On August 25, NVIDIA announced Jetson Orin Nano 2, focusing its entry-level edge AI platform on higher inference performance and better power efficiency. It is a development platform for robots, smart cameras and edge devices, not a consumer graphics card shipping today.
Specifications and performance
Jetson Orin Nano 2 offers up to 78 TOPS, 8GB of memory and an eight-core Arm CPU. NVIDIA says it delivers about twice the inference performance and uses 40% less power at the same performance; real results vary by model, quantization and workload.
The platform continues the Jetson software ecosystem and supports models and skills including Cosmos, Nemotron, Gemma 4 and Qwen 3. The value is not just the TOPS number but whether existing tools can deploy models to cameras, robots and devices that cannot stay connected to the cloud.
Why edge AI needs this kind of platform
In factories, warehouses, retail and mobile robots, video and sensor data may not be suitable for constant cloud upload. Local inference reduces round-trip latency and can preserve basic operation when connectivity is unreliable. Power, cooling, memory and model updates often matter more than one cloud benchmark.
Edge devices still have hard limits. Developers need to check whether a model fits in 8GB, whether the inference framework uses Jetson acceleration and whether cooling and power support sustained operation. More platform performance does not make every model real-time on a small device.
When will it be available?
NVIDIA currently points to the first half of 2027. Jetson Orin Nano 2 is still being prepared and is not shipping immediately after the announcement. Final pricing, developer-kit supply, regional channels and software versions remain to be detailed.
NVIDIA also cited partners including Cognex, Doosan Bobcat and Matic, along with more than three million developers in the Jetson ecosystem. These figures describe the platform's position, not proof that Orin Nano 2 is already shipping inside every partner product.
The takeaway for developers
Jetson Orin Nano 2 points to entry-level edge AI moving toward more performance, lower power and broader model support. It remains an announced platform planned for the first half of 2027. Whether to wait should depend on model size, latency, power budget and deployment volume, not only 78 TOPS or 2x performance.
