NVIDIA launches new Jetson Thor computer, equipped with intelligent agent skills, significantly reducing robot memory usage

Zhixi Author | ZeR0 Editor | Mo Ying Zhixi reported on July 16 that today, NVIDIA launched new modules T3000 and T2000 based on the NVIDIA Jetson AGX Thor architecture, which can support mass market robots and edge AI applications on a large scale. Both modules are planned to be launched in the first quarter of 2027. So far, NVIDIA has provided a scalable edge AI platform with performance ranging from 70TOPS to 2000TFLOPS, allowing developers to handle almost any edge AI workload. Jetson AGX Thor powers the next generation of humanoid robots and robotic systems. Companies such as 1X, Agile Robots, Amazon Robotics, Boston Dynamics, FANUC, Hitachi and Techman Robot are all developing based on this platform. The hardware foundation supporting these functions is the Jetson and IGX T3000 modules, which provide 865 FP4 TFLOPS of AI computing power in a compact form factor (about half the size and power consumption of the T5000). Jetson T3000 integrates NVIDIA Blackwell GPU, 8-core Neoverse Arm CPU, 32GB LPDDR5X memory and 273GB/s memory bandwidth, as well as 25GbE network connection. While providing the same performance, IGX T3000 also integrates functional safety functions and can seamlessly run the NVIDIA Halos for Robotics full-stack safety system to provide safety guarantees for robots working together with humans. Although the T3000 is smaller in size, its inference performance is comparable to the T5000 in terms of multi-modal workloads (including large language models, visual language models, visual language action models, and world-based models). With memory prices high, migrating to T3000 helps