News
Nvidia Unveils New Jetson Thor Chips to Bring AI Agents to Robots
Nvidia on July 15 introduced two new Jetson Thor computer modules, the T3000 and T2000, designed to run advanced AI models directly on robots, autonomous machines and industrial systems rather than in the cloud.
The modules are built on Nvidia's Blackwell GPU architecture and are intended to "move advanced robotics, visual AI and edge workloads onto compact, power-efficient systems," the company said in a blog post.
The T3000 delivers 865 FP4 teraflops of compute in a form factor roughly half the size and power draw of Nvidia's existing T5000 flagship module, according to the company. The T2000, aimed at less demanding applications such as visual AI agents, autonomous mobile robots and industrial manipulators, offers 400 FP4 teraflops and 16 gigabytes of memory. Both modules are scheduled to become available in the first quarter of 2027, with T3000 emulation support arriving this month through Nvidia's JetPack 7.2.1 software.
Running AI models locally, rather than routing data to a remote data center, lets robots and industrial systems make decisions without waiting on a network connection, a capability Nvidia and industry analysts have tied to warehouse automation, equipment inspection and manufacturing systems that require near-instant responses.
Nvidia framed the launch as part of a broader shift toward what it calls "physical AI," systems that go beyond generating text or images to perceive their surroundings and act in the physical world. "General-purpose robots and autonomous machines are moving from research labs to real-world mass-market deployment," the company said, citing demand for compact, power-efficient AI systems that can run foundation models at the edge.
The company also released new Jetson agent skills, which it said can automate memory optimization and deployment tasks, helping developers achieve "significant memory savings in days instead of weeks." Nvidia said early adopters, including humanoid robotics makers UBTech and Agile Robots, have used the tools to cut memory usage by up to 15 gigabytes, allowing some systems to shift to lower-cost memory configurations.
Nvidia said the new modules round out a product lineup that now spans from its entry-level Orin Nano chips to the high-end T5000, giving developers options across a wider range of cost and performance requirements. "As physical AI and embodied AI move toward mainstream deployment," the company said, the new Jetson Thor computers are meant to give developers "a scalable foundation for bringing intelligent humanoids and autonomous machines into the real world."