Buying guide · Advanced · 3 min read
Edge AI Boards
Jetson, Pi AI Kit, K210 and Coral
By Geekobot Learning Team · Updated 24 Oct 2026
On this page
Edge AI boards run machine learning models on the device instead of in the cloud, for object detection, face recognition, voice commands and anomaly detection. NVIDIA Jetson boards run full GPU frameworks and multiple camera streams. A Raspberry Pi with an AI kit runs fast vision models cheaply. K210 and ESP32-S3 boards run small TinyML models on very little power. Choose by the models you need and the software support, not just TOPS.
Key specs at a glance
- Jetson Orin Nano
- GPU-based, CUDA and TensorRT, Linux, about 7 to 25 W class
- Raspberry Pi AI Kit / HAT+
- Hailo NPU on a Pi 5, fast vision with supported models
- Coral USB / M.2
- Edge TPU for quantised TensorFlow Lite models
- K210 boards
- Small RISC-V chip with KPU, low-power vision
- ESP32-S3
- Vector instructions for TinyML: wake words, gestures
- Software
- Model format, toolchain and examples matter most
Which board for which job
| Board | Best for | Power | Notes |
|---|---|---|---|
| Jetson Orin Nano | Several cameras, robotics, ROS, custom models | High | Most flexible, highest cost |
| Raspberry Pi 5 + AI accelerator | Object detection, people counting | Medium | Large community, supported model zoo |
| Coral Edge TPU | Adding AI to an existing Linux PC or Pi | Low to medium | Needs quantised TFLite models |
| K210 / MaixCAM class | Low-cost camera vision | Low | Smaller models, own toolchain |
| ESP32-S3 | Wake words, gestures, sensor anomalies | Very low | TinyML, small models only |
- Jetson Orin Nano
- Best for:
- Several cameras, robotics, ROS, custom models
- Power:
- High
- Notes:
- Most flexible, highest cost
- Raspberry Pi 5 + AI accelerator
- Best for:
- Object detection, people counting
- Power:
- Medium
- Notes:
- Large community, supported model zoo
- Coral Edge TPU
- Best for:
- Adding AI to an existing Linux PC or Pi
- Power:
- Low to medium
- Notes:
- Needs quantised TFLite models
- K210 / MaixCAM class
- Best for:
- Low-cost camera vision
- Power:
- Low
- Notes:
- Smaller models, own toolchain
- ESP32-S3
- Best for:
- Wake words, gestures, sensor anomalies
- Power:
- Very low
- Notes:
- TinyML, small models only
What to check before buying
check that the board supports your model type and size after conversion and quantisation. Headline TOPS figures do not tell you whether your model runs well.
CSI camera support and the number of streams.
Jetson and AI HATs need good supplies and cooling; see heatsinks and fans.
active forums, maintained images and example projects save weeks.
on-device AI keeps images local and responds without the internet.
Getting started
Start with a pre-trained model from the vendor's examples (person detection or keyword spotting), get it running, then retrain on your own data. For Raspberry Pi 5 projects, the official AI kit examples and camera stack are the fastest path. For tiny devices, Edge Impulse and TensorFlow Lite Micro run on ESP32 and similar boards. Combine AI output with actuators: count people, sort parts with a servo or trigger an alarm.
Try it: Power supply sizing calculator
Change the values and the diagram and results update as you type.
Common mistakes to avoid
- Choosing a board by TOPS alone.
- A model that cannot be converted to the board's format.
- Weak power supply or no cooling.
- Expecting a microcontroller to run large vision models.
- Skipping vendor examples and starting with custom training.
Frequently asked questions
What is edge AI?
Running AI models on the device itself rather than sending data to a cloud server.
Jetson or Raspberry Pi AI kit?
Jetson for flexible GPU workloads and robotics; a Pi with an AI kit for low-cost vision with supported models.
Can ESP32 run AI?
Small TinyML models for wake words, gestures and sensor data, yes; large vision models, no.
What is a TPU or NPU?
A chip designed to run neural network maths efficiently.
Do I need internet for edge AI?
No. Models run locally once installed.
Parts for this guide
Calculators
Terms in this guide
Projects that use it
Put this into practice: 1 step-by-step build in our project library uses parts from this guide.
Keep learning
Sources
- NVIDIA Jetson Orin Nano developer kit documentation
- Raspberry Pi AI Kit documentation
- Google Coral documentation
- Espressif ESP-DL documentation
Written by the Geekobot team from manufacturer datasheets. Always check the datasheet for the exact part and variant you buy.




