Raspberry Pi gets a $70 Sony AI camera

Running real-time computer vision on a credit card-sized computer usually comes with a catch: either your CPU melts under the load, or you have to buy a bulky, power-hungry external accelerator.

Sony and Raspberry Pi think they have a better answer. The two companies just launched the Raspberry Pi AI Camera, priced at $70 and available immediately.

Instead of offloading visual data to a dedicated GPU or the host processor, this camera does the heavy lifting right on the silicon itself.

Neural processing directly on the sensor

The centerpiece of the new module is Sony's IMX500 intelligent vision sensor. It features a 12.3-megapixel resolution and can shoot high-resolution video at either 10 fps or 40 fps depending on the selected mode.

What makes it unique is edge processing. By handling AI tasks directly on the sensor die, it eliminates the need for expensive external accelerator chips or taxing the host board's resources.

The camera board also integrates an RP2040 microcontroller specifically tasked with managing neural network firmware. Out of the box, it comes pre-loaded with the MobileNetSSD object detection model, so makers can start spotting items without downloading extra weights.

It works across the entire lineup of Raspberry Pi single-board computers, including the flagship Raspberry Pi 5. On the software side, it taps straight into the standard libcamera and Picamera2 frameworks developers already know.

The first fruit of Sony's investment

The launch marks the first tangible consumer hardware result of Sony Semiconductor Solutions taking a minority stake in Raspberry Pi Ltd back in April 2023.

Eita Yanagisawa, general manager of SSS, described the camera as the first fruits of their strategic partnership, aimed at bringing edge AI sensing to the world's biggest developer community.

Raspberry Pi CEO Eben Upton echoed that sentiment, noting that the camera directly integrates Sony's deep imaging expertise into an accessible format for everyday tinkerers.

Why it matters

For builders and hobbyists, $70 for an on-sensor inference system is a steal. Building smart doorbells, tracking drones, or sorting robots previously required awkward USB accelerators or led to thermal throttling. Now, the main board just receives pre-calculated coordinates and labels while its CPU stays completely free.

For the embedded industry, this pushes the needle toward integrated sensing and computing. Instead of streaming raw video over bandwidth-constrained buses, the sensor delivers processed insights, which cuts latency and power draw drastically.

We think pairing Sony's high-end IMX500 with Raspberry Pi's dirt-cheap RP2040 was a brilliant packaging choice. It makes advanced edge AI feel as plug-and-play as an ordinary webcam. That said, it looks like custom models will be the real test. While the out-of-the-box MobileNetSSD works fine, shoehorning newer, heavier vision architectures into the sensor's tight memory limits could prove challenging for casual developers.