robotics, Boston Dynamics, Caterpillar, and Meta are integrating this technology into their advanced systems. Boston Dynamics, for instance, is incorporating Jetson Thor into its humanoid robot ‘Atlas’ to handle on-device computational tasks previously managed at the server level. Caterpillar highlighted the platform’s impact, stating, ‘NVIDIA Jetson Thor offers the AI performance we need to develop and deploy the construction and mining equipment of the future, enhancing precision, reducing waste and improving safety for our customers around the globe.’ Additionally, Agility Robotics plans to adopt Jetson Thor as its onboard computing platform, and Xiaomi has integrated it into its ‘YU7’ SUV to enhance driver assistance functions.
Also Read:
- Red Hat AI 3 Unveiled: Empowering Enterprise AI with Scalable Production Inference and Agentic Capabilities
- Orange Business, NVIDIA, LightOn, and Edarat Group Launch Sovereign Generative AI Platform for GCC Region
The NVIDIA Jetson AGX Thor developer kit is currently available for purchase, starting at $3,499, with shipping expected to commence in September 2025. Production modules, known as Jetson T5000, are also available through NVIDIA’s worldwide distribution partners, enabling broader access to this transformative edge AI technology.
