New Hailo-10H edge AI accelerator with on-device generative AI
Tel Aviv-based Hailo has officially launched the Hailo-10H, its latest edge AI processor designed to run generative AI models directly on devices—without relying on cloud infrastructure. The new chip is available for order, bringing native support for large language models (LLMs) and vision-language models (VLMs) to edge applications.
For eeNews Europe readers working in embedded AI, automotive, and industrial design, this development is notable and worth looking into. It combines high-performance inference, low power consumption, and support for cutting-edge AI models, all in a compact, edge-friendly form factor.
Generative AI at the Edge, No Cloud Required
The Hailo-10H builds on the success of the company’s Hailo-8, expanding from vision AI into full generative AI workloads. It enables on-device execution of advanced LLMs and VLMs, with practical applications in automotive cockpits, smart home gateways, telecom infrastructure, and retail systems. With real-time processing and ultra-low latency, the Hailo-10H delivers AI performance without the cost, bandwidth, and privacy concerns of cloud-based inference.
Orr Danon, CEO and Co-Founder of Hailo, stated: “With the Hailo-10H now available for order, we’re taking another major step toward our mission of making AI accessible to all. This is the first discrete AI processor to bring real generative AI performance to the edge, combining high efficiency, cost-effectiveness, and a robust software ecosystem.”
Importantly for European designers, the Hailo-10H supports data privacy regulations by keeping processing on-device. It is also automotive-qualified (AEC-Q100 Grade 2), making it a viable candidate for upcoming vehicle platforms, with production ramping up for 2026 designs.
Performance and Power Efficiency
Targeting applications constrained by power and size, the Hailo-10H draws just 2.5W while enabling inference on 2B-parameter models with less than 1-second first-token latency and throughput above 10 tokens per second. For video, it supports real-time 4K object detection using models like YOLOv11m.
The chip is fully compatible with the company’s established development tools, benefiting from an existing global user base of over 10,000 developers. This compatibility simplifies migration from previous designs and accelerates integration for new projects.
With generative AI moving rapidly from cloud datacenters to embedded systems, the Hailo-10H represents a key milestone. Product developers across multiple verticals can now explore natural language interfaces, multi-modal perception, and privacy-conscious AI features entirely at the edge.