NVIDIA has officially launched the new Jetson Thor robot computing platform. The product uses the latest Blackwell GPU architecture, with AI computing power reaching 2070 TFLOPS, which is 7.5 times higher than the previous generation Jetson Orin product, and energy efficiency has improved by 3.5 times.
According to the information, Jetson Thor is equipped with 128GB of ultra-large memory, which is a top configuration among edge computing devices. The launch of this platform aims to provide global developers with powerful tools to build physical AI robot systems, promoting deep interaction between robots and the real world.
NVIDIA CEO Jensen Huang stated at the product launch that Jetson Thor can run multiple AI models simultaneously and supports mainstream generative AI frameworks including Qwen and DeepSeek. He emphasized that this platform marks the official arrival of physical AI and the era of general-purpose robots.
In terms of hardware configuration, the GPU of Jetson Thor includes up to 2560 CUDA cores and 9 fifth-generation Tensor Cores, supporting multi-instance GPU technology, enabling efficient multitasking in complex scenarios. The processor uses a 14-core Arm Neoverse V3AE architecture, specifically designed for real-time control and task management, ensuring the stability of robot operations.
In terms of software support, the platform natively integrates the NVIDIA Isaac simulation platform, providing developers with a unified development environment from the cloud to the edge. Through FP4 quantization technology, the performance of some AI models can be doubled, effectively supporting real-time human-machine dialogue functions.
It is reported that several well-known Chinese enterprises have become early users of Jetson Thor, including Unitree Technology and Boston Dynamics. These companies have stated that the powerful computing power and flexibility of the new platform have significantly improved the navigation and interaction capabilities of robots in real environments.
Deepu Talla, Vice President of NVIDIA's Robotics and Edge AI, pointed out that true physical AI and robot development require the coordinated efforts of three stages: training, simulation, and deployment. Jetson Thor is a comprehensive solution designed based on this concept.
In terms of pricing, the Jetson AGX THOR developer kit is priced at $3499 in the U.S. market, approximately RMB 25,000. NVIDIA stated that the specific price in the China region will be announced separately.
Industry experts believe that the release of Jetson Thor will further reduce the barriers to robot development, accelerating the large-scale development and practical application of the robot industry. With the breakthrough of computing power bottlenecks, robot technology is expected to achieve commercial applications in more scenarios.
