Insider Brief
- Nvidia introduced two new Jetson Thor computer modules, the T3000 and T2000, designed to run robotics and edge AI workloads on smaller, more power-efficient systems.
- The Jetson T3000 delivers 865 FP4 teraflops of AI compute, while the Jetson T2000 provides 400 FP4 teraflops for visual AI agents, autonomous mobile robots, industrial manipulators and other edge systems.
- Nvidia also introduced Jetson agent skills for memory optimization and expanded its Cosmos 3 world foundation model family with Cosmos 3 Edge, a 4-billion-parameter model designed to run on Thor platforms.
Nvidia introduced two new Jetson Thor computer modules designed to bring robotics and edge AI workloads onto smaller, more power-efficient systems.
According to the company, the T3000 and T2000 modules are based on its Thor architecture and are intended for humanoid robots, autonomous machines, visual AI agents and other systems that need on-device AI processing.
Nvidia said Jetson AGX Thor is already being used by companies including 1X, Agile Robots, Amazon Robotics, Boston Dynamics, Fanuc, Hitachi and Techman Robot.
The new modules include:
- Jetson T3000: 865 FP4 teraflops of AI compute, a Blackwell GPU, an eight-core Neoverse Arm CPU, 32 gigabytes of LPDDR5X memory, 273 gigabytes per second of memory bandwidth and 25-gigabit Ethernet connectivity.
- IGX T3000: The same performance as Jetson T3000, with integrated functional safety and support for Nvidia Halos for Robotics, the company’s safety system for robots operating near people.
- Jetson T2000: 400 FP4 teraflops of compute and 16 gigabytes of memory for visual AI agents, autonomous mobile robots, industrial manipulators and other edge AI systems.
Nvidia said the T3000 delivers similar inference performance to the larger T5000 for multimodal workloads, including large language models, vision-language models, vision-language-action models and world foundation models.
The company also introduced Jetson agent skills, software tools designed to automate memory optimization, system configuration and deployment tasks across Jetson Thor and Jetson Orin devices.
Nvidia said UBTech, Agile Robots and Connect Tech reduced memory use by as much as 15 gigabytes through software optimization, allowing them to move from Jetson AGX Orin 64GB to the 32GB module. The company also cited memory reductions by SandStar, GROOVE X and NoTraffic in retail, companion robotics and traffic systems.
Nvidia also expanded its Cosmos 3 world foundation model family with Cosmos 3 Edge, a 4-billion-parameter model designed to run on Thor platforms. The company said the model is built for embodied systems that need to process the world, reason in real time and generate actions through on-device inference.
Developers can begin using T3000 emulation mode later this month with JetPack 7.2.1. T2000 emulation support will follow in a future release.
The Jetson T3000 and T2000 modules are scheduled to become available in the first quarter of 2027.
Image credit: Nvidia
