Specifications
Kit Basis
Hailo-10H M.2 Key-M AI acceleration module in a starter-kit bundle
AI Processor
Hailo-10H second-generation neural core architecture
AI Performance
40 TOPS at INT4; 20 TOPS at INT8
Generative AI
Runs LLMs, VLMs, Whisper speech models and diffusion-style workloads on-device
Vision Models
More than 150 supported vision networks from the Hailo Model Zoo
On-Module Memory
4 GB or 8 GB LPDDR4 / LPDDR4X, 4266 MT/s
Memory Architecture
Direct DDR interface, allowing scaling to large-model inference
Host Interface
PCIe Gen 3.0 x4 lanes
Form Factor
M.2 Key M — 2242 and 2280 module lengths, 22 x 80 mm
Typical Power
2.5 W typical consumption
Host Architectures
x86 and Arm hosts
Operating Systems
Linux, Windows and Android
AI Frameworks
TensorFlow, TensorFlow Lite, Keras, PyTorch and ONNX
Industrial Grade Option
Module offered in industrial grade spanning -40 C to 85 C
Thermal / Environment
Starter kit module rated 0 C to +40 C operating; industrial variant to 85 C
Humidity
20% to 85% RH operating, non-condensing
Physical
22 mm x 80 mm, approximately 6 g
Regulatory
RoHS, China RoHS, EU REACH, J-MOSS, WEEE; FCC, CE, RCM, BSMI, VCCI, UKCA
Kit Contents
Hailo-10H M.2 module plus quick start guide
Distribution
Sold through Hailo global distributors; volume quotation via QS Compute
Overview
The Hailo-10H M.2 AI Acceleration Starter Kit is the fastest path to evaluating Hailo's second-generation edge AI silicon. The kit bundles the Hailo-10H M.2 Key-M module with a quick start guide, so it drops into any x86 or Arm host with a free PCIe Gen 3.0 x4 M.2 Key-M socket and runs immediately under Linux, Windows or Android.
The Hailo-10H is the industry's first edge AI accelerator explicitly aimed at generative AI at the endpoint. It delivers 40 TOPS at INT4 (20 TOPS at INT8) while consuming about 2.5 W typical, and unlike the earlier Hailo-8 it carries a direct DDR interface with 4 GB or 8 GB of LPDDR4/4X on module. That combination is what lets it run LLMs, VLMs and Whisper speech models locally rather than only classic vision networks.
Practically, that means a starter kit can be dropped into an existing industrial PC, edge box or AI PC to add local generative inference without redesigning the enclosure — the module is specified at 22 x 80 mm and 2.5 W typical, so it can be treated as a peripheral rather than as a new thermal problem. Vision coverage is not sacrificed either: more than 150 models from the Hailo Model Zoo run on the same silicon.
For series production the Hailo-10H is available in an industrial-grade variant covering -40 C to +85 C. QS Compute supplies starter kits for evaluation and quotes volume pricing on the module for production builds.
Key Benefits
Generative AI without a GPU: 40 TOPS INT4 and a direct DDR interface run LLMs, VLMs and Whisper locally. 2.5 W typical: adds edge AI to a sealed or fanless box without reworking thermal design. Drop-in M.2 Key-M: fits existing x86 and Arm hosts with a PCIe Gen 3.0 x4 socket. Vision and GenAI on one part: over 150 Model Zoo vision networks alongside large-model inference.
Applications
On-device LLM and VLM assistants in industrial HMIs, private speech recognition and Whisper transcription at the edge, intelligent camera and video analytics nodes, retail and access-control AI appliances, medical and lab instruments requiring local inference, robotics perception modules, and AI PC or workstation add-in acceleration.
Request a Quote — HAILO-10H M.2 AI ACCELERATION STARTER KIT — 40 TOPS GENERATIVE AI AT THE EDGE
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →