Dev Tools · 1h ago
AI model runs on $10 microcontroller using Google's per-layer embeddings
A developer ran a 28.9-million-parameter language model on an ESP32-S3 microcontroller costing $10. The technique uses Google's per-layer embeddings to store a lookup table on 16MB of flash memory. The model is slow and limited but shows AI inference can run on ultra-low-cost hardware.
Meridian48 take
This is a clever hack that pushes the boundaries of edge AI, but the practical utility of a tiny, slow model on a microcontroller remains niche.
Read the full reporting
AI developer runs 28.9-million-parameter model on $10 ESP32-S3 microcontroller — uses Google's Per-Layer Embeddings technique, stores table on 16MB Flash memory →
Tom's Hardware
edge-aimicrocontroller