Ambarella and Ultralytics bring YOLO models to CVflow edge AI chips
Ambarella and Ultralytics have announced a technical collaboration to support deployment of the Ultralytics YOLO model family on Ambarella's CVflow-based edge AI system-on-chips (SoCs). The partnership is intended to let developers and product teams build real-time computer vision applications directly on low-power edge hardware, without relying on cloud processing for inference workloads.
The integration covers the full range of YOLO capabilities, including object detection, instance segmentation, image classification, pose estimation, and oriented bounding box detection. Ambarella says a common CVflow architecture and unified software environment span its entire edge AI SoC portfolio, so developers can carry perception software forward as a product scales in performance or shifts to a lower cost point.
The collaboration
On the Ambarella side, the work centres on the Cooper Developer Platform and the Ambarella Developer Zone, a resource hub launched at CES 2026 that consolidates optimised models, agentic blueprints, developer kits and documentation. The aim is to meet developers in the frameworks and model families they already use, rather than requiring bespoke porting work.
Fermi Wang, president and chief executive of Ambarella, said that Ultralytics YOLO is among the most widely adopted model families in computer vision, and that supporting it on CVflow "broadens the set of teams that can build on Ambarella silicon." Glenn Jocher, founder and chief executive of Ultralytics, framed the deal as accelerating the path from development to deployment: "This collaboration combines Ultralytics' accessible computer vision models with Ambarella's low-power AI processing, helping move vision AI from development to deployment more efficiently."
Neither company disclosed financial terms, a revenue-sharing arrangement, or a specific timeline for when YOLO model support will be fully certified and available through the DevZone.
Market context
The edge AI inference market is intensifying as semiconductor vendors compete to capture workloads that are migrating away from cloud compute, driven by latency requirements, bandwidth costs, and data-sovereignty concerns. Ambarella competes in this space against dedicated edge AI chip makers such as Hailo and Kneron, as well as the edge-facing product lines of larger semiconductor companies including Qualcomm, NXP, and Texas Instruments, all of which have pursued partnerships with popular model providers to reduce developer friction.
YOLO's position as a de facto standard in real-time object detection gives Ultralytics significant leverage in such partnerships. The model family's open-source distribution, with over 330 million package downloads and more than 140,000 GitHub stars according to the company's own figures, means hardware vendors that support it natively gain access to an existing developer community rather than having to build one. For Ambarella, the collaboration complements its stated installed base of more than 50 million AI SoC units across physical security, automotive, drones, and robotics.
Regulatory and standards read-across
Edge AI deployments in automotive and physical security are subject to growing regulatory scrutiny in both the EU and the UK. The EU AI Act classifies certain automated surveillance and vehicle safety systems as high-risk, requiring conformity assessments and data governance documentation before deployment. Developers building on the Ambarella and Ultralytics stack for those verticals will need to account for those requirements at the platform design stage, not as an afterthought. The collaboration's emphasis on efficient on-device processing also aligns with emerging guidance on data minimisation: running inference locally means raw video does not need to leave the device, a meaningful advantage for compliance in environments where personal data is captured.
The next milestones to watch are formal benchmark results comparing YOLO inference performance on CVflow against competing edge AI silicon, and the first named customer deployments across the targeted verticals of smart cameras, robotics, automotive, and industrial systems.