Edge for AI

Purpose-built NPU, VPU and TPU hardware for running inference on-device, where sending video to a server is not an option.

207 products from 11 vendors, from $19 to $5,500 (median $399). Most common accelerators: Jetson Orin NX, Jetson Xavier NX, Jetson Orin Nano, Jetson Nano, Jetson AGX Orin.

What to compare

VRAM is the wrong measure here — these devices are rated in TOPS and judged on performance per watt. What decides a deployment is usually thermal envelope, whether it runs fanless, and whether the vendor ships a toolchain for the models you already have.

Who buys these

Robotics, industrial vision and anyone deploying inference somewhere with a power budget and no rack.