Press release

DEEPX NPU Turns Raspberry Pi Into a Physical AI Platform That 'Sees and Understands'

AI summary

Raspberry Pi's official news site featured DEEPX NPU technology, which runs everything from image recognition to vision-language AI on about 3 watts.

Key points
  • Raspberry Pi's official news site features DEEPX NPU technology, from image recognition to vision-language AI
  • 25 TOPS of computing performance at about 3 watts supports real-time, always-on AI inference
  • Global open AI ecosystem connects developers' first experiments to industrial products
Sixfab AI HAT+ board for Raspberry Pi 5 with a DEEPX DX-M1M NPU
The Sixfab AI HAT+, an AI accelerator board for Raspberry Pi 5 built on DEEPX's DX-M1M NPU. (Raspberry Pi)

SEOUL, South Korea, Oct. 15, 2026 — Neural processing unit (NPU) technology from DEEPX, a developer of ultra-low-power AI semiconductors, has been featured on the official news site of Raspberry Pi, the global small-computer platform.

Raspberry Pi highlighted that DEEPX NPUs can handle everything from image recognition to small vision-language AI models on about 3 watts, and pointed to potential uses across physical AI fields such as robots, smart cameras and industrial equipment. The feature also fits DEEPX's strategy of building on the global developer ecosystem to bring more AI technology into industrial products.

Raspberry Pi recently published a technical article on DEEPX and Sixfab, an official design partner. The article covers how to run AI directly on the device, and what it can be used for, with the Sixfab AI HAT+, an AI accelerator board for Raspberry Pi 5 built on DEEPX's DX-M1M NPU.

DX-M1M delivers up to 25 TOPS of computing performance while drawing about 3 watts in typical continuous operation. That figure covers the NPU alone, excluding the Raspberry Pi board and peripherals. Low power draw and heat reduce the cooling burden on small devices and make it easier to keep AI features running continuously.

The range of applications extends beyond detecting objects to understanding and explaining situations. Convolutional neural networks (CNNs) can detect objects and anomalies in video, small vision-language models (VLMs) can interpret the context of a scene, and small language models (SLMs) can understand commands or sum up information in short sentences.

For example, a camera over a workbench could detect a 3D printer error, analyze the scene and explain the problem to the user. The same approach can extend to quality monitoring on production lines, environmental awareness for robots and situational analysis in smart spaces. Running AI on-site also reduces dependence on external data transfers and network connections, helping with real-time response and privacy.

To help developers build these applications, DEEPX provides model conversion and optimization tools, runtime software, video processing tools and pre-optimized AI models. The Sixfab AI HAT+, an edge AI expansion board and the ALPON X5 AI industrial AI gateway share the same DEEPX runtime and model format. This software continuity means models validated on a development board can also be used in industrial products.

This support is part of DEEPX's open ecosystem strategy, which connects developers with familiar hardware, AI models and software tools. DEEPX is the only Korean NPU company to have built partnerships across global hardware and AI software ecosystems including Raspberry Pi, Baidu's PaddlePaddle and Ultralytics. Through these partnerships, DEEPX is lowering the barriers for developers to adopt AI chips and building an environment that carries them from early technology development through prototyping to mass production.

DEEPX sees particular value in the fact that developers around the world can experiment with AI on Raspberry Pi and grow those experiments into real products. Because the technology developers first pick up and learn can lead to future industrial product development, the company expects that strengthening its position in the global developer ecosystem will play an important role in building a long-term customer base.

In June, DEEPX and Sixfab launched the Sixfab AI HAT+, an AI accelerator board for Raspberry Pi 5 built on DX-M1M. The 25 TOPS version is on sale now, and a 13 TOPS version is scheduled for release by the end of the year.

"For physical AI to spread, developers around the world need to be able to easily build the AI features they need and turn them into real products. We will support the whole development environment, from chips and software to mass production, so that small experiments that start on Raspberry Pi can grow into products sold in markets around the world." — A DEEPX official

  • #deepx
  • #raspberrypi
  • #sixfab
  • #npu
  • #physicalai
  • #edgeai
  • #dxm1m