Skip to content
All insights

Edge AI: running AI on the device, next to the data

6 October 2026

From cameras and IoT sensors to smartphones, AI is moving off the cloud and onto devices. That means faster responses, less bandwidth and private data that stays local.

Talk of AI usually brings to mind large models running in the cloud. Growing just as quickly is edge AI: running the model on the device itself, whether a camera, an IoT sensor or a smartphone, instead of sending everything to a server to be processed.

Comparison diagram: before, a camera or sensor sends all raw data to a central system around the clock; with edge AI the device analyses data itself and sends only the events that matter. Benefits: faster response, less bandwidth, private data stays local.

Why edge AI is taking off

  • The hardware is ready: many new smartphones and IoT boards include a processor built for AI work (an NPU).
  • Smaller models, more capable: techniques such as quantization (compressing a model) and distillation (teaching a small model from a large one) have made on-device models far more capable.
  • Privacy: images and audio never leave the device; only the result of the analysis is sent.

Benefits you can measure

Faster response

There is no round trip to a server, so decisions that must be immediate, such as an on-site alert, happen in a fraction of a second, and keep working when the connection drops.

Less bandwidth, lower cost

A camera streaming video to the cloud all day uses a huge amount of data. Analyse on the device and send only the events that matter, and the volume falls sharply. That suits sites running on a SIM card or a limited network.

Where it is used

  • Traffic and security cameras: count vehicles and detect incidents, sending only event images and statistics.
  • Factory machines: listen to sound or measure vibration to catch the signs before a breakdown (predictive maintenance).
  • Environmental monitoring stations: screen out faulty sensor readings before sending, so the data reaching the central system is more reliable.
  • Mobile apps: read documents, scan text or translate with no internet connection.

Limits to know about

On-device models are still smaller than cloud models, so very complex tasks still need a server. Managing many devices also matters: updating models remotely and keeping each device secure.

Most good systems are hybrid: devices filter and make urgent decisions, while the cloud gathers data, analyses the big picture and improves the models.

Kathi Studio builds IoT devices, applications and back-end systems. If you want the data from your sites to work harder, we can help design everything from the device to the dashboard.