NVIDIA says Jetson Orin Nano 2 doubles edge-AI inference

At a press briefing, NVIDIA put a new computer in the place where a robot has to make decisions: inside the machine, not at the far end of a network. The Jetson Orin Nano 2 delivers 78 trillion operations per second (TOPS) of AI compute, and NVIDIA says it doubles inference performance compared with the Jetson Orin Nano Super.

The gain comes from improved Tensor Cores and higher memory bandwidth, while the board keeps the same compact form factor. Inference—the process of running an AI model—also becomes less demanding on power: in 15-watt mode, NVIDIA says the Nano 2 uses 40% less power than its predecessor at the same peak-to-peak performance. Those are the manufacturer’s figures.

The board is built for edge AI, where compact devices run models locally during deployment. It has 8GB of memory and an eight-core Arm CPU, and NVIDIA says its existing open software stack can run on it. The company also says the computer can support memory-efficient versions of LLMs, or large language models, and VLMs, or vision-language models, including Cosmos, Nemotron, Gemma 4 and Qwen 3.

That changes the options for developers working on machines that must interpret their surroundings while moving. Matic Robotics is using the Nano 2 in home-cleaning robots for conversational AI, gesture detection, precision mapping and semantic understanding. Wing, Alphabet’s delivery-drone subsidiary, plans to evaluate it for more responsive perception and reasoning. NVIDIA says more than 3 million developers are building on its robotics stack, while more than 10,000 companies are shipping or developing Jetson-based products.

Concretely, the immediate opportunity is not a finished robot arriving for consumers, but a smaller computing platform that developers and robotics companies can test in compact devices. It could let Matic add richer interaction and help Wing explore quicker, more energy-efficient deliveries, while preserving compatibility with existing Orin software. The Nano 2 is newly introduced and is being used or evaluated by partners; its performance and power claims still come from NVIDIA rather than an independent comparison.