NVIDIA Supercomputer for Robots

🤖 NVIDIA has launched Jetson Thor — a new computer for the next generation of smart robots.

The American corporation officially introduced the Jetson Thor robotics module, built on the Blackwell architecture. The device is designed to take the computational capabilities of smart robots to a whole new level, providing 2070 FP4 teraflops of power, making it an ideal tool for developers in robotics and physical artificial intelligence (AI).

Performance Breakthrough

Compared to its predecessor, Jetson Orin, Jetson Thor offers:

• 7.5 times more AI computations,
• 3.1 times higher CPU performance,
• twice the memory capacity.

These improvements open new possibilities for processing complex data directly on the device: from high-speed sensors to visual and multimodal analysis.

Integration into Humanoids

Agility Robotics has already confirmed the use of Jetson Thor in the sixth generation of its Digit robot, actively used in logistics. CEO Peggy Johnson stated:

“Powerful on-device processing provided by Jetson Thor will take Digit to a new level — increasing real-time responsiveness and expanding its set of advanced skills.”

Boston Dynamics is integrating Jetson Thor into its Atlas robot, allowing server-level data processing directly on the device, speeding up decision-making and enhancing robot autonomy.

Scientific Research and Innovation

Jetson Thor is actively used by Stanford University, Carnegie Mellon University, and the University of Zurich for research in robot navigation, perception, and planning. Associate Professor Sebastian Scherer explained:

“There used to be a large gap between computer vision and robotics because there weren’t enough computations for real-time decision-making. Now, models and computing resources allow robots to perform much more complex tasks.”

AI and Generative Model Support

Jetson Thor is optimized for generative models, including Cosmos Reason, DeepSeek, Llama, Gemini, Qwen, as well as specialized robotics solutions like Isaac GR00T N1.5. The module is fully compatible with the NVIDIA AI ecosystem: Isaac (robotics), Metropolis (video analytics), Holoscan (sensor processing).

Pricing and Availability

• Jetson AGX Thor Developer Kit — from $3499

• Jetson T5000 modules — from $2999 per batch of 1000 units

  
  

  NVIDIA has also opened pre-orders for DRIVE AGX Thor for autonomous vehicle solutions, with deliveries scheduled for September.

AI Investments

The company announced plans to invest up to $500 billion in AI infrastructure in the U.S., emphasizing its long-term strategy for developing robotics, autonomous systems, and artificial intelligence.

⚡ Conclusion
Jetson Thor ushers in a new era of robotics: more powerful on-device computing enables the creation of humanoids and autonomous systems capable of real-time decision-making, working with sensors and multimodal data, and integrating with modern generative AI models. Companies and research labs gain a tool that accelerates innovation and makes robots smarter and more autonomous.