Nvidia has unveiled the Jetson Orin Nano 2, a compact computing module designed for robotics and edge artificial intelligence. The new product delivers a twofold increase in inference performance while maintaining the same form factor, opening up the direct possibility of running modern language and multimodal models directly on the device, without relying on cloud resources.
Architectural breakthrough in a compact chassis
At the core of the Jetson Orin Nano 2 is an 8-core processor based on the Arm architecture, 8 GB of RAM, and computing power of up to 78 TOPS. The key improvement is upgraded Tensor Cores and increased memory bandwidth, which together provide double the inference efficiency compared to its predecessor, the Jetson Orin Nano Super. At the same time, engineers have preserved the compact dimensions, and in the 15 W power consumption mode, the device delivers the same performance while reducing energy consumption by 40%. This is critically important for autonomous systems operating in real time: robots, drones, and computer vision systems that must process data streams without a constant connection to data centers.
Physical AI: from perception to action
Nvidia positions the platform as the foundation for "physical AI"—systems capable not only of analyzing their surroundings but also of reasoning and then acting independently. The module supports models optimized for edge inference, including NVIDIA Cosmos, NVIDIA Nemotron, Gemma 4, and Qwen 3. Developers gain tools for image and speech recognition, navigation, human interaction, and decision-making locally. Notably, modern compact models already achieve accuracy comparable to large systems of the previous generation, allowing generative functions to shift from the cloud directly to end devices.
Ecosystem and early deployments
Nvidia's robotics stack is already used by more than 3 million developers. Among the first integrators of the Jetson Orin Nano 2 are machine vision systems manufacturer Cognex, Doosan Bobcat equipment, and robotics company Matic, which uses the platform for home robots that combine indoor mapping, object recognition, and autonomous task execution. Another illustrative case is Alphabet's Wing division, which uses the previous generation Jetson Orin Nano in delivery drones; the new version is being considered to accelerate data processing and improve energy efficiency.
Strategically, Nvidia is betting on democratizing generative and agentic AI, lowering the entry barrier for developers of small autonomous devices. The Jetson Orin Nano 2 occupies the lower segment of the lineup but gains computing capabilities that previously required significantly more powerful hardware. This is not just an evolution of hardware but a paradigm shift: AI is becoming truly local.
My take: A twofold increase in performance with reduced power consumption is not a marketing gimmick but a response to a real demand in the autonomous systems market. By 2027, when the "ChatGPT moment" for robots is predicted to arrive, such platforms will become the foundation for the mass adoption of embodied AI. The only question is how quickly the developer ecosystem adapts to the new capabilities of edge inference.