License YOLO
Technology partner

Pioneering innovation in edge AI semiconductors with DEEPX

Ultralytics partners with DEEPX to bring fast, efficient YOLO inference to NPU-powered edge AI silicon.

Export Ultralytics YOLO to DEEPX NPUs with one command. Achieve GPU-level vision AI accuracy at a fraction of the power for robotics and industrial automation.

About DEEPX

DEEPX is an AI semiconductor innovator delivering the world’s most efficient edge AI hardware—purpose-built to run models like Ultralytics YOLO at unmatched frames-per-second per watt (FPS/W). By redefining performance standards around real-world efficiency rather than theoretical metrics, DEEPX enables developers to achieve GPU-class accuracy with only a fraction of the power budget. DEEPX’s ultra-low-power, high-performance chips are already being validated in mass-production systems across the U.S., Korea, Taiwan, and China, powering robotics, factory automation, drones, and industrial inspection solutions. By combining YOLO-optimized efficiency with real-time adaptive intelligence, DEEPX is enabling the next era of autonomous robots, smart factories, and edge AI systems that are both powerful and sustainable.

Why choose DEEPX for Ultralytics YOLO?

Optimized for Ultralytics YOLO

Full Ultralytics YOLO support out of the box. Deploy with a single format=deepx command.

Edge-native performance

Purpose-built NPU architecture for real-world physical AI workloads.

Real-time inference

On-device inference across all seven YOLO tasks, with real-world speed and reliability.

Lower cost of ownership

A fraction of the power cost of GPUs, with none of the performance trade-offs.

Easy integration

Production-ready pipeline with zero-friction model quantization and validation.

Future-proof

Dedicated CI/CD pipeline ensures compatibility with every new Ultralytics release.

The complete solution

Build and refine your custom YOLO models on Ultralytics Platform. From Ultralytics YOLOv8 to YOLO26, access state-of-the-art architectures, curated datasets, and powerful optimization tools, all designed to get you to production faster.
Convert and quantize your trained YOLO model for DEEPX NPU deployment in a single command. The export process leverages DEEPX's DX-COM compiler to compile your model into an optimized .dxnn package for high-performance NPU inference.
Deploy your compiled model on DEEPX NPUs for real-time, on-device inference, with low latency, minimal power draw, and the reliability needed for production edge AI workloads. Powered by the DX-RT runtime, ensuring seamless integration across x86 and ARM-based edge systems.
Train with Ultralytics YOLO

Frequently asked questions

  • We work with established hardware, cloud and robotics companies, enterprise software platforms with an active user base, and system integrators deploying vision AI at production scale.

  • Early-stage startups can join our Startup program, students and community builders our Ambassador program, and anyone can license YOLO to build with it today.

  • Strategic partners work with us on joint engineering and go-to-market. Technology partners have a supported YOLO integration documented in our docs.

  • YOLO exports to formats such as ONNX, TensorRT, OpenVINO, CoreML and LiteRT, so it runs on CPUs, GPUs, NPUs and edge devices. See all integrations in our docs.

  • Ultralytics YOLO is available under AGPL-3.0, which requires you to open-source your project, or under an Enterprise license for private and commercial use.