Overview
- NVIDIA announced the Jetson Orin Nano 2 on Thursday as a palm‑sized edge AI computer that delivers 78 TOPS of AI throughput, 8 GB of memory, and an eight‑core Arm CPU.
- The company says the module gives roughly twice the inference speed of the prior generation while using up to 40% less power in the same form factor and can match older Nano performance in a 15W low‑power mode.
- NVIDIA presented example runtimes for recent models, saying Nemotron 3.5 Lightning runs at about 115 tokens per second, and it framed the product as enabling on‑device language, vision and multimodal reasoning.
- Early customers and evaluators named by NVIDIA include Cognex, Doosan Bobcat, Matic and Alphabet’s Wing, and NVIDIA plans to ship the module and a developer kit in the first half of 2027.
- Analysts welcome the silicon and software push but warn that the module’s 8 GB memory will limit how large a model can run fully on‑device, and they note that so‑called 'small' models have grown in size, which will shape real‑world use cases.