Nvidia has unveiled the Jetson Orin Nano 2 — a compact computing module designed specifically for use in robotics and edge AI systems. The new generation of the device delivers twice the performance on neural network inference tasks while retaining the same form factor, enabling modern language and multimodal models to run directly on end devices without requiring a constant connection to cloud infrastructure.
According to the published technical specifications, the Jetson Orin Nano 2 is equipped with an 8-core Arm-architecture processor, 8 gigabytes of RAM, and delivers AI computing performance of up to 78 trillion operations per second (TOPS). Nvidia representatives note that thanks to enhanced Tensor Cores and higher memory subsystem bandwidth, the device delivers twice the inference efficiency compared to the previous Jetson Orin Nano Super model. The company has also preserved the compact form factor of the previous version. In 15-watt power mode, the new module delivers the same level of performance while consuming 40% less energy. This characteristic is of critical importance for devices operating directly at the network edge — autonomous robots, drones, and computer vision systems that need to process data in real time without continuously sending information to cloud data centers.
Nvidia positions the Jetson Orin Nano 2 as a foundational platform for the development of so-called physical AI — intelligent systems capable of perceiving the surrounding world through sensors, analyzing information, and independently making decisions to carry out physical actions. The computer supports modern artificial intelligence models specifically optimized for edge inference, including NVIDIA Cosmos, NVIDIA Nemotron, Gemma 4, and Qwen 3. Developers will be able to use these models to address a wide range of tasks: image and speech recognition, autonomous spatial navigation, natural human interaction, and on-device decision-making. Nvidia representatives note that advanced compact models are already capable of achieving accuracy comparable to larger systems of the previous generation, making it possible to move more and more generative AI functionality from centralized data centers directly to end devices.
According to company data, its robotics software stack is already used by more than 3 million developers worldwide. Among the first companies to adopt or test the Jetson Orin Nano 2 in their products are machine vision system manufacturer Cognex, industrial equipment maker Doosan Bobcat, and robotics company Matic. The latter is using the platform to develop home robots that need to simultaneously perform several complex tasks: mapping spaces, recognizing objects, interacting with people, and independently carrying out household chores. Another promising application scenario is autonomous drones. Wing, a subsidiary of Alphabet, uses the previous-generation Jetson Orin Nano in its delivery systems and plans to evaluate the new version to improve data processing speed and enhance the energy efficiency of its aircraft.
Nvidia's strategy is aimed at making generative and agentic artificial intelligence technologies accessible not only to large robotics companies, but also to developers of relatively small autonomous devices. The Jetson Orin Nano 2 occupies the lower price segment of the company's product lineup, yet gains computing capabilities that previously required significantly more resources to run modern models directly on a device. In the context of industry development, it is worth noting the forecast of ACE Robotics Chairman Wang Xiaogang, who expects an "ChatGPT moment" for embodied AI-based robots by the end of 2027. Such assessments reflect market participants' expectations regarding the acceleration of commercial deployment of autonomous robotic systems in the coming years.