Cheap NPUs are quietly turning SBCs into local-inference boxes
A wave of low-cost accelerator HATs and NPU-equipped boards is pushing real on-device inference down to single-board-computer prices. The interesting part isn't the TOPS number — it's where the work moves.