NVIDIA has introduced Jetson Orin Nano 2, a compact computer for running artificial intelligence in robots, drones, and vision systems. The device provides 78 TOPS of AI compute, 8 GB of memory, and an eight-core Arm processor.
AI directly in robots and drones
Edge AI processes information close to the device that produces it. A robot can analyze images or interpret instructions without continuously sending all its data to a remote server.
This approach may reduce latency and dependence on a permanent connection. However, it does not guarantee that an application will work entirely without internet access or that its data will automatically remain private.
Jetson Orin Nano 2 keeps the compact form factor of Jetson Orin Nano Super. The 78 TOPS figure describes the theoretical number of certain AI operations the device can perform per second, but it does not determine the actual speed of every application.
Twice the inference performance
Inference is the process in which a trained model analyzes new information and produces a result. In a robot, it can support object recognition, image interpretation, or instruction processing.
According to NVIDIA, Jetson Orin Nano 2 delivers up to twice the inference performance of Jetson Orin Nano Super through improved Tensor Cores and increased memory bandwidth.
Actual results will depend on the model, software, numerical precision, and memory required by each task. The comparison does not mean that every application will run at exactly twice the speed.
Lower power use in 15-watt mode
Energy efficiency is important for battery-powered devices and systems with limited space for cooling.
In 15-watt mode, NVIDIA says the new device can deliver the same performance as its predecessor while using 40% less power. This figure applies to a specific operating condition and is not a constant reduction for every workload.
Language and vision models
The device can run language models and models that combine vision and language when they are optimized for its available resources. NVIDIA cites the Cosmos, Nemotron, Gemma, and Qwen families as examples.
Possible applications include camera interpretation, object or condition detection, scene analysis, instruction processing, and inspection support. Some projects will require smaller models, quantization, and other adjustments to operate within 8 GB of memory.
What it provides and what is still required
Jetson Orin Nano 2 provides computing capacity, but it does not automatically turn a machine into an autonomous system. A complete product still requires sensors, software, motors, permissions, and safety mechanisms.
NVIDIA lists Cognex, Doosan Bobcat, and Matic among the first organizations exploring or adopting the device. This does not confirm that every company has completed a commercial product based on the new computer.
Availability remains to be confirmed
NVIDIA announced Jetson Orin Nano 2 on August 25, 2026. The verified information does not establish its price, delivery date, or immediate availability in every country.
