Nvidia's Jetson computing platform is becoming a standard building block for running artificial intelligence directly inside physical machines, from factory robots to construction equipment, rather than relying on remote data centres.
The shift matters because machines operating in the real world cannot always afford to wait for a response from a distant server: a robotic arm on a production line, a self-driving vehicle or a piece of heavy equipment needs to process information and react in milliseconds, with or without an internet connection.
Jetson packs the computing power and memory needed to run modern AI models into a compact module designed to sit inside a machine rather than in a server rack, with entry-level models capable of running on hardware starting at the Jetson Orin Nano 8GB, one of Nvidia's smaller and more affordable configurations.
Caterpillar, the construction and mining equipment maker, has built an AI voice assistant called Cat AI that runs entirely on a Jetson chip inside its machines, using Nvidia's Nemotron speech models to understand and respond to spoken commands without a cloud connection.
Franka Robotics is running Nvidia's GR00T N1.6, a general-purpose robot control model, entirely onboard its robotic arms using the same platform.
Nvidia also described agentic use cases, in which an AI system is given a goal and works through a series of tasks to complete it overnight without human supervision, illustrated by a researcher instructing a Jetson-powered system: "Go to sleep. Everything will be ready by morning."
Related reading
- NVIDIA invests $4bn in Lumentum and Coherent to scale optics manufacturing
- NVIDIA partners with Lumentum to develop next-gen AI data centers using advanced optics
- Nvidia takes AI-powered mobile network technology from lab to live field trials
The platform supports a range of open-source models alongside Nvidia's own, including Gemma 3, Mistral 3 and Qwen, giving developers flexibility to choose or customise the AI that best fits their application.
Nvidia said tutorials are available on the AI development platform Hugging Face, with further demonstrations planned at its GTC conference next month.The recap
- Open models run locally on NVIDIA Jetson edge modules.
- Jetson Thor delivers 120 action tokens per second.
- NVIDIA will demo edge models at GTC next month.