NVIDIA has made Jetson Thor modules generally available as a new class of robotics computers designed to run real-time Artificial Intelligence workloads at the edge. Jetson Thor delivers 7.5x more AI compute, 3.1x more CPU performance and 2x more memory than its predecessor, the NVIDIA Jetson Orin, enabling concurrent processing of high-speed sensor streams and visual reasoning in dynamic environments. The company positions the product as a way to reduce cloud dependence by shifting generative reasoning and inference to on-device execution.
Several robotics companies and use cases are highlighted as early adopters. Agility Robotics has integrated NVIDIA Jetson into the fifth generation of its humanoid robot Digit and plans to adopt Jetson Thor for Digit´s sixth generation to improve real-time perception and decision-making. Boston Dynamics is integrating Jetson Thor into its humanoid robot Atlas to bring server-level compute and high-bandwidth data processing on device. NVIDIA also expects Jetson Thor to accelerate other applications such as surgical assistants, smart tractors, delivery robots, industrial manipulators and visual AI agents by enabling larger, more complex models to run in real time.
Jetson Thor is optimized for generative reasoning models and supports transformer, vision language and vision language action models. NVIDIA lists compatibility with models and toolchains including Cosmos Reason, DeepSeek, Llama, Gemini, Qwen and domain-specific robotics models such as Isaac GR00T N1.5. The modules run the full NVIDIA software stack, including Isaac, Metropolis and Holoscan, and will continue to gain throughput and latency improvements through the NVIDIA CUDA ecosystem and future software releases.
Research labs at Stanford, Carnegie Mellon and the University of Zurich are evaluating Jetson Thor for advanced perception, planning and navigation. At Carnegie Mellon, teams use NVIDIA Jetson for autonomous robots that conduct medical triage and search and rescue, and plan to upgrade from Jetson AGX Orin to the Jetson AGX Thor developer kit to improve MAC-VO model performance, sensor fusion and fleet experimentation. The Jetson family includes a developer kit with a Jetson T5000 module, reference carrier board, active heatsink and power supply. Hardware and sensor partners such as Advantech, Aetina, ConnectTech, MiiVii, TZTEK, ADI, e-con Systems, Infineon, Leopard Imaging, RealSense and Sensing are building production systems and using Holoscan Sensor Bridge to stream sensor data into GPU memory. Pricing details in the article are not stated. NVIDIA also announced the DRIVE AGX Thor developer kit is available for preorder with deliveries slated to start in September.
