NVIDIA's Jetson Orin Nano 2 doubles edge AI inference for robotics

NVIDIA asserts that as AI models become more efficient, more devices can become autonomous, but developers need compact, energy-efficient computers built for edge AI. The company introduced the Jetson Orin Nano 2, an entry-level computer for edge AI that it says enables millions of developers worldwide to build systems for physical AI applications.

Deepu Talla, vice president of robotics and edge AI at NVIDIA, said that today's small and medium frontier models have reached the accuracy of last year's largest frontier models, unlocking real-time intelligence for edge devices. The Jetson Orin Nano 2 brings this breakthrough within reach of developers, delivering the performance and energy efficiency needed for real-time reasoning in smart drones, robots, and vision AI systems.

During a press briefing, Talla noted that open models powering physical AI have evolved significantly over the past three months. A year ago, frontier models were 600 billion or 1 trillion parameters; now that level of accuracy is available in entry-level edge AI products like the Orin Nano 2. NVIDIA's Jetson Orin Nano 2 is part of the company's 'three-computer, full-stack' approach, with Omniverse and Cosmos providing simulation for testing, DGX supporting training, and Jetson serving as the 'robot brain' for runtime deployment.

The Jetson Orin Nano 2 delivers a significant leap in AI and video processing performance in a cost-effective, power-efficient system. It features 78 trillion operations per second (TOPS) of AI compute, 8GB of memory, and an eight-core Arm CPU.

Compared with the Jetson Orin Nano Super, the new module doubles inference performance through improved Tensor Cores and higher memory bandwidth, while maintaining the same compact form factor. In 15-watt mode, it consumes 40% less power to deliver the same peak-to-peak performance as its predecessor, according to Talla.

NVIDIA's existing open software stack can run on the Nano 2, which is built on the same GPU architecture as data centers. It is designed as a drop-in replacement for existing Orin customers. With Jetson agent skills and the company's ecosystem, developers can run the latest large language models (LLMs) and vision language models (VLMs) optimized for memory-efficient edge inference, including open models such as NVIDIA Cosmos, Nemotron, Gemma 4, and Qwen 3.

Talla said this unlocks a level of intelligence that was impossible before, allowing frontier AI models to run on robots and physical AI applications. He believes this will speed up the development and deployment of AI in the physical world. NVIDIA has also refreshed its higher-end Jetson offerings, including the Orin NX, Orin, and T3000/T2000 for mainstream applications and the Jetson T5000/T4000 for advanced uses.

NVIDIA reported that more than 3 million developers are building on its robotics stack, and partners such as Cognex, Doosan Bobcat, Matic, and Wing are evaluating or using the Jetson Orin Nano 2 for edge AI in compact devices. Talla noted that thousands of companies ship with entry-level edge AI on Jetson, and over 10,000 companies are shipping or developing products built on Jetson.

Alphabet's Wing is using the Jetson Orin Nano Super and NVIDIA's software stack in its delivery drones and plans to evaluate the Nano 2 for real-time AI perception and reasoning. Dinuka Abeywardena, head of perception at Wing, said the company is exploring the Nano 2 for more responsive and energy-efficient drones.

Matic Robots is using the Jetson Orin Nano 2 for its home cleaning robots, adding conversational AI, gesture detection, precision mapping, semantic understanding, and greater autonomy. CEO Navneet Dalal said the module enables state-of-the-art AI models at the edge in a compact home robotics platform.

Talla also mentioned that companion robots previously lacked good intelligence, but now frontier-level intelligence can run in real time, enabling a new class of applications. He cited Hugging Face's Reachy Mini, which uses a small LLM for speech recognition. NVIDIA listed many partners building carrier boards, hardware, and customized AI software.

Talla expects NVIDIA's technology to enable generalized navigation and dexterous manipulation, as well as improvements in robot training, safety, and independence. The Jetson Orin Nano 2 module and developer kit will be available in the first half of 2027.