NVIDIA Jetson Thor unlocks new robotics potential
The NVIDIA Jetson Thor has officially launched, bringing unprecedented power to robotics and physical AI applications. Built on NVIDIA’s Blackwell architecture, the module delivers 2,070 FP4 teraflops, providing the compute and memory required for real-time reasoning. This leap forward will empower roboticists to process high-speed sensor data, enable visual reasoning at the edge, and handle complex multimodal workflows that were previously too slow for real-world use.
Compared to its predecessor, Jetson Orin, Jetson Thor offers 7.5x more AI compute, 3.1x more CPU performance, and double the memory. This makes it a breakthrough platform for humanoid robotics, industrial automation, autonomous vehicles, and advanced physical AI systems.
Adoption in humanoid robotics
Agility Robotics has confirmed that Digit, its humanoid robot, will transition to Jetson Thor for its sixth generation. The upgrade will allow Digit to expand real-time perception and decision-making skills, enabling more complex logistics operations in warehouses and factories.
Boston Dynamics is also integrating Jetson Thor into Atlas, its humanoid platform. By doing so, Atlas gains server-level compute and AI acceleration directly on device, supporting advanced mobility and manipulation tasks without heavy reliance on cloud resources.
Read Also
Real-time reasoning at the edge
Jetson Thor is optimized for generative reasoning models and can run large transformer-based AI frameworks locally. It supports Cosmos Reason, DeepSeek, Llama, Gemini, Qwen, and Isaac GR00T models, all at the edge with minimal latency.
This allows robots and AI agents to perform reasoning tasks, analyze live sensor streams, and respond instantly in dynamic environments. For fields like autonomous vehicles, smart spaces, or industrial safety monitoring, this advancement represents a critical step forward.
Expanding research and enterprise adoption
Leading universities including Stanford, Carnegie Mellon, and the University of Zurich are already using NVIDIA Jetson Thor in robotics research. At Carnegie Mellon’s Robotics Institute, Jetson Thor powers triage and search-and-rescue robots, enabling models like MAC-VO to operate at the edge with advanced perception and navigation.
Enterprise adoption is also growing. Hardware providers such as Advantech and ConnectTech are developing Jetson Thor-based systems, while software companies including Solomon and Vaidio are building AI workflows optimized for the platform. From humanoids to surgical assistants and smart tractors, the applications are broad and expanding quickly.
NVIDIA Jetson Thor ecosystem
The Jetson Thor ecosystem includes developer kits and production modules. With the Jetson T5000 module, reference carrier boards, and full support for NVIDIA’s AI software stack, developers can accelerate robotics workflows faster than ever. Platforms like NVIDIA Isaac, Metropolis, and Holoscan enable robotics, video analytics, and sensor processing applications directly on device.
Pricing starts at $3,499 for the developer kit, while T5000 modules are priced at $2,999 for 1,000-unit orders. With over 2 million developers already using Jetson technology, Jetson Thor is poised to redefine robotics development and physical AI innovation across industries.
