Tony Kim
Aug 25, 2025 10:33
NVIDIA introduces Jetson Thor, a comprehensive platform designed to revolutionize physical AI and robotics, offering unmatched performance and scalability with the new Blackwell GPU architecture.
In a significant leap for robotics and artificial intelligence, NVIDIA has launched its latest platform, the Jetson Thor, designed to enhance the capabilities of physical AI and robotics. This announcement, made at GTC 2025, marks a pivotal moment in the evolution of robotics, as Jetson Thor offers unparalleled performance and scalability, making it the ultimate platform for developing next-generation humanoid robots.
Redefining Robotics with Jetson Thor
The Jetson Thor platform is built on NVIDIA’s Blackwell architecture, featuring a powerful GPU and 128 GB of memory, delivering up to 2070 FP4 teraflops of AI compute. This allows it to run the latest generative AI models efficiently, all within a 130 W power envelope. Compared to its predecessor, Jetson AGX Orin, it provides up to 7.5x higher AI compute and 3.5x better energy efficiency.
Jetson Thor’s capabilities are further enhanced with the new Blackwell Multi-Instance GPU technology, a robust 14-core Arm Neoverse-V3AE CPU, and a suite of accelerators, including a third-generation Programmable Vision Accelerator (PVA) and dual encoders and decoders. These features collectively empower developers to build adaptable robots that can perform diverse tasks across varied environments, reducing costs and expanding practical applications beyond what specialist robots can achieve.
Advancing Generative AI and Multimodal Processing
One of the standout features of Jetson Thor is its support for generative reasoning models, which are crucial for robotics platforms that simulate possible sequences of actions, anticipate consequences, and generate high-level plans. This capability leads to robotic systems that are significantly more flexible and capable of human-level reasoning in real-world settings.
Jetson Thor’s performance in handling generative AI models is impressive, offering speedups of up to 5x compared to Jetson Orin. With FP4 and speculative decoding, developers can achieve an additional 2x performance speedup, allowing for real-time responsiveness even when handling multiple models and a large number of sensor inputs.
Comprehensive Software Support
The Jetson Thor platform is complemented by a full-stack software suite, including NVIDIA Isaac for robotics, NVIDIA Metropolis for visual AI, and NVIDIA Holoscan for sensor processing. This software stack is optimized for low-latency, high-throughput applications, ensuring responsive and intelligent behavior in complex systems such as humanoid robots and autonomous machines.
Further, Jetson Thor supports the new Cosmos Reason, an open, customizable reasoning model for physical AI and robotics, enhancing its capabilities in deploying AI agents at the edge.
Implications for the Robotics Industry
With Jetson Thor, NVIDIA is not only advancing the technical capabilities of AI and robotics but also providing a robust ecosystem to accelerate time-to-market for developers. The platform’s extensive connectivity options and seamless integration with existing systems make it an attractive choice for innovators in the field.
The introduction of Jetson Thor signifies a major step forward in the development of generalist robotics, paving the way for more adaptable, efficient, and intelligent machines. As the robotics industry continues to evolve, NVIDIA’s latest offering positions itself as a cornerstone for future advancements.
For more information, visit the official NVIDIA page on NVIDIA.
Image source: Shutterstock
Source: https://blockchain.news/news/nvidia-jetson-thor-advanced-robotics-ai
